Archive for: September, 2007
September 28th, 2007
More facts and less hysteria on Vista, please!
In the latest round of hysteria to be written about Windows Vista, Don Reisinger regurgitates the usual hysteria about Windows Vista mixed in with a pinch of facts here and there. Don spouts off the usual nonsense about sales, UAC, and even DRM. Despite the fact that bashing Vista is quite the popular sport these days, I’m going to see if I can set him straight with an honest and factual assessment of Windows Vista.
Are Vista sales really poor?
Everyone knows that Windows Vista retail box sales are poor, but does that matter when Microsoft relies overwhelmingly on sales to OEM PC makers? If you focus only on the retail box sales, you’re missing the real picture because Vista has sold more than 60 million licenses and ~78% of those sales are Vista Premium edition. Don complains about Windows Vista Ultimate edition and I actually agree with him that it’s overpriced and under delivers but Microsoft doesn’t need to “save itself” if Vista Ultimate fails, more like an “oh well”.
Does it matter if a few people revert to XP?
Even if a whopping 20% of computer buyers downgrade and revert to Windows XP for whatever reason, that still leaves 80% who stay with Windows Vista. That means hardware makers and ISVs (Independent Software Vendors) have to deal with Windows Vista now or later whether they like it or not if they want to stay in business. The fact that 60 million copies were sold in the first 6 months since launch pretty much confirms Vista will become the dominant OS by default.
How about Vista drivers?
There are no questions about it, a fair number of Vista drivers during the first 2 months stunk badly. Vista implements a brand new driver model which offers a little more separation between the driver and the kernel so that a bad driver is less likely to crash the entire system. The price for this is that there is a brand new learning curve and it took a few months for the hardware companies to get it right. For the most part, everything is working well but there are still some older devices that don’t have drivers and will never get drivers for Windows Vista and much of that is because the hardware vendors want you to buy new hardware.
Is it fair to expect a hardware company to write drivers for a 3+ year old device? Probably not. Is it fair to expect users to buy new hardware because they can’t get Vista drivers for a one year old device? Definitely no! Reputable hardware companies that want to keep customer loyalty will go back as far as they can to create drivers for older hardware. Was it Microsoft’s fault that the drivers didn’t work well at first? Technically no but that isn’t going to matter to consumers and they’ll take their anger out on Microsoft and the hardware maker. Fortunately, the driver situation has stabilized but it’s always good to check for drivers before you upgrade a computer to Vista and before you buy a piece of hardware.
How about application compatibility in Windows Vista?
There’s no question about it, applications will break in Vista and it’s probably the #1 reason some people are reverting back to Windows XP. This is primarily due to the fact that many applications never followed Windows development guidelines set since 2000. One of the worst offenders is Intuit which refused to properly write QuickBooks right up to the 2006 version. Intuit never followed Windows development guidelines that have existed since Windows 2000 and XP for Windows logo certification and they - like many other software makers - used Vista as an opportunity to sell a new version of QuickBooks 2007. If you bought QuickBooks 2006 or earlier, you were out of luck and it wouldn’t run on your new computer and you had to buy QuickBooks 2007.
Microsoft asked developers for 7 years to clean up their act but drew the line in the sand with Windows Vista which comes with UAC (User Access Control) on by default. That finally forced vendors like Intuit to properly code their application and not violate security best practices. If UAC does nothing else and even if people turn it off, it has had the desired effect of cleaning up the Windows development community.
Vista and Internet Explorer 7 also breaks a lot of applications in the name of drawing a line in the sand for security. Microsoft will get criticized for not getting rid of things like Active X but they’ll also get criticized for breaking dangerous coding techniques and the vast majority of Active X controls have been disabled in Windows Vista by default. There are still plenty of web applications that don’t work inside Windows Vista and Internet Explorer 7 and vendors like Kodak will try to drag their feet but they will have to deal with it sooner or later unless they want to alienate the 60 million (since summer) and growing Vista user base. My colleague David Berlind questions why Microsoft needs to break so many legacy applications and the answer is security. It’s a known fact that until something is hard broken, no one will change anything. Is this going to be painful? Certainly. But it has to be done if we want a more secure computing environment.
September 27th, 2007
Sun's new servers get intimate with Intel and Microsoft
John Fowler (Executive VP Sun Microsystems) announced a pair of new Servers yesterday at Sun’s head quarters in Menlo Park California. Fowler showed off one of the servers last week at Intel IDF in San Francisco an posed with Pat Gelsinger (standing on the left in picture to the left) where he announced products based on Intel processors running Microsoft Windows on top of a Solaris Hypervisor platform. The same X4450 2U quad-processor 16-core (two rack units at 3.5″ tall and 19″ wide) server was also shown at the Tigerton launch three weeks ago and I wrote extensively about Sun packing a big bang in a 2U server. Sun has now packed a dual-processor server in to the X4150 1U chassis with a whopping 8 SAS 2.5″ hard drives (pictured below).
Having this many number of drives in addition to 16 FBDIMMs in a 1U server is definitely unusual and impressive. The hard drives brackets had to be redesigned to push the drives back to allow plenty of air-flow to cool the two quad-core Intel 5300 series processors and 16 Fully Buffered DDR2-667 memory DIMMs. One of the three PCI-Express slots is loaded with a battery backed up RAID controller that allows worry-free write-back-cache operation to speed up write operations when using striping technology. Sun made it clear that despite the fact that Solaris uses ZFS which includes RAID-like redundancy features implemented in software, Sun was going to support the Windows, Linux, and Solaris world with hardware based RAID-0, RAID-1, RAID-5, RAID-6, RAID 1+0, and RAID 6+0.
One of the things a reporter asked at the launch event was why Sun wasn’t supporting “embedded virtualization” like other server vendors. I had to contain myself since this is another one of those marketing slogans that make me roll my eyes. As I pointed out when I covered the 2U X4450, embedded virtualization is nothing more than a $20 USB stick shoved in to a port on the mother board booting the Hypervisor software and it’s no different that a $300 PC booting off the USB port for the last couple of years. The 2U X4450 and the 1U X4150 both have internal USB slots you can insert a USB stick in to for booting anything under the sun including Hypervisors from any of the virtualization players. John Fowler had to thread the needle to answer politely but I’d suggest to him that he should explain that Sun does have the capability like everyone else in the future.
Another question asked from the audience is the appropriateness of these servers for HPC (High Performance Computing). This put Fowler in a tough position sitting next to Intel’s Pat Gelsinger because this particular server with the Intel 5000 series “Bensley” chipset wasn’t the best HPC platform in the world. Intel dominates on the general purpose server market benchmarks like SPECint_rate2006, SAP, SPECjbb2005, SPECweb2005, and TPC-C but not on high memory bandwidth HPC applications. Regardless of the fact that AMD might be exaggerating their lead by advertising scores from a non-existent model 2360SE processor not due until end of year and advertising an unlabeled processor in full page ads on the Economist, the current leader in the dual-processor HPC market is based on AMD’s “Barcelona” Opteron 2350 which has the highest published score on SPECfp_rate2006 and a lower power consumption level. Fowler of course wasn’t going praise AMD in an event with Intel so these servers were presented as good solutions for the HPC market. Sun does offer a 2U AMD 16-core equivalent to the X4450 (which I covered here) though it’s not clear if Sun has a 1U dual-processor 8-core equivalent to the X4150 server.
I followed on with a question if Sun was going to follow on with a “Stoakley” platform (faster Inter chipset with Fully Buffered DDR2-800 FSB1600 support) with Harpertown (45nm Penryn server variant CPU) version of this server since the new platform and CPUs were due from Intel this November 12th. This question is very relevant to the previous question about HPC performance since Intel showed off SPECfp_rate2006 performance estimates at last week’s IDF event that were even higher (chart below) than AMD’s 2360SE product due at the end of the year.
Gelsinger jokingly told Fowler that he wanted to know when Stoakley was going to be put in to Sun’s servers too but Fowler declined to discuss future product offerings but did say that the server platform was very flexible and that they wouldn’t design it without future processors in mind. I spoke to Sun’s engineer after the event and he explained that the power regulator was detached and modular and they could easily replace the motherboard and CPU with a Stoakley platform.
September 27th, 2007
Nike fixes e-commerce site by implementing HTTPS
After weeks of emails bouncing back and forth between me and Nike representatives about the lack of HTTPS SSL security on Nike.com, I finally got an email yesterday that Nike has fixed the problem. I was first made aware of the issue by fellow blogger David Berlind’s post on Nike’s e-commerce failing to implement HTTPS which makes it impossible for shoppers to know if they’re looking at the real Nike.com or if they’re feeding their credit card information to a criminal posing as Nike.com.
There was a similar case with a large number of American banks that did the same thing and failed to implement HTTPS for their online banking site and most of the banks silently fixed the issue after a few months without informing anyone that anything was wrong in the first place. I wasn’t sure if I was getting through to Nike and I was just about to write a blog exposing the situation when I got the email explaining that Nike fixed the problem after some internal meetings and tests to verify nothing would break with their flash-based e-commerce site. Now you go to Nike.com and pick out what you want and right and hit the check out button, you’ll be redirected to an HTTPS site before you are asked to enter your credit card information.
There was some initial concern that this couldn’t be done while maintaining the session so I was going to suggest encrypting the entire shopping session since encryption is basically free (from a hardware standpoint) these days but that wasn’t necessary. I am happy with Nike’s overall responsiveness to my complaints and I hope every online site learns from this incident and not make the same mistake again.
September 26th, 2007
Europe's new 'monopoly' tariff on Microsoft bypasses WTO
The European Commission has just levied a new $689,900,000 “fine” (read: tariff) on American software company Microsoft under the pretense of anti-trust which conveniently bypasses WTO agreements. The Brussels based think tank Globalization Institute has published a paper (PDF) where it recommends a ban on OS (Operating System) bundling for all PCs sold in Europe. At the end of the paper it writes:
Policy recommendation
This paper’s recommendation is that the European Commission should require all desktop and laptop computers sold within the EU to be sold without operating systems.
Scott M. Fulton, III wrote an excellent news piece here where he covers the key issues and points of views and our own bloggers John Carroll and David Berlind weighed in on the issue. Alex Williams of the Adam Smith Institute says “This neo-protectionist economic agenda is forming a policy cloak for the anti-Americanism of many European Commissioners, and it is European citizens who stand to suffer from it.” and I agree with him.
The Globalization Institute says their recommendation will produce more “choice” but I can’t possibly see how this would produce more choice when 90% of the population wants an Operating System (not necessarily Windows) bundled with their computer and they have no desire to install their own OS or pay someone to do it. I can even agree on a matter of principle that computer makers should be forced to sell no-OS computers as an easy option for consumers or businesses though the savings won’t be as big as some people think since hardware makers don’t pay full OEM (Original Equipment Manufacturer) pricing or anything close to it. But to tell European Consumers and they can’t buy a computer with a heavily discounted pre-configured Operating Systems is asinine in my opinion and it is the absolute antithesis of choice.
When a PC maker bundles Windows Vista Home and sells the entire computer to the consumer at $350, does anyone think they pay $100 single-quantity OEM cost let alone the $199 retail price? There’s absolutely no way and I would venture to guess that the true cost of Windows Vista Home is in the vicinity of $60 because Microsoft sells at a significant quantity discount. The computer comes with the OS and hardware qualified drivers integrated in to the system and everything works out of the box which is what 90% of the population wants. Dell (and other PC makers) have started offering users the option of getting Linux bundled with the PCs because of MARKET demand but now some Bureaucrats in Brussels wants to tell Dell and others that this is now going to be against the law?
The last time the EC (European Commission) in their infinite wisdom decided to ban the bundling of software forced Microsoft to ship a version of Windows without Windows Media Player installed. To the EC’s consternation, no one bought that crippled version of Windows and they kept buying Windows. Now some of these same people want to consider crippling PC companies and force them to sell worthless hunks of metal to people with no operating system installed and people will have to figure out how to install their own OS and device drivers or pay someone else to install it for them not to mention the additional cost of buying single-unit OEM OS.
The European Commission is frustrated that despite all their meddling these last few years, Microsoft has doubled their market share in the “Workgroup Server” market from 40% to 80% thought this is another one of those arbitrary definitions like the Apple iTunes monopoly definition. When you factor in all those unregistered or roll-your-own copies of Linux running in the market place, you can hardly declare Microsoft a monopoly in the server space. Within that narrowly defined market segment, perhaps the EC should consider the fact that people prefer paying $600 perpetual licenses for Windows Server plus a very occasional $250/incident support fee (typically 4 times a year for all Microsoft issues for my old company) over a $1300/year/server support contract for Red Hat Enterprise Linux. Those tens of thousands of copies of Linux being run by Google and millions more by other companies don’t count in the eyes of the EC when they need to define Microsoft a monopoly.
The EU Competition Competitor Neelie Kroes said that the EU now expects a “significant drop” in Microsoft’s overwhelming market share. In fact Kroes even hinted that perhaps somewhere around 50% but not exactly is the correct market share. Kroes’ spokesman Jonathan Todd clarified that:
“Once illegal abuse has been removed and competitors are free to compete on the merits, the logical consequence of that would be to expect Microsoft’s market share to fall,”
So I can translate this (via the contrapositive rule of logic) that if the market share doesn’t fall, then that “logically” must means that free competition doesn’t yet exist and illegal abuse must still be rampant. That leaves absolutely no other possible explanation for Microsoft’s dominant market share so what’s next if crippling PC makers doesn’t work? Will the EC then order ISVs (Independent Software Vendors) to port all of their applications to Linux with equivalent performance, functionality, stability, and validation if they wish to continue doing business in Europe? Where does the madness end?
I have no doubt some people are jubilant about the fact that someone is sticking it to Microsoft, but do they honestly believe that an EC that tastes the fruit of their fines (tariffs) will stop with just Microsoft? They’ve already declared Apple’s iTunes a monopoly so what is to keep them from imposing a new WTO-bypassing tariff on Apple? What happens when the EC declares Cisco a monopoly in routers because their market share is too big and not because they’ve actually broken any anti-trust laws? Will the EC come up with all sorts of creative remedies to force Cisco to drop their market share to ~50%? What happens when the EC declares Oracle a monopoly in their respective market? Should Intel’s market share be knocked down to ~50% too?
The American people and their politicians need to wake up to the fact that the EC is imposing tariffs under the guise of anti-trust merely on the basis of market share. Europeans need to realize that their politicians are doing no favors for them with these draconian rules and that they will end up paying higher prices and greater hassles. Trade is a two-way street and there will have to be repercussions and the side that has the trade surplus bleeds the most.
Loading ...September 25th, 2007
Why OLPC mesh wireless networking won't work
One of the touted features of the $200 OLPC laptop is the peer-to-peer mesh topology networking feature that can theoretically bring an Internet infrastructure where there is no network infrastructure. The problem is that peer-to-peer wireless LAN mesh topology sounds better than it actually works and there’s a good reason it isn’t used commercially.
[UPDATE 9/27/2007 - I should clarify that OLPC mesh technology applies to the XO laptop shown on the left or to the Intel Classmate [current version of Classmate doesn't support mesh]. Intel is also on the board of OLPC so it’s not OLPC versus Intel. Intel is also providing some help on technology based on the centralized Access Point and Bridge model. OLPCs can also work with centralized wireless LAN infrastructures and that is the point of this blog; that the two technologies work best together and that they’re not mutually exclusive. A $60 Linksys router running modified Linux and a $20 antenna can provide fast and reliable infrastructure for the entire school.]
The word “mesh” is traditionally highly regarded in the networking world because every IT student is taught in Computer Networking 101 that “mesh topology” is the most advanced form of networking. Mesh topology traditionally conjures up the image of multiple redundant links with high-performance distributed loads but that only applies to the wired networking world when multiple physical links are used to build the network. High-performance and load-distribution does not apply to wireless mesh topology especially when we’re talking about typical implementations that use a single radio and a single radio frequency. In fact, every wireless relay adds another hop and the relay action doubles the radio contention because the same data has to be retransmitted on the same radio frequency.
Even if we ignore the delay and contention problems of mesh topology wireless LANs, there’s an even more fundamental problem facing the peer-to-peer mesh technology being implemented in projects like the OLPC. The radios and antennas are so small that it would take hundreds of OLPC devices with perfect spacing to replace a single high-powered Access Point with high-gain antennas. Consider the illustration below where I compare OLPC laptops that are capable of transmitting up to 50 meters with their small 30mW radios and small antennas versus a centralized AP that’s capable of 400 meters range.
Mesh versus Access Point topology:
Note that I’m being very conservative with the 400 meter range with a 300mW Access Point because those things can easily go twice as far. But even with a mere 8:1 advantage in range, it would take more than a hundred OLPC laptops to cover the same area. If we’re talking about a more realistic 16:1 advantage in range, then it would take more than 400 OLPC laptops to cover the same area and they would all have to be spaced out perfectly. We also have the possibility of using 500mW radios and 16 dBi antennas for even longer range in rural areas. When we consider the fact that a single failure in one of the mesh nodes due to battery drainage, moving out of range, software hang will cause the entire mesh scheme to break, there simply is no way to get around the centralized architecture.
Last week at Intel’s IDF convention in San Francisco, Intel’s “World Ahead Program” was showing off some cheap commodity technology and blueprints that would empower schools with wireless networking and Internet access. These blueprints and part lists allow the schools to build their own wireless infrastructure with cheap off-the-shelf components. The all-in-one Wireless Access Point and Wireless Bridge box (dual radio) allows remote locations that lack wired Internet uplinks to bridge wirelessly to the central uplink. I came up a slightly modified version shown in the illustration below to show the flexibility of this architecture.
AP and bridged extension wireless LAN (full size):
With a few of these “towers” with sufficient transmit power and high-gain omni-directional antennas for client access and directional antennas for the backhaul; we can reliably cover a very large campus.
September 24th, 2007
Finally a reasonably cheap 80 Plus power supply
One of the key criteria to getting a new EPA Energy Star rating on a new computer is having an “80 Plus” power supply. A computer power supply that is more than 80% efficient at 20%, 50% and 100% load can earn an 80 Plus logo. Finding an 80 Plus power supply for a do-it-yourself PC at a reasonable price has been challenging and too many 80 Plus power supplies are oversized and overpriced.
A large price tag prices out a large portion of the market and oversized power supplies don’t actually deliver the kinds of efficiencies their marketing implies. A 500 watt power supply that is rated 80 Plus will perform at well below 80% efficiency when paired with a modern PC that typically uses 50 to 100 watts of power because that’s less than 20% loading. The solution is to find a small and inexpensive power supply in the 200 watt range and Sparkle Power Inc. may have produced the right solution with their SPI220LE power supply.
The SPI220LE is small (fits in 1U chassis and some MicroATX or FlexATX chassis), reasonably priced at $55 MSRP (available online), and it’s completely silent since the fan doesn’t even move until you get to 50% loading which is almost never. In my quest to build a 50 watt PC, I hit a point of diminishing returns because my SeaSonic S12 330 watt PSU couldn’t operate that efficiently at 15% loading. I could not get the computer to use less than 50 watts no matter how hard I tried and 51 watts was as low as I got.
With the SPI220LE, power consumption on a G33 motherboard and Intel E6750 dropped down to 47 watts when it was 52 watts [UPDATE 1:00PM - using an already efficient Seasonic S12 330W power supply. The 330W PSU while efficient at 20% to 100% loses efficiently at loads of 15%]. A D201GLY dropped down to 31 watts in idle power consumption with the SPI220LE while the same computer measured 46 watts when I used a cheap PSU that came with a $16 MicroATX chassis which meant it was consuming an extra 48% in power. That’s a huge difference in efficiency level and it’s no wonder the EPA is prioritizing on power supply efficiency when it comes to computers. Now I just wish more vendors carried the SPI220LE power supply so that I wouldn’t have to pay full retail price.
September 21st, 2007
Mini PC cottage industry, is dead at 6
The Mini PC cottage industry began life in 2001 with the birth of Mini-ITX lived a modest life in certain niche markets, died this year by commoditization. It was 6 years old.
*
Because of the boutique status of Mini PCs, they commanded hefty premiums over their larger and more capable cousins and sold for twice the money while having only 1/4 the capabilities. Prices ranged from $500 to $1000 for a silent low-power Mini PC. But tragedy struck with commoditization this year when Intel launched an all-out price war on Mini-ITX motherboard/CPU market with a new product line geared after the “next billion PC” emerging markets.
The Intel D201GLY is a PC motherboard that’s referred to as a MicroATX when it fact it’s the same size as Via’s popular Mini-ITX motherboard. The product comes with a “Celeron” Model 215 1.33 GHz CPU with 27 watt TDP (Thermal Design Power) based on a single core version of the “Yonah” processor which is slightly better than a Pentium 4 2 GHz processor. The processors code-named Yonah because the original Core Duo and Core Solo processors used in laptops and even some servers 10 months before Intel launched its new Core 2 Duo line of processors.
I stopped by Fry’s last night and picked up one of these boxes for $69.99 (same price at here) and it was sitting next to a Mini-ITX Motherboard and Via CPU box with slightly better capabilities because of an integrated Firewire port. One major difference was the fact that the Via product had an extra 1 in front of the price tag selling at $169.99 which was what it usually went for in the glory days of the Mini PC cottage industry.
At IDF this week, Intel showed a newer model called the D201GLY2 with an extra 2 SATA ports in addition to the existing PATA IDE port and it comes with a new Intel “Celeron” Model 220 1.2 GHz processor that can be cooled completely passively. That would seem to indicate that the new processor is probably in the 10W to 15W TDP range if it can be cooled without an active fan on top of the CPU. Pricing is expected to be similar to the slightly older D201GLY and should be available soon.
The heat on Via is on and it will undoubtedly force them to drop their prices once these Intel combos costing 2.4 times less begins to really sell in bulk. Farewell to the days of premium prices for Mini PCs and I can’t really say I’ll miss them.
* Note that the Mini PC picture shown is actually one of the cheap commodity fanless PCs from eWay from Justin James’ review and tutorial. It’s a photo that CNET Networks owns the rights to so it was used as a model for the Mini PC.
September 20th, 2007
Leaked - Skulltrail 8-core Penryn benchmarks at IDF
All sorts of things have been mysteriously falling in to my lap lately and IDF has been no different. Someone who does not work for ZDNet or CNET Networks handed me a sheet of paper with some very interesting benchmarks for the newest “Skulltrail” enthusiast game/workstation. Since I signed no NDAs and never agreed to one verbally, I’m free to disclose the leaked performance numbers.
The Skulltrail is a computer based on a dual-socket motherboard that’s probably a derivative of “Stoakley” which supports FSB1600 with Fully Buffered DDR2-800 DIMMS and two quad-core Penryn 45nm CPUs. The system shown below is based on two “pre-production” 3.4 GHz quad-core Penryn processors with two 2 GB DDR2-800 CL5 FBDIMMs. It also has two NVIDIA 8800GTX graphics cards in SLI configuration but the benchmarks in question weren’t graphics intensive so they merely showcase the capability of the CPU.
I took a high-resolution photo of the benchmark results below
[Update 1:30PM - converted to text below]
| Test system Configuration | System A | System B |
| PROCESSOR | 2x Pre-production 45nm Hi-k Intel quad core processors (12MB cache, 3.40Ghz, 1600MHz FSB) | Pro-production 45nmmi High-K quad core processor (12MB cache, 3.00GHz. 1333Mhz FSB) |
| MOTHERBOARD | Skulltrail dual socket motherboard | Pre-production Intel X38 - motherboard |
| BIOS | Pre-Production Engineering Bios | Pre-Production Engineering Bios |
| GRAPHICS CARD | 2 x NVIDIA GeForce 8800GTx | 1 x NVIDIA GeForce 8800GTX |
| GRAPHICS DRIVER | NVIDIA pre-production driver | NVIDIA ForceWare Release 162.22 |
| MEMORY | 2×2GB Pre-production Micron DRD2-800 CL5 FBDIMMs | 2×1GB Corsair CM3X 1024-1333C9DHX |
| HARD DRIVE | Seagate Barracuda 320GB 7200.10 7200RPM | Seagate Barracuda 320GB 7200.10 7200RPM |
| OPERATING SYSTEM | Vista x86 32-bit Ultimate | Vista x86 32-bit Ultimate |
| Benchmark or software application | ||
| 3DMark*06 V1.1.0 Pro - CPU (score) | 6359 | 4569 |
| 3DMark*06 V1.1.0 Pro - Overall (score) | 17006 | 11899 |
| Cinebench* R10 Beta (CPU Benchmark) | 21521 | 11810 |
| TMPG Inc. * TMPGEnc 4.0 Express rev 4.3.3.9999 Beta (seconds) | 53 | 72 |
The 3DMark *06 V1.1.0 Pro CPU scores didn’t seem to reflect a heavily threaded benchmark so it didn’t really show much of a difference. I’m not sure what to make of the TMPGEnc 4.0 encoding numbers other than the fact that it appears to not scale too well beyond 4 cores. TMPGEnc appears to be one of the latest applications that added SSE4 support for vastly improved video encoding.
I was in the demo booth for the SSE4 Penryn demos and Adobe Premier was also one of the applications being showcased with SSE4 support and it seemed to scale perfectly up to 4 threads but it wasn’t demoed on Skulltrail which as 8 cores. I’m not sure how it compares to a pre-SSE4 processor but a beta version of DiVX was showing more than a doubling in performance per core at the same clock frequency compared with the existing Core 2 Duo 65nm processors.
September 18th, 2007
Intel demos Nehalem 2nd generation 45nm CPU running Windows
Intel not only showed off its next generation 45nm Penryn processors at Intel fall IDF 2007 during CEO Paul Otellini keynote, they showed a three-week old second generation 45nm Nehalem processor running Windows XP.
What’s notable about the Nehalem processor is that not only does this processor improve upon the execution engine in Penryn; it also includes a brand new memory architecture called “QuickPath” (AKA CSI) that will vastly improve the memory subsystem to allow the chip to scale multiple-cores. Each Nehalem processor will have 4+ or 8+ cores with two threads per core which means it will be able to run 16 execution threads in hardware. Current x86/x64 processors from Intel and just recently AMD have 4 cores and a total of 4 execution threads so Nehalem should be a massive jump in thread count, scalability, and performance. [Update 3:00PM - I got a clarification that the first version of Nehalem will be 4 cores in 2008, the second version of Nehalem will be 8 cores on a single monolithic die in 2009.]
Pictured below is Glenn Hinton (Chief Nehalem Architect, Intel Fellow) along with CEO Paul Otellini demonstrating one of the first Nehalem processors that barely got “taped out” a month ago. The PC booted Windows XP and played back an audio file with a synthetic computer voice saying “I am Nehalem, I am three weeks old”.
September 18th, 2007
Is the 50 watt PC possible using commodity parts?
I’ve been testing a bunch of systems from AMD and Intel in the last few months trying to build the ultimate low-power PC using commodity components and I can say I’ve gotten close. I wanted to get to the magic 50W marker at least for idle power consumption which is very important for desktop computers because they spend most of their time close to idle and this is essential for a PC that stays on close to 24 hours a day. Keeping the peak power consumption below the 100W marker is also important if you’re trying to contribute to medical research projects like folding at home.
The good news I can report is that I got very close and what allowed me to get close to the 50W marker is a high-efficiency power supply though it looks like it still isn’t small enough.
First I will disclose the shared components and methodology on the test systems
- SeaSonic 330W S12 PSU
- Two DDR2-667 DIMMs
- 120mm chassis fan (1 watt)
- 3.5″ Seagate Barracuda 160 GB 7200 RPM HDD (on AMD systems)
- 3.5″ Seagate Barracuda 500 GB 7200 RPM HDD (on Intel systems). This drive HDD uses one watt more than the 160 GB model.
- 16x DVD +/- R/RW burner
- USB keyboard
- USB mouse
- Peak CPU loads generated with WPrime
PC power consumption at system level
Note a 1 watt handicap on the Intel based systems from HDD difference.
To get an idea of how these CPUs perform, Tomshardware has this database of speed comparisons between a whole slew of systems. Note that the AMD BE-2350 and Intel E6750 2.66 GHz processor are highlighted in red on that comparison chart. The AMD BE-2350 based system is slower but its peak power consumption is higher than the two Intel CPUs. I haven’t had time to test an Intel E2160 1.8 GHz processor yet but that processor takes even less power than the relatively new Intel E6750 2.66 GHz processor.
The AMD BE-2350 “45 watt TDP” low-power processor designed for the HTPC (Home Theater PC) market does indeed idle at the lowest power level but it’s not much lower than the other systems. The Intel GD965WH motherboard is an older board based on the G965 which had a much less efficient chipset which was fixed with the “3-series” chipset motherboards like the G33 and P35 motherboards.
The Gigabyte motherboard in the chart above is based on an Intel G33 integrated graphics chipset motherboard and coupled with a new Intel E6750 2.66 GHz CPU it was able to show some impressive energy efficiency and performance. When the CPU voltage was dropped from the normal 1.25V to 1.0V and locked to a 6x CPU multiplier forcing the CPU to operate at a constant 2.0 GHz clock speed, it produced some stellar idle and peak power measurements of 52W and 63W. That would put it in to the notebook range. With a much smaller power supply, we could easily see low 40s on idle wattage.
It is interesting to note that most review sites on the web are measuring 60 watts or more for the AMD BE-2350 system and that’s almost certainly due to the fact that they’re using larger power supplies. Because I’m using a 330 watt power supply, I’m at about 15% of the rated capacity which already puts me below the ideal 25% to 75% range that a power supply operates at. So even if my power supply is 80% or more efficient, it might only be 65% efficient at 15% utilization. If someone uses a 500 watt power supply, then they’re only operating at 10% utilization which drops down in efficiency even worse which is why I’m one of the few people measuring well below 50 watts. If I can find a very efficient 150 watt PSU which takes me out of the commodity market, then I can get 85% efficiency, then I can get my idle power consumption down into the 40 watt range.
September 14th, 2007
AMD's Barcelona benchmarks: Omissions and disclosure matter
[Update 9/16/2007 - After doing a little more digging, it appears that SPEC's rules require the clear statement of a basis of comparison www.spec.org/fairuse.html rule 3.b. Without a basis of comparison, there's no telling what the marketing department is talking about. Without the basis of comparison, it's like calling someone the best pitcher in baseball without remembering to fully disclose the basis of comparison that:
"he is the best at E.R.A., but not strikeouts, during day games that are part of a doubleheader on the home field with temperatures above 90 degrees against one-legged batters whose mothers are named Susan"
So if I were to say Michael Jordan was the best basketball player of his time, I do not need to list any qualifiers just as Intel doesn't need any qualifiers for posting the best published SPEC score. But if I said Yao Ming is the best player in the NBA today without any qualifiers, then that would be a problem. If I put an * next to Yao Ming's name and then put "* Among Asians" on the bottom of the slide, that would be acceptable. AMD failed to put * no auto parallelization on the bottom of the slide and that is a violation of SPEC's acceptable use rule.]
A lot of people in the press have been scratching their heads at some off the performance claims made by AMD at the Barcelona launch Monday night including my colleague Tom Krazit who also covered the event for news.com (sister site to ZDNet). Not only were results omitted for questionable reasons, but they were omitted without a proper disclosure. The media then picked up on these slides and presented it to the public as is.The slide in question (shown below with link to full size version) claimed to be a comparison of the best published SPEC CPU 2006 (Standard Performance Evaluation Corporation) results of the best Intel compiler results versus the best PGI compiler results as of September 9, 2007. What AMD didn’t tell you was that they omitted better Intel results because of an optimization technique called “auto parallelization”. I only learned of this because I’ve compiled the officially published scores from SPEC for the Barcelona launch and AMD’s numbers didn’t match mine and I asked AMD about the disparities.
AMD used these slower Intel E5345 results which had auto-parallelization turned off instead of using the best Intel E5345 results which would have meant AMD’s 2350 should have trailed by 13.4% and not just by 5% as shown in the slide. Nothing at the live event disclosed this omission. AMD Spokesman Phil Hughes told me that the link to the slower SPEC results were included in some of the slides sent to the media but you would have had to follow the link and then manually look up all the compiler flags which is fairly cryptic.
[UPDATE 9/30/2007 - Arian Wong points out that the charts are badly skewed with a non-zero baseline.]
I asked AMD why they didn’t disclose omitting auto parallelization (referred to as autopar in quote below) numbers from Intel. AMD Spokesmen Phil Hughes and John Taylor sent me the following responses.
Phil Hughes: As I mentioned after I sent you those slides yesterday, I didn’t include the backup that had the SPEC links. As John indicated below, it is something we will consider doing moving forward. Does Intel indicate Autopar is turned on for their performance charts?
John Taylor: Autopar is a (SPEC legal but must be explicitly disclosed) way to minimize the Intel memory bottleneck. Instead of running the full SPECint workload, it runs a fraction of the workload on multiple paths. We’ll look into including a notation that we are sourcing against the more appropriate, but non-autopar, number. Because we don’t have a memory bottleneck, we have no reason to do autopar testing that reduces the workload. So for AMD, autopar is inherently a non apples-to-apples SPECint comparison.
But wait just a minute, every minor detail is already spelled out ad nauseam on the actual SPEC disclosure page (example here). To turn the question around on Intel as if they did something underhanded in including the best results with all the optimizations seems strange to me since it is accepted practice to quote the best official results from SPEC on marketing slides without having to regurgitate every minor tweak that’s already fully disclosed on SPEC. I once made the mistake of not labeling something as an “estimate” since it wasn’t officially published and I learned first hand how rigorously SPEC enforces its policies when I received an email and formal snail mail asking me to immediately fix my error. SPEC does not require that you post the entire disclosure so long as the dates and the numbers are accurate, but to silently omit your competitor’s best scores and then present it as the best score available as of September 9th 2007 is not ok. As a result of this, the media walked away with a false impression and reported it as news which is why AMD needs to be called out on this one.
Switching gears to AMD’s technical explanation for its choice, John’s explanation that auto parallelization helps minimize Intel’s memory bottleneck seems strange to me because the more parallelization of code there is, the higher the memory utilization. But even if it did get around the memory bandwidth problem for Intel, why would someone who consciously purchased an Intel compiler choose to cripple a feature that gives you better performance? If auto parallelization is an underhanded technique that improves application performance, PLEASE give me more of those underhanded techniques. Intel’s 10.0 compiler auto parallelization feature will also boost AMD processor performance and Intel has explained to me in the past that their compilers produce some of the best performance results for AMD processors. Furthermore, AMD’s favorite compiler PGI also supports auto-parallelization and here’s an excerpt below.
“In addition to the data parallel capabilities of the PGHPF® compiler, the PGI CDK package includes the PGF95(tm) Fortran 95 compiler; the PGF77® FORTRAN 77 compiler; the PGCC® ANSI C compiler; and the PGC++(tm) ISO/ANSI-compliant C++ compiler. All of these compilers support automatic parallelization for SMP workstations using a simple compiler switch, and full native support for OpenMP directive-based SMP programming.”
So is AMD promising a moratorium on auto parallelization from now on or only when it hurts Intel CPUs? But cherry picking Intel processor scores doesn’t seem to be the end of it and it seems that AMD is even willing to omit better Opteron dual-core scores to make its new generation Barcelona processor improvements look bigger than they actually are.
One of the first performance slides that were shown at the Barcelona launch showcased the performance boost of AMD’s new “Barcelona” Opteron 2350 quad-core CPU over AMD’s previous generation Opteron 2222 dual-core processors on various SPEC benchmarks. When I saw the numbers, I had to wonder how AMD turned a 46% improvement over the 2222 on SPECint_rate2006 in to a 57% improvement showcased by AMD at the launch of Barcelona. The 46% improvement number is what I derived from officially published SPEC scores as of September 9th 2007 so I thought it was strange when the AMD Barcelona slides indicated a 57% improvement.
I contacted AMD for an explanation and Phil Hughes explained to me that the best 2222 SPECint_rate2006 results from Sun Microsystems (published June 26th 2007) were omitted because they were “8222SE”. “SE” is AMD’s designation for high wattage 120W TDP (Thermal Design Power) CPUs instead of the normal 95W TDP parts that scored 56.4 on SPECint_rate2006 from Fujitsu Siemens. Hughes argued that this is a fair comparison between the “Barcelona” Opteron 2350 quad-core and Opteron 2222 dual-core because both chips operate within the 95W envelop. But that would imply that the 120W 2222SE somehow runs faster because it can use more power and that’s ridiculous because there is no performance difference between an “SE” high-wattage part versus a normal-wattage part.
The only difference between the Opteron 2222 and 2222SE is the fact that the 2222 lower-wattage part is a better quality yield that happens to leak less power than the 2222SE. The performance difference comes from differences in the system and/or compiler Sun implemented in the system they benchmarked and Sun’s system showed the true potential of the Opteron 2222 processor. Had Sun used a regular 2222 in the exact same server with the same binaries used to benchmark the 2222SE version, it would have gotten the same results (within normal variations). By omitting the best 2222SE results, AMD can claim that their new Barcelona 2 GHz product is 57% better than the previous generation 3 GHz product when in fact it was only 46% better.
September 13th, 2007
Why the ban on mandatory RFID implants should be Federal
The California legislature recently banned employers from mandating RFID (Radio Frequency Identification) implants for their employees. While I’m glad I’m covered in my state, why isn’t this ban being implemented at the Federal level to cover every citizen? I’m not suggesting that we ban the devices; I’m suggesting that no one should be forced to stick on of these in their body just to get a job. I’ve covered the issue of RFID many times before and I’m not fundamentally opposed to RFID technology or RFID implants, but I do oppose the idea that anyone should be forced to implant one in their body and it would be just as offensive if my employer asked me to tattoo a bar code on to my forehead.
It would be just as offensive if my employer asked me to tattoo a bar code on to my forehead
Verichip RFID implants are worthless from a security standpoint because they’re essentially passing clear text data over the radio waves and it can easily be cloned. If it’s cloned, you’ll have to undergo knife treatment to get a new one unless the chip is reprogrammable. Even if Verichip stopped using clear text authentication and switched to strong NSA Suite B grade crypto, I wouldn’t want it inside my body. Is any material item in this world worth life or limb? If someone wants my access device and password at the point of a gun, I’d give it to them. I don’t want them to have to cut it out of my body.
Last summer there were some issues raised about the privacy and safety of RFID enabled passports. While the scenarios were arguably remote and the privacy concerns overblown because someone can copy the same information from a regular passport, there is no reason to have the RFID in the passport since an optical or contact based system would have the same effectiveness. RFID in the traditional sense gives you more flexibility and convenience because of its long wireless range but the usable range for RFID passports is literally a few millimeters away. RFID in the Passport implementation is effectively a contact based solution that has none of the flexibility but all of the security liabilities of a wireless solution.
What about the argument that we need RFID implants for our children? I have two kids and I can tell you that RFID isn’t going to make me feel any better. First of all, that RFID implant isn’t going to be a “LoJack” device for children and you’re not going to be able to track them down if they’re abducted unless you’re within a few feet of the child. Second, having the RFID implant might mean the abductor will cut it out of your child to take out the implant. I might consider an external device hidden in a watch or something that has an active transmitter with some effective range but implants are simply out of the question.
As critical of RFID as I am, I’m not so sure why some people are so anti-RFID that they don’t even want the devices to exist in the first place. RFID implants can make sense in medical areas. If it makes it easier for emergency workers to identify a patient’s special needs, that’s great so long as the consumer gets to voluntarily place it in their own body. There’s also new technology being developed for diabetics where the RFID sensor can wirelessly report glucose levels without you having to prick your finger every day. RFID inventory tracking and logistics can simplify and automate many things so we must distinguish between good RFID devices and bad ones.
September 12th, 2007
AMD: Is closing the quad-core deficit enough?
AMD kicked off the “premier” of their latest microprocessor offering with a launch party at the Herbst International Exhibit Hall Monday night. Partners like VMware, HP, Dell, Sun, Oracle, IBM, Microsoft and others were on hand or were there by video link to celebrate the launch of AMD’s single-die quad-core milestone processor. Barcelona is critical for AMD since it was getting battered by Intel’s ten-month head start in quad-core processors which used a cheaper-to-manufacture dual-die process.
AMD argues that its single-die “native quad-core” process with its lower latency is architecturally superior to Intel’s dual-die process. But the challenges of manufacturing a massive 283mm squared die with high yields and high clock speeds is daunting and the fact that Barcelona is 6 months late and 600 MHz short makes this fact painfully clear. AMD’s executive VP Mario Rivas admitted back in March that he wished AMD had “immediately done a MCM - two dual cores and call it a quad-core” if he could do it all over again. Intel takes the easy manufacturing route of combining two 143mm squared dies which allows Intel to mix and match the best combinations. Intel’s soon to launch 45nm chip takes the die size down to an even more manageable 107mm squared.
Earlier this year, AMD had told several news organizations such as ZDNet and TGDaily that Barcelona will outperform Intel’s Clovertown 2.66 GHz quad-core processor by margins of 20 and 50 percent on both integer and floating point. It was initially implied that AMD was comparing a 2.6 GHz Barcelona processor, but it wasn’t confirmed until AMD distributed 2.6 GHz benchmarks in July with similar performance claims. The actual launch speed for Barcelona was 2.0 GHz and it fell short of AMD’s original claims by a significant margin. The actual benchmarks which were leaked to me last Friday which are now confirmed by published SPEC.org results indicate a 24% deficit on SPECint_rate2006 and a 15.4% lead on SPECfp_rate2006 over Intel’s best two-socket processors which is a far cry from the 20 and 50 lead claimed by AMD earlier in the year.
<Next page - SPEC CPU 2006: Intel Clovertown/Tigerton and AMD Barcelona>
September 11th, 2007
Sun shows 2U 16-core Barcelona server with FSB1600
After last week’s launch of Sun’s 2U 16-core Intel Tigerton server last week, Sun followed up with an AMD Barcelona version last night at AMD’s Barcelona launch. Sun kept the same compact 2U footprint while keeping all the features. So far, no other Server vendor has attempted to squeeze this many goodies in to a 2U chassis. Sun’s CEO Jonathan Swartz talks about Sun’s entrance in to the x86 world two years ago and the success they’ve had. Swartz announced the intention to take over 4th place in the x86 Server Market and these new 2U four-socket servers appear to be Sun’s strategy to differentiate itself.
The new Barcelona four-socket server comes with dual TCP offloading enabled gigabit NIC cards, redundant power supplies, and 32 DIMM slots for up to 256 GBs of memory capacity (128 GB is more realistic now since 8 GB DIMMs are priced out of this world for now). In the image above, the hand is holding a tray with two Opteron quad-core servers and 16 DDR2 DIMMs. Below that tray are two more quad-core CPUs and 16 more DDR2 DIMMs. Half the memory and CPU are stacked on top of the other half and this is a rather unusual but innovative design.
What’s notable about this system is that it’s capable of running DDR2-667 or DDR2-800 though the 800MHz DDR2 DIMMs and the higher clocked Barcelona processors haven’t launched yet. Due to the limitations of DDR2 memory, only the first half can be populated if maximum memory clock speeds are required. Once we go past 4 DDR2 DIMMs, the memory clock speed will drop down one notch. DDR2-667 memory will drop down to DDR2-533 and DDR2-800 memory will drop to DDR2-667.
Sitting in front the CPU and memory bank are two rows of redundant cooling fans which are hot-serviceable. You get an optical drive in the front along with 8 hot-swap SAS 2.5″ hard drives that can be configured for up to RAID Level 6 though you’ll most likely want to configure it for four independent mirrored sets of RAID Level 1. Having all that redundant storage at your disposal means you can safely store the hard drive images for the 128 virtual guest machines you’ll be hosting on this virtualization monster. This could possibly be one of the fastest HPC servers in its class with DDR2-800 memory since AMD’s processor architecture dominates on memory bandwidth.
For those who prefer a blade server, Sun also showed off a blade version of this server.
The new AMD Barcelona processors have a new feature called “Rapid Virtualization Indexing” (nested paging) which improves Virtualization performance. AMD announced a 79% improvement on VMware’s benchmark (beta version that supports nested paging) on the new Barcelona 2.0 GHz processor over existing Opteron dual-core 3.0 GHz processors. However, it’s difficult to compare this to Intel since Intel claims a 167% improvement on the vConsolidate virtualization benchmark with Tigerton over previous generation Intel Tulsa four-socket servers. The benchmarks are different and the baselines are different and there’s no way to tell which platform runs better until we can compare apples to apples. Both these benchmarks have been submitted to SPEC.org but it is very confusing at this time. I’ll try to clear some of these questions up at VMworld tomorrow and in the coming weeks.
September 7th, 2007
Leaked - AMD Barcelona versus Intel Clovertown and Tigerton
[UPDATE 9/10/2007 - My leaked extrapolated scores below are now officially confirmed on SPEC.org for Integer and Floating Point. Since I only had two significant figures on percentage gain to work off of from the leaked slides, I was off by up to 0.3 which is to be expected. AMD's official Barcelona launch announced newer scores submitted on 9/9/2007 which are now one point higher on SPECint_rate2006 moving up to a score of 89 which isn't too much of a change. The SPECfp_rate2006 scores may have changed up half a point or so but it's hard to tell given the fact that AMD only gives two significant digits off their percentage gain.]
[Update 9/9/2007 7:30PM - IBM seems to have confirmed these leaked numbers with this PDF file labeled for release on 9/10/2007. The scores are for an AMD Opteron model 2347 1.9 GHz quad-core and they're a little lower than the Barcelona 2.0 GHz scores I extrapolated from the leaked AMD slides for Monday's Barcelona launch. It's still quite possible there are better scores for AMD's Barcelona but IBM's SPEC CPU 2006 scores seem to suggest that the numbes below are reasonable.
[Calculations on Intel advantage fixed at 7:30PM]
IBM also gives us the single threaded results for SPEC CPU 2006 which allows us to see how fast each core can perform on its own. This is very important for desktop applications (like games) and even server applications because there are some tasks that will always be single-threaded by nature. From IBM’s latest data, we can see that Intel still holds more than a 31.2% advantage on SPECint_2006 performance and more than an 18.2% advantage on SPECfp_2006 at 1.9 GHz. This is calculated by looking at the Barcelona 1.9 scores of 11.3 and 11.2 on SPECint_2006 and SPECfp_2006 versus an Intel E5335 2.0 at 15.6 and 14 adjusted down by a ratio of 1.9/2 which is conservative performance for Intel at a theoretical 1.9 GHz.
[Estimates of Barcelona over K8 advantage added at 7:30PM]
Over an existing Opteron “K8″ 2.0 GHz processor that gets 10.7 on SPECint_2006 and 10.9 on SPECfp_2006 adjusted down by a ratio of 1.9/2, the Barcelona 1.9 GHz core boosted SPECint_2006 by ~11% and SPECfp_2006 by ~8%. This pretty much confirms what I’ve been saying that Barcelona will not significantly improve single threaded IPC (Instructions Per Cycle) over AMD’s K8 architecture.
Intel still holds a massive clock-for-clock lead when it comes to single threaded performance and this is primarily due to the 4-issue execution engine in Intel’s Core Micro-architecture versus AMD’s 3-issue execution engine. What allows AMD to narrow the gap when we get to multiple CPU sockets is the Integrated Memory Controller per Processor and HyperTransport memory architecture. But in order for AMD to beat Intel on performance, they will need to ramp their CPU close to 3 GHz but they have a long way to go at this point.]
It appears that some slides for AMD’s Barcelona launch next Monday has come in to my possession and SPECint_rate2006 benchmarks were included. AMD caught wind that I was going to be analyzing and posting the leaked data and they just called me explaining that they wanted to give me some updated slides. I don’t know exactly what it was but I know that some of the numbers posted for Intel on SPECfp_rate2006 were outdated. I explained that I never signed (or verbally agreed to) an NDA on these slides and I didn’t intend to sign an NDA at this late point in the game and the gentleman from AMD was fine with that. But once I explained what I was going to do, AMD’s rep asked me for 30 minutes to check on some numbers.When he called back, I never got any updated slides but I did get a strong hint that I might be posting inaccurate information on Barcelona performance and that maybe I wouldn’t be journalistically accurate. The gentleman strongly hinted that the numbers may change on the official launch next Monday compared to the numbers that I’ve extrapolated for now. Now this sounded nothing like the first call and it sounded to me like AMD just didn’t want the numbers posted. Of course it’s quite possible that newer SPEC benchmark numbers were submitted and the slides that I have are using older numbers. Since the word “change” was used, that tells me that the numbers I have are legitimate but they may have been tweaked a bit since.Since the leaked slides indicated that the performance estimates were submitted to SPEC.org on 8/20/2007, that looks fairly new to me though it does not rule out the possibility of newer and better numbers for AMD. I *cannot vouch* for the accuracy of the Barcelona leaked numbers and I cannot rule out the existence of newer Barcelona numbers. But since the slides look authentic and the performance numbers are within the expected range, I’ve decided to post the performance numbers and let the reader decide if they’re authentic or not.
Barcelona SPEC CPU numbers allegedly submitted 8/20/2007[Update 11:00PM] Added September numbers for Intel. Note that it is possible that AMD has numbers newer and better than the ones submitted on 8/20/2007 in the leaked slides so it’s possible the number for AMD will go up a little.
The numbers here are as of August 2007.
Barcelona SPEC CPU numbers allegedly submitted 8/20/2007The 160 score for the Barcelona 8350 2.0 fall within 4% of my “mid 150s” estimate. Unless AMD has better numbers for Monday’s Barcelona launch (they are hinting they do), it appears that even a Tigerton 1.86 GHz 4-socket Server beats a Barcelona 2.0 GHz 4-socket Server on SPEC CPU 2006 16-thread integer performance. This latest picture on integer performance also explains why AMD priced Barcelona so low against Intel.[Update 9/8/2007 - Calculation fixed for Barcelona. 35% advantage should have been over August E5345 scores and not off September E5345 scores.] Here are the SPECfp_rate2006 peak numbers which are used for the niche HPC market. The leaked Barcelona 2.0 numbers are based on AMD’s slide that claims a 35% advantage over Intel’s 5345 as of 8/28/2007 (this is the date that these slides were generated). The numbers below show updated September numbers for Intel but it is possible that AMD has newer and better scores than the ones submitted on 8/20/2007.
September 7th, 2007
Will AMD Barcelona kill off high-margin AMD CPUs?
AMD was riding high a year ago dominating sales in the dual-processor Server market and virtually owned the multi-processor (4+) Server market. Then Intel launched the Core Micro-architecture on the desktop and dual-processor Server market and momentum swung from AMD to Intel overnight in both of those markets. The one exception was the multi-processor Server market where AMD continued to reign virtually unchallenged. That all changed this week when Intel launched the 7300 series “Caneland” platform with the “Tigerton” 7300 series CPU.
Intel flips fortunes on multi-processor servers
Intel went from being a virtual non-player in the multi-processor Server market to a dominator overnight on everything except for the niche HPC (High Performance Computing) market. AMD quickly responded by saying that Intel’s Tigerton CPU “falls short”. But how much more of a beating does AMD want if a 91% increase over AMD’s 8222 3.0 GHz 4-processor server on SPECint_rate_base2006 isn’t enough? While SPECint_rate2006 already represents a lot of mainstream workloads, Intel also broke 4-processor x86/x64 records for SPECjbb2005, SPECweb2005, and TPC-C by large margins.
Next week’s “secret” Barcelona launch
One of the worst kept secrets is next week’s imminent launch of AMD’s new Barcelona quad-core CPU. Even AMD admitted in their response to Intel’s 7300 platform that Barcelona will be a few days away. Even the leaked prices have been widely distributed and AMD fans are cheering at the arrival of sub-$300 Barcelona quad-core DP (Dual Processor) CPUs and the $1004 (at quantity 1000) top bin Barcelona 2.0 quad-core MP (Multi Processor) CPU.
Note: AMD declined to comment on these prices saying that they won’t comment on “speculation”. But being this close to the launch date, it’s very unlikely the leaked prices are wrong.
Barcelona: The Opteron dual-core killer
If we stop to think about the ramifications of cheap low-frequency Barcelona processors, AMD could be facing a real pricing dilemma. Barcelona at 2 GHz will only threaten high-margin AMD CPUs but it will not threaten high-margin Intel CPUs because the clock speed isn’t high enough yet. When a $250 Barcelona DP quad-core kills a $750 Opteron DP dual-core processor on performance, who will buy that $750 part? When a $700 Barcelona MP quad-core kills a $1600 Opteron MP dual-core processor, who will buy the $1600 part? Barcelona will effectively erase all of the high-margin CPUs in their line-up next week when it launches.
AMD has priced the Barcelona processor for volume which is critical to keeping those Fabs busy and critical in maintaining market share. This is pretty much what Intel has been forcing AMD to do for the last year in the desktop market and AMD had to choose between losing margins and losing market share. Three quarters of back-to-back ~$600M losses makes it clear which choice AMD made.
What makes a quad-core a quad-core?
Of course this wasn’t what AMD intended because they had planned on launching Barcelona in the mid 2 GHz range but things fell through. AMD likes to boast about their “native quad-core” technology where they build a quad-core processor out of a single huge 283mm squared die, but the brutal challenges of manufacturing this ambitious chip has meant that AMD has had to ship Barcelona 6 months late and 600 MHz lower than originally planned. If underperformance and delays is what “native quad-core” buys you, Intel’s strategic decision to hold off on native quad-cores until late 2008 seems to be the right choice.
It’s no wonder AMD’s executive VP Mario Rivas admitted back in March that he wished AMD had “immediately done a MCM - two dual cores and call it a quad-core” if he could do it all over again. Intel’s executives have stated that single-die processors are nice in concept but even Intel with its manufacturing expertise would not attempt it in 2007. Those two dual-core dies on a single Intel Tigerton processor seem to be smashing plenty of performance records just fine so it appears to be a wise decision. Note that it isn’t just a trivial matter of “gluing” the dual-core chips together to make a quad-core processor; it takes some serious packaging skills to pull it off correctly.
Predictions on Barcelona performance
Barcelona will launch at 2 GHz which is a full 600 MHz lower than originally thought. At 2 GHz, there is simply no way it will come close to reclaiming the performance crown for AMD for the majority of the market. AMD proudly displayed their SPEC CPU 2006 floating point performance numbers and they will reign supreme in the niche High Performance Computing market. AMD has refused to comment on SPEC CPU 2006 integer performance which represents the workloads that the majority of the market demands.
To get an ideal of how AMD’s Barcelona 2.0 GHz product will perform on SPECint_rate2006, I ran some calculations and I’ve compared my numbers with other reputable sources and their calculations are in line with mine. I estimate that a Barcelona model 8350 2.0 GHz 4-socket will probably have a peak SPECint_rate2006 score in the mid 150s plus or minus 5% which surprisingly positions it slower than Intel’s low-voltage Tigerton L7345 4-socket 1.86 GHz processor which scored 166. [Update 9/8/2007 - Leaked - AMD Barcelona versus Intel Clovertown and Tigerton]
However, the Barcelona 8350 will most likely perform better than Intel’s low-end E7320 2.13 GHz processor which has half its cache crippled. Because the Barcelona 8350 is priced below the Intel E7320, AMD will be very competitive on the low-end of the market but its high-margin parts will be virtually non-existent until AMD can crank up the clock speeds on Barcelona. This is the same margins for market share game AMD has played in the last year. If AMD can deliver on 2.3 GHz Barcelona parts by the end of the year, then they will be in a much healthier position but they won’t be able to regain the performance crown until they can hit the high 2s. But by the time AMD ramps closer to 3 GHz, we’re looking at Intel Tigerton’s 45nm successor called Dunnington so the pressure from Intel is unrelenting.
September 6th, 2007
Sun packs big bang in tiny 2U Intel 7300 16-core server
One of the servers previewed at Intel’s Caneland/Tigerton launch yesterday was Sun’s full-blown 4-socket 16-core Intel Xeon MP 7300 series server. What was unique about this server is the fact that it was the only 2U server shown yesterday with all the amenities you’d expect from a high-end 4-socket server to make an ex-IT guy like me blush. All the other servers from HP, IBM, Dell, Fujitsu, Lenovo were all 4U in size.
It comes with dual TCP offloading enabled gigabit NIC cards, redundant power supplies, and 32 DIMM slots for up to 256 GBs of memory capacity (128 GB is more realistic now since 8 GB DIMMs are priced out of this world for now). In the image above, you can see the hand holding the memory tray that holds 32 DIMMs which neatly sits on top of the 4 record-breaking “Tigerton” 7300 CPUs. Sitting in front the CPU and memory bank are two rows of redundant cooling fans which are hot-serviceable.
You get an optical drive in the front along with 8 hot-swap SAS 2.5″ hard drives that can be configured for up to RAID Level 6 though you’ll most likely want to configure it for four independent mirrored sets of RAID Level 1. Having all that redundant storage at your disposal means you can safely store the hard drive images for the 128 virtual guest machines you’ll be hosting on this virtualization monster. I noticed an internal USB port on the motherboard so there should be the potential to plug in a USB memory stick that can be used to boot a dedicated Virtual Machine Hypervisor.
Because you won’t be able to hot-migrate Virtual Machines off of a server that is hosting the virtual hard drive images locally, you’re going to want a pair of these servers and make sure you have an instance of your application across both servers. Virtual Machine hot-migration requires the virtual hard drives to be stored on a separate SAN (Storage Area Network). While hot-migration brings additional robustness and flexibility, it does add complexity and you need to make sure that your SAN is physically AND logically redundant. If the SAN hosted hot-migration model of Virtualization is your cup of tea, Sun (as well as HP and IBM) has a blade version of a 4-socket 7300 series server.
Note: A colleague of mine told me that his office suffered a long and nasty outage when their SAN died and all of their Virtual Machines went down because they all lost their virtual hard drives.
Of course this type of server isn’t just limited to Virtualization hosting; it’s an ideal machine for SAP, OLTP (Online Transaction Processing), Java Enterprise server, or web hosting. The Intel 7300 series platform broke records for all of these server disciplines in a four socket configuration and Sun has shown that they’re able to move beyond the realm of SPARC and do it with competence. While I haven’t had a firsthand review of the server to give it a definite recommendation, I must say that I was impressed by the form-factor and design of Sun’s latest Intel server.
On an interesting note, Sun’s latest Intel x86/x64 server is nearly triple (2.73 times) the speed of Sun’s recently launched UltraSPARC T2 server on SPECint_rate2006 peak. I haven’t seen final pricing yet but I would guess that they both end up in the same price range.
September 6th, 2007
Intel Tigerton and AMD Barcelona price list
Intel has launched the Xeon 7300 platform codenamed “Caneland” and the Core micro-architecture Xeon MP Processor codenamed “Tigerton”. Since AMD is set to launch “Barcelona” quad-core Opterons next week and DailyTech leaked the Barcelona price list, we can now compare the prices for the new MP (multi-processor platform) CPUs.
*Note that all the unit prices listed below are for quantities of 1000
** IBM X3950 M2 X4 chipset (DDR2-based)
*** Unisys 32-socket chipset
| Intel Tigerton XEON MP price list (4, 8-16**, 32*** socket) | |||
| Model | Frequency | TDP | Price* |
| X7350 quad-core 8MB cache | 2.93 GHz | 130W | $2301 |
| L7345 quad-core 8MB cache | 1.86 GHz | 50W | $2301 |
| E7340 quad-core 8MB cache | 2.4 GHz | 80W | $1980 |
| E7330 quad-core 6MB cache | 2.4 GHz | 80W | $1391 |
| E7320 quad-core 4MB cache | 2.13 GHz | 80W | $1177 |
| E7220 dual-core 8MB cache | 2.93 GHz | 80W | $1177 |
| E7310 quad-core 4MB cache | 1.6 GHz | 80W | $856 |
| E7210 dual-core 8MB cache | 2.4 GHz | 80W | $856 |
The following AMD Barcelona prices were leaked by DailyTech
| AMD Barcelona Opteron quad-core price list (4 and 8 socket) | |||
| Model | Frequency | TDP | Price* |
| 8350 | 2.0 GHz | 95W | $1004 |
| 8347 | 1.9 GHz | 95W | $774 |
| 8347 HE | 1.9 GHz | 68W | $861 |
| 8346 HE | 1.8 GHz | 68W | $688 |
Barcelona performance numbers (other than AMD estimated SPEC CPU 2006 PEAK floating point score of 69.5 for two AMD 2350 processors) have not been announced yet so we can’t compare the price/performance yet. If I were to make an *educated guess* for SPEC CPU 2006 integer performance, I would expect the top-end AMD Opteron 8350 to be close (faster or slower) to Intel’s slightly more expensive E7320 on performance. The Opteron 8350 will win on energy efficiency because of Intel’s FBDIMM memory unless we’re talking about a system based on IBM’s X4 DDR2 chipset.
[update 5:07PM - I should clarify this "win" I'm granting AMD on energy efficiency against is at the low-end. When you get to Intel's 80W E7340 2.4 GHz 8MB Cache model, you get a lot more work done which offsets the small power differential which changes the performance/watt metric significantly. [UPDATE 9/8/2007 - Furthermore, Sun's engineer explained that the Barcelona processors only support 4 DDR2-667 memory DIMMs because that's a limit of DDR2 technology UNLESS you're willing to drop down to DDR2-533 memory which Sun is not willing to do. This is why Sun's next 4-socket Intel server will have double the memory capacity of Sun's 4-socket AMD server so it isn't a black and white argument on DDR2 versus FBDIMM. This information I got isn't correct. Each Opteron chip has two channels which means any 4-socket Opteron server will natively support 32 DDR2-533 or 667 DIMMs. I could have sworn 2-socket servers had 16 DIMM slots which they do so I thought the info I got from the Sun engineer sounded funny at the time but I figured he would know better.] IBM went to a very expensive and exotic design with their X4 chipset to be able to utilize DDR2 at high capacities.
We also need to recognize the fact that the 4-socket market cares about performance and memory capacity first and foremost. Customers of 4-socket servers are willing to pay 4x the money for 2x the performance. Someone willing to pay a $10K to $20K premium up-front for just a little more performance isn’t that worried about saving $100-$200 on power per year.]
We will not know exact performance numbers for Barcelona Opteron quad-core until hopefully next week. Of course when they do release performance numbers, I’ll do a detailed price, watt, and performance shootout. AMD by all estimates will dominate memory-intensive HPC (High Performance Computing) but they will not likely threaten anything above the Intel E7320 on mainstream server performance until AMD can ramp the clock frequencies on Barcelona.
If you also want to see leaked prices for Barcelona dual-socket processors, see this post.
September 5th, 2007
Intel launches Caneland MP platform and Tigerton CPUs
Intel has announced the launch of the “Caneland” 7300 series MP (multi-processor) 4-socket chipset and the Tigerton XEON MP 65nm CPU. The clock speeds for the new chip are 1.6, 1.86, 2.40, and 2.93 gigahertz and they are based on Intel’s Core Micro-architecture which launched last summer with the Core 2 Duo desktop and Xeon 5100 series dual processor server platform. The 50 watt version goes up to 1.86 GHz, 80 watt up to 2.40 GHz, and the 2.93 GHz Tigerton CPU peaks out at 130 TDP (Thermal Design Power). [Update 9/6/2007 - See Intel Tigerton and AMD Barcelona price list and Sun packs big bang in tiny 2U Intel 7300 16-core server]
Multiple speed records in the 4-socket x86/x64 market were broken. IBM’s System x3850 M2 is based on IBM’s own X4 DDR2 chipset came in with a record SPECint_rate_base2006 score of 184. All the other servers announced today were based on the announced Intel 7300 chipset. The Lenovo R630 G7 came in second at 178 SPECint_rate_base2006 and a peak SPECint_rate2006 score of 214. By contrast, the highest performing AMD 4-socket server using the AMD Opteron 8222 dual-core CPU comes in at 96.1 base and 108 peak for SPEC CPU 2006. AMD will launch their quad-core processors later this month at 2.0 GHz which will reduce the 91% base performance advantage in SPEC CPU 2006 that Intel opened up. More benchmarks on other SPEC metrics, SAP, TPC can be seen in Intel’s press release below.
The IBM’s System x3850 uses IBM’s own X4 chipset which greatly differs from the Intel Caneland chipset because it uses DDR2 memory instead of FBDIMM (Fully Buffered DIMM). DDR2 requires massive pin count and it is expensive to implement on the motherboard, but it has significant power savings over FBDIMM. The X4 architecture uses “re-drive” chips to boost the distance between the memory and the X4 memory controller. Each M2 Server holds 4 CPUs and 32 DIMMs and four of these servers can be linked together through a proprietary interface to create a massive 16-processor 128-DIMM server. Each re-drive chip takes 7 watts and there is one re-drive chip per 4 DIMMs taking a total of 56 watts. [UPDATE 9/8/2007 - However, those 56 watts are offset by a potential savings of 256 192 watts saved from using 32 DDR2 DIMMs which results in a net savings of 136 watts. However, it isn't clear how much extra power the X4 chipset itself uses but IBM reps told me that the memory subsystem uses 25% less power than an Intel 7300 series chipset so the total system savings would be significantly less than 25%] See a gallery of the IBM X385 M2.
September 4th, 2007
No more distorted faces with content-aware resizing
If you hadn’t seen the new image scaling techniques of Shai Avidan and Ariel Shamir yet, take a look at the video above and be impressed. I was already impressed with it two weeks ago when I first saw it but the more I look at it the more I like it. Adobe seems to agree since they went out and hired Shai Avidan.
Image scaling has been one of my pet peeves and I’ve always hated how wide screen TVs scaled 4:3 images in a distorted way (or how regular TVs scaled wide screen). It would be interesting to see if this technology could be applied in real-time 24 or 30 frames per second to video. The Photoshop users will probably be happy to get a plug-in soon.
George Ou is Technical Director of ZDNet. See his full profile and disclosure of his industry affiliations.
SponsoredWhite Papers, Webcasts, and Downloads
- Easily Monitor Virtual/Physical/Cloud and Save Budget. up.time - Free Trial Uptime Software Easily manage, measure, and monitor all your physical,virtual, and cloud assets across platforms, applications, domains, and multiple datacenters. Download Now
- Top 7 Things You Should Know About Activation and Genuine Windows Microsoft Learn the top 7 things you need to know about volume activation, including product keys, experience, deployment, virtualization, resources and more. Download Now
- Getting personal with business continuity: Five critical success factors in overcoming workforce disruptions IBM Corp. An event that disrupts your business, no matter how limited or broad in ... Download Now
- See why AND is the new OR. Read "Why NetApp."
- Imagine Virtually Anything™ w/ Cisco®, NetApp®, & VMware®.
- See how much space you can save with our calculator.
- Stay current on the latest trends in our blogs.
Recent Entries
- Saying goodbye to ZDNet
- 55W PC power supply powering the dual-core computer
- Fixing the unfairness of TCP congestion control
- HDMI survival guide for home theater
- The cheapest way to do VoIP is still analog
Blogs From Our Sponsors
Top Rated
Premier Vendor Content Whitepapers, webcasts & resources from our Power Center Sponsors
Archives
ZDNet Blogs
- A Developer's View
- All About Microsoft
- The Apple Core
- Between the Lines
- BriefingsDirect
- Collaboration 2.0
- Dev Connection
- Digital Cameras & Camcorders
- Ed Bott's Microsoft Report
- Emerging Tech
- Enterprise Web 2.0
- Forrester Research
- Googling Google
- GreenTech Pastures
- Hardware 2.0
- Home Theater
- iGeneration
- Irregular Enterprise
- IT Project Failures
- Laptops & Desktops
- Lawgarithms
- Linux and Open Source
- Managing L'unix
- The Mobile Gadgeteer
- On Sustainability
- The Semantic Web
- Service Oriented
- Smartphones and Cell Phones
- Social Business
- Social CRM: The Conversation
- Software & Services Safari
- Software as Services
- Storage Bits
- Team Think
- Tech Broiler
- Technology and the Global Supply Chain
- Tom Foremski: IMHO
- The ToyBox
- Virtually Speaking
- The Web Life
- ZDNet Education
- ZDNet Government
- ZDNet Healthcare
- Zero Day
White Papers, Webcasts, and Downloads
- Top 7 Things You Should Know About Activation and Genuine Windows Microsoft Learn the top 7 things you need to know about volume activation, including product keys, experience, deployment, virtualization, resources and more. Download Now
- Unlocking Hidden Value from Investments in SAP NetWeaver Business Warehouse IBM Find out how IBM Cognos 8 software and solutions can drive maximum value from your current investment in SAP NetWeaver Business Warehouse. Download Now
- Nucleus Research Guidebook: Leveraging Value from SAP with IBM Cognos IBM SAP users: Discover why IBM Cognos software can help you enjoy less costly deployments, faster time to benefit, and greater productivity improvements. Download Now
SmartPlanet
- Thought-provoking progressive ideas on diverse topics that intersect with technology, business, and life, and matter to the world at large. Visit SmartPlanet
- More from IBM
-
- How to Drive Better Business Outcomes with Exceptional Web Experiences Download the eBook
- Driving Business Agility through SOA Connectivity & Integration Read the White Paper from IBM
- Linking Decisions and Information for Organizational Performance Read the Tom Davenport study
