Loading ...
As April Fools’ Day approaches, I remember Pavlov’s dog and just how conditioned we all can be – particularly in terms of technology. Swap the “n” and “m” keys on someone’s keyboard or change the language setting on your friend’s Internet browser and see how long it takes your victim to figure out exactly what has happened. These changes are hard to catch because the assumptions you make when you sit at your computer – the location of keys or language settings of your software – aren’t meshing with reality. You have to challenge the validity of your assumptions before you can fully comprehend the situation.
If someone asked you to identify the top performing and most energy efficient server processors that have been shipping for the last four months – what would you say? If you answered Intel “Harpertown” or “Nehalem” processors you would be wrong. The 45nm Quad-Core AMD Opteron™ processor (code name “Shanghai”) has been shipping since November 2008, and has steadily gained solid marks in a variety of benchmarks, application performance evaluations, and power comparisons. More importantly, it’s gained the accolades of end customers who deploy AMD-based servers and rely not only on its performance, but on its energy saving properties and ease of management. Once again, assumptions and reality don’t always mesh.
Now you can accuse me of being an AMD “fanboy,” but the facts, not hype, support my position. Take a look at a couple of recent articles in Ars Technica and InfoWorld that make the same case I just did for “Shanghai” using third party performance and power evaluations.
One area where the Quad-Core AMD Opteron processor really shines is virtualization. Currently the AMD-based Sun Fire X4600 MS server holds the record for the most virtual machines (114) on a server with VMware’s VMmark test running with VMware ESX 3.5U3, the currently shipping version of this hypervisor. A demo video posted on YouTube showcases the ability to perform a live migration between all generations of Quad-Core AMD Opteron processors with VMware ESX 3.5U3 – even our new 6-core “Istanbul” product which is due to release in the second half of 2009. John Troyer from VMware’s VMTN Blog was a guest on the AMD Virtualization blog and he showcased the combined benefit of AMD-V™ Rapid Virtualization Indexing and VMware ESX 3.5 for scaling a web serving environment on a HP ProLiant DL585 G5 server running theSPECweb2005 benchmark with Apache web serving software. Now that’s a mouthful, but it’s a real-world scenario that could easily be taking place in enterprise data centers today.
Let’s add another dose of reality. According to the survey by Enterprise Strategy Group the average number of virtual machines per physical server is between 5 and 10 – a far cry from the record 114. Live migration, such as VMware’s VMotion, is a much in demand feature but it requires a specialized infrastructure and does not support heterogeneous (AMD and Intel) processor environments. And while VMmark and SPECweb2005 benchmarks provide a way to evaluate performance aspects of servers, they don’t take into consideration what are perhaps the two major decision factors for most IT groups – the cost of the system and its power consumption. These realities don’t make the Quad-Core AMD Opteron processor any less of a virtualization powerhouse – but it does show you what happens when you go beyond easy assumptions. As technology providers we should have an obligation to provide you with both “hype” and facts.
The server industry is on the cusp of a huge hype cycle that will go on for the next few weeks and you will be pounded with information meant to drive buying decisions solely based on assumptions. My advice – let’s not be fooled into making decisions based on automatic conditioning. Otherwise we night end up with many errors we need to correct.
Margaret Lewis is a Product Marketing Director at AMD. Her postings are her own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
(4.93 out of 5)
#1 by Robert Hallock - March 31st, 2009 at 13:12
How does AMD plan to position its Shanghai or upcoming Istanbul processors against Intel’s offering, given that the Nehalem EP now holds both the Virtualization and performance/watt records?
Anandtech: http://it.anandtech.com/IT/showdoc.aspx?i=3536&p=12
“There is little doubt that the newest Xeon is also the champion in virtualization.”
Anandtech: http://it.anandtech.com/IT/showdoc.aspx?i=3536&p=14
“Still, it is clear that when you compare the best Intel and AMD CPUs, the best performance/Watt figures come from Intel by pretty large margin.”
Or does AMD take issue with the conclusions AMD has drawn in these benchmarks?
#2 by Peter G. - March 31st, 2009 at 13:50
Margaret, I don’t think too many people will be fooled. Anyone worth their salt will research their purchase and see that the data doesn’t lie. You’re right, over the past four months, AMD’s server processors have done great things. It’s clearly lined out in your blog. But four months is a long time in hardware cycles, and things change.
Four months ago, Shanghai set the world on fire.
Yesterday, Shanghai was shanghaied.
The very sources you cite in this article have shown your competitor’s new product overall offers vastly superior performance/Watt, and in some cases does this while costing less. Instead of focusing on the past four months, lets hear about what the future holds. Honesty, integrity, and transparency are what’s needed now, not a rehash of history.
#3 by Margaret Lewis - March 31st, 2009 at 14:58
Thanks for the comments. In my blog I point out major decision factors that customers request that go beyond benchmark numbers – one is power consumption and the other is the cost of the system. Nehalem EP has posted leading virtualization performance numbers on VMmark – but this benchmark does not give you any information on power consumption while the benchmark is running or the price of the system. I suggest that you go on line to HP or Dell and price the system configurations that achieved these numbers. You can then use this system cost to figure out a cost per VM (system cost/total number of VMs). This will help you understand how the “raw” performance score correlates to a price/performance score. In terms of performance/watt – I believe you are referring to SPECpowerjbb? This shows you the power performance in one specific workload and does not necessarily give you info about power consumption for VMware, database, web serving, etc.
The Intel “Harpertown” platform will continue to be sold by Intel – if you look at industry predictions of Nehalem ramp it takes several quarters before it even reaches 50% of the market. Shanghai continues to be strong contender to Harpertown in terms of the “raw performance” and a strong contendor in price/performance against Nehalem.