You've made a MAJOR error in the article. Intel manufactures "Xeon" processors, and AMD manufactures "Opteron" processors. In its current state, the article is quite confusing to say the least.

twenty-something dollars saving per server per year? so for a 4-year life cycle it's what? 100 dollars? so like, for a two-CPU server it's $50/CPU savings. So basically what are we talking about? if intel's CPU is more than $50 cheaper than AMD's at the same performance level then it's a winner. And you bet it is with the Core2-based Xeons. What exactly is AMD's marketing machine bragging about with all this "energy efficiency" bull???

Who are Neal Nelson and Associates? Who funded the study? (I noticed the company thanked AMD for help but not Intel.) Was the methodology fair? What was Intel's response? Why were performance-per-watt numbers ommitted? In this industry it's rare that the favored company isn't behind a study like this. Apparently you people are either too lazy or too biased to do anything more than regurgitate a press release.

Why are you comparing an Opteron to a Xeon which is atleast 1 year old - given that there have been over 3 product releases since Woodcrest (June 2006).

Of course the intel processor used more power. to start with the power supply was 100 watts larger in the intel server. Also look at the fact that the intel Xeon Dual core CPUs have been on market for a while now and AMD is just playing catch up. AMD nuts have been at this kind of comparison for a while now.. comparing intel apples to genetically enhanced, unreleased apples (amd). Now lets talk about stability and reliability. Chipzillas got that

look closely at the specifications: two differences besides the processor. Two different power supplies are used, and FBDIMM used on Xeons and DDR used on Opterons.

Let me guess, Neal Nelson and Associates loaded up 8 FBDIMMs they can get their hands on to compare power consumption. Well news to you, that 7% less power Opteron draws is just due to the FBDIMMS. It also means that Opteron is 20% slower. So performance per watt for woodcrest is higher.

The two different power supply might have different efficiency ratings. Delta 600W on the Xeon vs Delta 500W on opteron. Conveniently "Neal Nelson and Associates" measured the power draw at the wall, totally ignoring the difference of power efficiency(usually lower on higher power ratings).

After all the power saving is a mere 20 dollars. Given the typical lifespan of 3 years for those servers, 60 bux is nothing compared to the performance difference. Woodcrest owned Opterons. so I respectfully ask this question: how much did AMD pay nelson and associates?

Thanks to everyone who has commented this far.

First, to those of you who saw the first version of the article where I confused the Opteron and the Xeon, I apologize. I fixed it within minutes, but not before the damage was done, evidently.

Second, the questions and comments many of you pose are good ones, and I am looking forward to seeing responses from Intel as well as the conversation this report generates.

To those of you who think I did a rush job, well, in a way, I agree: The purpose was to draw people's attention to the report as soon as I learned about it. But I assure you, I put out inquiries to Intel and AMD immediately. AMD was quick to provide a comment. Not surprisingly, Intel needs time, since they likely just learned of it Friday, late day.

Do stay tuned. And you might want to check out this thread on the subject over at the XtremeSystems forums: http://www.xtremesystems.org/forums/showthread.php?p=2321617#post2321617.

One other comment aside from the ones I had planned to make [already covered by others].
The 3GHz Xeon has a higher power draw than other Xeons. The slowest is the most lean on consumption, all the mid-range chips are rated at 65W, that fastest.. 3Ghz.. is rated at 80W typical.
Given the comparable performance numbers, if the consultants used 2.6GHz Xeons v. 3GHz Opterons the performance would have been on par and the thermals would even out.
Of course, given the inefficiencies of AC-to-DC power supplies, moving the Xeon box to a 500W power supply should put the power advantage to Intel.

Note: I do realize that Intel systems [ignoring CPUS] consume more power than AMD systems due to the FBDIMMS and system controller. The reality, however, is that the typical power between comparable systems is really very similar between similarly configured boxes.

I'm pretty sure Tom's Hardware or Anandtech did an article comparing two nearly identical systems (PSU, etc) and still came out with AMD being more power efficient than Intel. I read this recently.

gyll - My goal is to report power-efficiency data so customers can make value decisions like the one that you have outlined.

Jack - I have been a Chicago area VAR since 1973. Click Company Information -> Company History on my company web site http://www.nna.com for more information. I did not receive any funding from anybody. Not AMD. Not Intel. Nobody. I thanked AMD (and Novell/SUSE, and Seagate, and MySQL) because each of them loaned me some resource that I used in the test. I did not thank Intel because they refused my request to borrow a machine. I released some test data in May that did transactions/KWH (AMD won that test also) but I prefer this new methodology where I subject each machine to exactly the same transaction load.

kaiwai - This is the launch of my 2nd generation power-efficiency benchmark. I plan to test many more machines for future comparisons. These two were the ones I had available at this point in time.

matt - Both test machines were Colfax 1U rack mount servers. The Xeon came from Colfax with a 600 watt PS and the Opteron came with a 500 watt PS. Note the differences in power at Idle vs 500 users. If PS efficiency were the major factor there would not be this difference. The major change between Idle and 500 users is not in the PS but the activity within the CPU.

paidamdmoron - Colfax shipped the Xeon with an Intel MB with 8 FB-DIMMs. They shipped the Opteron with a Tyan MB with DDR-II DIMMs. I think Intel has been pushing Intel MBs (with FB-DIMMs) so I was not surprised to see that combination. You suggest a very intersting follow-on test. A Tyan or Asus MB with DDR-II and the same pair of Xeon 5160s would highlight the power impact of the Intel MB and the FB-DIMMs. But how does your FB-DIMM theory explain the difference between Idle and 500 users?

Steven Bandyk - "thermals would even out", "power advantage to Intel", "really very similar". Can you provide us with any actual test results to support these statements? I would be happy to compare 2.6 GHz Xeons to 3.0 Ghz Opterons. Can you help me borrow a pair of 2.6 GHz Xeons? I would only need them for about a week and I will run the test for free.

Looks like another propaganda from AMD!!!

Neal, if the Xeon machine was just a couple of watts higher under load then the Opteron, why don't you ask Colfax what would warrant the Xeon to have an inherently more inefficient power supply, especially when it's not even reaching 300W, never mind 600W?

Can you also give us the price of the systems? There should be no secret there. I think the AMD based would cost significantly more because of the 8xxx parts. That can't be a mistake since you did the same on the last test.

It's fascinating how everyone sees the down side. It would be exactly the same if AMD showed lower results then Intel. Cause they would say, Well you didn't use the fastest Intel chip. Guess human nature.

As for power saving. Consider not a single machine. But a cluster. Lets say 10,000 system. On the level of super computers.

So, during a 4 year cycle single system = 100$. A Super computer cluster = 1,000,000$. Guess that makes a difference.

As for FB-DIMMS. It was Intel's decision, and they went the wrong path. And from the Wild Wild Web the information seems to be that future chipsets will not support FB-DIMMS due to lack of performance vs the power it consumes.

To me the comparison is reasonable. But one thing missing is the price of the two systems. Since someone who is buying (more then 1 machine) would consider purchase cost and maintenance cost (power consumption) along with performance.

As for the Power supplies. If they are both from a single manufacturer, the efficiency will be nearly identical at same power usage. Look at various "Standard" ATX PS. Seasonic, one of the "top end". Their power efficency between at 400W for the 500W and 600W models is identical, as both power supplies are identical. So there is a good chance that these power supplies follow the same scheme. If its the same series of Power Supplies.

Additonally, people don't buy these systems, companies do. And the difference is that they don't look deeply into the configs as normal "geeks/nerds" and fanboys "AMD/Intel" like me and you. They look at cost/performance. And this article points to that fact.

And no I'm not a AMD or Intel fanboy. Just a computer Geek :) On the desktop side Intel has AMD beat on performance. But server wise, they shot off their leg when they decided FB-DIMMS and not DDR II like AMD. And now they suffer.

Luckily in computer world, things quickly change. I personally can't wait for the future to become reality.

This is a joke. The guy compares $2,149 processor (see http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_609,00.html?redir=CPT301) not yet available in retail stores to an old processor anyone can buy online for $699 (see www.pricewatch.com/cpu/).

George - Both Toms and AnandTech have recently reported Xeon versus Opteron power-efficiency.
The Toms link is: http://www.tomshardware.com/2007/07/11/energy-efficiency-intel-left-out-in-the-cold/page11.html#energy_index_amd_unbeatable
and the AnandTech link is
http://www.anandtech.com/IT/showdoc.aspx?i=3039&p=6

Both of them report that the AMD systems are more power-efficient than the Intel systems.

The mother lode of power savings is AMD's Pacifica hardware support for Virtual Operating Systems coupled with hot plug HyperTransport. Some studies report the typical server is active 7% of the time and idle 93%. A Virtual Operating System can dynamically allocate resources over the hundreds of cores in a rack to keep them busy. In case of a surge of demand another rack can be connected by hot plug. Potential savings appears to be fourteen fifteenths of hardware and power costs. Any IT manager who does not address AMD's Pacifica should be relieved of fourteen fifteenths of his salary.

Prodeous: yes you save 100 bux by spending 2000 bux more to buy the 8xxx processors vs the woodcrest. So to save your 1,000,000 dollars in a cluster, you would have to spend 20,000,000 dollars more for the processors.

back at Neal Nelson:
"Colfax shipped the Xeon with an Intel MB with 8 FB-DIMMs. They shipped the Opteron with a Tyan MB with DDR-II DIMMs. I think Intel has been pushing Intel MBs (with FB-DIMMs) so I was not surprised to see that combination."
-ok, so "Colfax made those servers" is sufficient

"You suggest a very intersting follow-on test. A Tyan or Asus MB with DDR-II and the same pair of Xeon 5160s would highlight the power impact of the Intel MB and the FB-DIMMs. "
-No, I never suggested the follow on test. I never suggest a test that's impossible. You obviously don't know that 5160s can only use FB-DIMMs.

"But how does your FB-DIMM theory explain the difference between Idle and 500 users?"
-It's not a theory. Everyone knows why FB-DIMM use more power than regular DDR. The issue of Idle vs 500 users only tell you that opteron's cool and quiet is working better than the Xeon's idling mode. But who buys a quad system to let it sit idle most of the time.

- Another thing: Nelson: as someone pointed out...the CPUs have a price differential of over 3000 dollars for the pair. AMD lost

"hoaxwatcher - An 8-way enabled chip is priced higher than a 2-way enabled chip."

That's quite the understatement. LOL. I think it's an atrocity for you to "assume" (MP makes no difference) one thing yet are set to "prove" (AMD supposedly more EFFICIENT) than others.

http://www.tomshardware.com/2007/07/11/energy-efficiency-intel-left-out-in-the-cold/page16.html Here they conclude the L2 E6300 are "the best processor" based on EFFICIENCY (work per unit, be it price or watts), then in the end, praise AMD for idling lowest. If a chip consumes 50W but finishes a task in 1s, and another consumes 10W but finishes in 100s, Toms would conclude the 10W to be better.

So let's recap. The systems have different PSUs when there is negligible difference between systems at load. You get assistance from AMD and not Intel, silly "independents" (while Anandtech makes effort to communicate to all vendors). As admitted (indirectly) by Neal, the AMD system is much more expensive, therefore, money saved on watts will not be enough to recuperate costs. And the webpage itself, all show AMD in a favorable light, and when not AMD vs Intel, use AMD.

I think it's more than a coincidence that Neal did not reveal the system's prices. This is a bigger scandal than Intel "not having perf/watt" lead (Clovertown's not even compared, wide variety of workloads are needed to substantiate that claim, this test is just 1 workload).

Based on the previous comments on the blog I will:

1) Try and re-test the Opteron with a pair of 3.0 GHz 2222 CPUs. They are about the same price as the Intel 5160s. This will give us two things. First, a reading about whether the 8222s and 2222s consume the same power and second, a comparison between Intel and AMD where the servers have similar prices as well as the same clock speeds.

2) Re-test both the Xeon and Opteron with 4 DIMMs in each machine (rather than the current 8 in each).

Can anybody tell me how to find a motherboard with dual LGA771 sockets and DDR-II RAM? Every board I have found for Xeon Core is also FD-DIMM. Is there a northbridge for Xeon Core that supports DDR-II?

I cannot move the same power supply between the two 1U servers that I have here. Both machines have a custom SAS/SATA backplane with cooling fan connectors and non-standard power connectors. I have not been able to find a Delta 600 watt for the Opteron nor a Delta 500 watt for the Xeon.

1) Are you going to get backhand low voltage engineering AMD samples and crappily binned Intel chips?
2) Clock speed is not a factor of purchase! Suppose that AMD was stuck at 3GHz and dual core, while Intel is at (theoretically) 3.33GHz and quad core. It's not "unfair" to compare 3.33 v 3 or dual vs quad. Only unfair to poor AMD maybe but more unfair to readers that they cannot get the most data. So why don't you pop in a 2.33 Clovertown and see what happens? If nothing, then that's a problem with your bench.
3) There will be eventually in the 2nd half
4) If you're willing to seek out different motherboards, why not seek new cases too?
5) Your bench is not representative of everyone's needs, therefore you can't claim "New Test Shows AMD Still More Power-Efficient Than Intel"

I'd much rather see server comparison on power utilization. The CPU metrics are meaningless, there are many places a server can save power. Your CPU consumes 100watts, but servers come with 500-1500Watt power supplies. Show me most efficient servers, besides Sun's T2000, or most efficient storage systems besides NexSAN's SATABeast.

Thanks