17

It seems that one of the main advantage of CPUs that are marketed for servers (e.g. the AMD Opteron and Intel Xeon) over CPUs marketed for consumers (e.g. AMD Phenom II and Intel i7) is related to reliability under thermal strain.

The response given on superuser.com to the question "Advantages of server-versions (Xeon, Opteron) of CPU's" totes a common line in my searches on Google for an answer. The claims typically involve three categories of advantage (which are quite regularly repeated):

1. Heat resistance

"Server" CPUs are better able to withstand more heat for longer than their "consumer" counterparts.

Presumably the advantage would be that the "server" CPU would last significantly longer before dying if the fan cut out, or alternatively requires less cooling.

2. Performance and optimization

"Server" CPUs have faster speeds or are optimized for server tasks, for example from the link from techpowerup.com stating the advantages of a "server" version:

Tweaked memory controllers, Cache sizes, thermal outputs, voltages, testing scemes, HT links, pricing, included heatsink, speed binning, many things.

First of all opty's are tweaked for higher throughput vs latency. Also their cache sizes in reguard to their speed is also larger. Voltages and TDP are lower compared to their equivelint x2 or phenom units.

3. Features

"Server" CPUs have certain features the "consumer" CPUs do not have, e.g. "HyperThreading and Virtualization Technologies" (from the Superuser link), or support for error correcting code (ECC) memory.

Of the above advantages commonly toted as the basis for CPU manufacturers selling their "server" versions at significantly higher prices than the "consumer" versions, I'm particularly skeptical of the claim of thermal resilience. So let's ignore claims #2 (performance) and #3 (features).

I'm skeptical of greater reliability in general, but I think it's fair to say that a significant portion of CPU failure is, in regular consumer and server usage, related to thermal resistance (i.e. heat increases the rate of failure).

Rather than differentiation between reliability based on "server" or "consumer" branding, I'd expect that all CPUs of a similar die size would have similar thermal resistance qualities. In other words, I think die size plays a much larger role than branding.

Are server CPUs more resilient to the effects of heat than consumer CPUs of the same die size?

EDIT: Edited the above to limit the claims to thermal strain (i.e. removed "features" and "speed"), as per the comments of Fabian and others.

Community
  • 1
Brian M. Hunt
  • 17,999
  • 13
  • 99
  • 176
  • For the latter point, you have to consider that even a short downtime could have serious consequences in a commercial environment, it's not just the cost of component replacement. – Phoshi Apr 30 '11 at 23:14
  • Please provide an example of the claim. Your question is quite confusing. – Sklivvz Apr 30 '11 at 23:32
  • @Sklivvz: The link in the second paragraph to the answer on Superuser.com shows one of the quintessential examples of the claims (of the several examples linked). How might you suggest improving the question? @Phoshi: I agree, and there's nothing in the question limiting the analysis to component costs, albeit tantalizing for its objective quantifiably - one must remain conscious of the subjective cost of downtime, though only a reader may assess that based on their own circumstances. – Brian M. Hunt Apr 30 '11 at 23:53
  • 1
    @Brian, I can't in all honesty find the claim anywhere. I would suggest a much, much simpler version of the question: "Intel claims > X, Y, Z (source). Is this supported by independent evidence?" – Sklivvz Apr 30 '11 at 23:57
  • @Sklivvz: Feel free to edit the question to shore it up if you can make sense of it :) – Brian M. Hunt May 01 '11 at 00:03
  • @Brian, I can't without the claim! :-) – Sklivvz May 01 '11 at 00:09
  • @Sklivvz: In a sentence, the claim is: the Opteron and Xeon have better heat resistance, features and/or performance than their non-server equivalents (the Phenom II and i7, respectively). – Brian M. Hunt May 01 '11 at 00:18
  • 1
    There are a lot of different claims in this question, I would regard the features part as off-topic (and also not answerable in general, as nearly no two recent Intel processors have the same features enabled), the interesting on-topic claim for me would be the better reliability/heat resistence. – Mad Scientist May 01 '11 at 09:17
  • Shouldn't this be on a stack more focused on computer technology? I don't see any skeptical angle. – Scott Hamilton May 01 '11 at 13:49
  • @Fabian: I agree that the heat/durability claim is the best skeptics question - do you think it's worthwhile/practical to break it out into its own question? – Brian M. Hunt May 01 '11 at 21:14
  • I remember reading somewhere that one of the major factors in the price difference is not actually in the performance but in the quality control: CPUs that are marketed for the server market undergo much more rigorous testing that provides for a guarantee of a lower failure rate. – ESultanik May 02 '11 at 13:04
  • From what I understand, server CPUs are the "cream of the crop." Meaning, as mentioned, that they are put through the highest of quality standards. To what exact extent I don't know, but I do believe it to be the case. Also, you can't underestimate power/heat issues with a server CPU. Server CPUs need to run 24/7 without a break, being efficient and keeping cool are at a premium here. Especially if you are in a crowded server room where cooling itself is at a premium. – Mr. Shickadance May 03 '11 at 11:43
  • Three claims in one question? Does not look good to me, no matter how upvoted it is. – Suma Jul 18 '11 at 19:31
  • @Suma: I've given that issue quite a bit of thought, as it clearly seems to be a difficulty (and will surely come up again in the future). What is the best way to deal with multiple-claims in one question? – Brian M. Hunt Jul 18 '11 at 20:09
  • 1
    I think multiple questions are best. – Suma Jul 18 '11 at 20:35
  • As the current answer is about features, I suggested editing this question to be about features only, and if you still think the other two claims are clear and notable enough, then ask them as a separate questions. – Suma Jul 18 '11 at 21:22

1 Answers1

11

For the majority of server CPUs the difference is that they support multi-CPU configurations whereas consumer CPUs don't. These multi-CPU versions are where you get the very high prices that you may be talking about, but there are no consumer versions of these CPUs.

However, there are a number of direct comparisons that can be made:

For example, the Intel Xeon E3-1285 and Intel Core i7-2600 are very similar chips with almost identical specifications. If you look on the linked pages, you can see that the Xeon version supports ECC memory which is often important for servers whereas the i7 does not. There is a price premium for the Xeon, but it is fairly modest at $339 vs $294.

The Intel Xeon W3580 and i7-975 Extreme Edition are also very similar. Again the Xeon supports ECC when the i7 doesn't and again the prices are very similar at $1090 vs $1050 on newegg.com. In this case the server CPU also supports faster RAM. Where the price premium does kick in is with the dual-socket version of this CPU which is the Xeon W5590 which costs $1680 on newegg.com. This version does need additional hardware to be able to support two sockets and does also support a larger amount of RAM, but I find it difficult to believe that the price difference could be justified by those differences. I suspect that the price premium here is more about being able to charge more in the dual-socket market than anything else.

At least in these two cases, it looks like your original assertion of manufacturers charging significantly more for server versions is incorrect.

Sklivvz
  • 78,578
  • 29
  • 321
  • 428
Martin Sheppard
  • 119
  • 2
  • My mistake. The answer actually is ok as is. – Sklivvz May 02 '11 at 03:55
  • That's helpful insight comparing the price. Supposing they are priced the similarly, if the server versions are better why would one ever buy the consumer equivalent? – Brian M. Hunt May 02 '11 at 18:01
  • @Brian: Because of hardware incompatibilities for the server versions, I'd assume. I doubt the boards supporting these server-CPUs have the same featureset/price as the consumer versions. – ProdigySim May 02 '11 at 21:03
  • @Brian: If they are almost identical and consumer one is slightly cheaper, why buying the more expensive one if you do not use the additional features it offers? – Suma Jul 18 '11 at 20:38
  • @Martin: Thanks for your effort putting this answer together. I'm sorry for changing the question on you to one of thermal resistance. – Brian M. Hunt Aug 05 '11 at 18:27