Server Processors

[ben chi(f4)]

Why would I want a Xeon instead of a i7?

Does the Xeon Processor crunch data better?

 

[drescherjm]

Why would I want a Xeon instead of a i7?

These are tested more rigorously. As an end user this probably does not mean anything. If you run your business off of the machine you want as close to 100% reliability as you can get.

Does the Xeon Processor crunch data better?

No.

 

[__Miguel_]

Why would I want a Xeon instead of a i7?

Does the Xeon Processor crunch data better?

I remember at least back in the C2D/C2Q era that LGA771 Xeon CPUs were slightly different from regular Core-based CPUs. I think they were optimized for server workloads, but the difference was not that great (I'm basing my post on what I've read around the time Skulltrail was launched).

I don't know if that difference still exists, but the biggest difference should be reliability, as it has already been said.

However, do keep in mind that sometimes Xeon CPUs come in models not available for non-Xeon CPUs, like more cache, slower or faster speeds, etc..

Last, but not least, and specifically on single-socket Nehalem-based CPUs, the Xeon models have lower TDP (130W for the desktop version, 95W or 105W for the server version). At least from what I can tell, of course. Check the Intel ARK website for this info.

Cheers.

Miguel

 

[Dan_D]

These are tested more rigorously. As an end user this probably does not mean anything. If you run your business off of the machine you want as close to 100% reliability as you can get.

No.

There is absolutely no proof that this is true. AMD said this about their S939 Opterons and people take that to mean that Intel does the same thing. The W35xx Xeon's are identical to their Core i7 counterparts excluding the fact that they support ECC. The Xeon E55xx series is another matter. Those I wouldn't say are more rigorously tested so much as they have all their QPI links enabled along with their ECC feature. Hardware wise the dies are all the same. The difference is really microcode and that's it. There may be binning involved but I don't think that counts for "more rigorous testing."

If there is any additional or different testing placed on these CPUs, it's never been discussed publicly by Intel as far as I know.

I remember at least back in the C2D/C2Q era that LGA771 Xeon CPUs were slightly different from regular Core-based CPUs. I think they were optimized for server workloads, but the difference was not that great (I'm basing my post on what I've read around the time Skulltrail was launched).

No! Where does this type of misinformation come from? There are no differences between LGA771 and LGA775 CPUs excluding two things. SMP support (multi-processor capable) and the Xeon's have slightly different TDP ratings. (Sometimes, but not always.) Xeons with lower TDP ratings are simply CPUs that come from the better bin. They passed their QC tests with better thermal characteristics than the standard consumer chips may have. Given Intel's very consistent manufacturing capabilities the differences are probably less pronounced in this area than you might think. Hence why underclocking works so well on consumer line CPUs.

I don't know if that difference still exists, but the biggest difference should be reliability, as it has already been said.

Prove it. I've never seen anything to substantiate this idea. Intel's CPUs are already extremely reliable and the erratum for both chips is the same. I've seen consumer processors in servers plenty of times and they have never proven to be "less stable" than their Xeon branded counterparts. If there is more extensive (and I stress the *IF*) testing it is on the motherboard/chipset side rather than the processor side. The Xeon may go through a longer validation period with OEMs, but in the end that means very little to you or I. I can promise that aside from ASUS, few manufacturers bother to do much testing with Xeon branded CPUs in X58 chipset based boards. (ASUS actually validated the E55xx series CPUs and I don't know of anyone else that has for sure.)

However, do keep in mind that sometimes Xeon CPUs come in models not available for non-Xeon CPUs, like more cache, slower or faster speeds, etc..

This has been true in the past. Usually there are more Xeon models than there are equivalent desktop / consumer market CPUs.

Last, but not least, and specifically on single-socket Nehalem-based CPUs, the Xeon models have lower TDP (130W for the desktop version, 95W or 105W for the server version). At least from what I can tell, of course. Check the Intel ARK website for this info.

Cheers.

Miguel

That's comparing the Xeon E55xx to the Core i7. The W35xx sequence processors such as the W3520 are identical to their Core i7 920 counterparts (same TDP) excluding the fact that the Xeon supports ECC memory officially and the feature is disabled on the Core i7 920.

These CPUs are the SAME. The lower TDP CPUs were probably from the "best, or better" bins but that's it. I don't think there are differences in QC. Microcode is all that gets changed from one CPU to the next in regard to Nehalem based CPUs.

 

[drescherjm]

There is absolutely no proof that this is true. AMD said this about their S939 Opterons and people take that to mean that Intel does the same thing.

I know AMD employees still say that current Opteron cores and testing are not the same between Server and Desktop lines. For intel, I remember seeing some posts here about that but are correct that this information is not from Intel.

 

[ben chi(f4)]

So... Data Crunching go with 2xXeon Quads or 2xi7s top of the line processors.

(Once again I say price doesn't matter as the government is paying)

 

[rflcptr]

So... Data Crunching go with 2xXeon Quads or 2xi7s top of the line processors.

(Once again I say price doesn't matter as the government is paying)

You're going to have to go with the 2 x 55xx series Xeons. Core i7s do not have 2 QPI links enabled for dual processor setups.

 

[__Miguel_]

No! Where does this type of misinformation come from?

Well, after digging up almost two years of motherboard reviews on two of my favorite tech sites, I ended up with this quote, from Anandtech:

Despite using a Xeon socket, the Core 2 Extreme QX9775 isn't a Xeon. It turns out there are some very subtle differences between Core 2 and Xeon processors, even if they're based on the same core. Intel tunes the prefetchers on Xeon and Core 2 CPUs differently so unlike the Xeon 5365 used in its V8 platform, the QX9775 is identical in every way to desktop Core 2 processors - the only difference being pinout.

Intel wouldn't give us any more information on how the prefetchers are different, but we suspect the algorithms are tuned according to the typical applications Xeons find themselves running vs. where most Core 2s end up.

As I said, I was only recalling what I had read. If it's true or not, I'll let each one of you to figure it out.

Prove it.

Don't really have to. I said "should be", not "is". From my point of view, and what makes more logical sense, is that Xeon and Extreme-Edition cores are binned first from anything else. And those get tested to lower VIDs (usual on even LGA775 Xeons) and/or higher speeds. That extra validation step (you don't want to risk having a QX9770 having heat or degrading issues just because poor binning, right?) tells me that, if those CPUs aren't more reliable, they at least ought to be more reliable (especially the parts with lower VIDs, since less voltage = less electromigration).

Is that difference so big you should care? For desktop systems with a 4-year (tops) life expectancy, and not 24/7 use, no. For critical enterprise systems working 24/7 for sometimes more than 4 years, that little extra something might be the difference between a running system and millions of dollars lost in revenues (mind you, I'm thinking cascade-collapse because of system overloading after a failed node, much like happened to Google a while back on the router department - and yes, that's a long shot).

Do I think there is a great difference between them? Probably not that much (though the lower Xeon VIDs are always nice), and the reason Intel doesn't disclose failure data might actually be because the difference is so tiny that it would loose the whole 3xxx series Xeon lineup if everyone knew they are so close to each other...

That's comparing the Xeon E55xx to the Core i7.

Ups, my bad.

I somehow managed to compare the i7-940 with the X3470, which is an LGA1156 part (oddly enough, I think that happened way before the official LGA1156 lauch...) Anyway, moving on, you're absolutely right on this one.

Shame, though, that low power parts (like the LV Xeons) seldom find their way to the desktop arena (like the S series). And even more shame those low power parts that do find their way to the desktop arena are so damn expensive...

Cheers.

Miguel

 

[Dan_D]

Well, after digging up almost two years of motherboard reviews on two of my favorite tech sites, I ended up with this quote, from Anandtech:

Interesting, he might be right. I've never talked to Intel about it directly, so there could be truth to this. I don't think that difference is earth shattering and differences in performance can normally be attributed to the motherboards and chipsets which are vastly different between the two.

As I said, I was only recalling what I had read. If it's true or not, I'll let each one of you to figure it out.

Again this might be true.

Don't really have to. I said "should be", not "is". From my point of view, and what makes more logical sense, is that Xeon and Extreme-Edition cores are binned first from anything else. And those get tested to lower VIDs (usual on even LGA775 Xeons) and/or higher speeds. That extra validation step (you don't want to risk having a QX9770 having heat or degrading issues just because poor binning, right?) tells me that, if those CPUs aren't more reliable, they at least ought to be more reliable (especially the parts with lower VIDs, since less voltage = less electromigration).

Well Intel may or may not put potential Xeons through more validation tests. I have no idea. Much of Intel's manufacturing process has always been kept secret. That's not likely to change. Now parts with lower VID are most certainly from the best bins. They'd have to be.

Is that difference so big you should care? For desktop systems with a 4-year (tops) life expectancy, and not 24/7 use, no. For critical enterprise systems working 24/7 for sometimes more than 4 years, that little extra something might be the difference between a running system and millions of dollars lost in revenues (mind you, I'm thinking cascade-collapse because of system overloading after a failed node, much like happened to Google a while back on the router department - and yes, that's a long shot).

Point taken.

Do I think there is a great difference between them? Probably not that much (though the lower Xeon VIDs are always nice), and the reason Intel doesn't disclose failure data might actually be because the difference is so tiny that it would loose the whole 3xxx series Xeon lineup if everyone knew they are so close to each other...

This is what I suspect as well.

Shame, though, that low power parts (like the LV Xeons) seldom find their way to the desktop arena (like the S series). And even more shame those low power parts that do find their way to the desktop arena are so damn expensive...

Agreed.