What are the advantages of using server grade processors

[turtlebud]

I noticed that with the new dell precisions workstations that the standard configuration is as a server, not really a desktop which made me wonder what advantages server grade machines offer over desktop machines.

With the Mac Pros, one of the reasons why they are so expensive (well, a lot of the reason why) is because they use processors that intel created for servers and workstations as opposed to desktop computers. What is the advantage of these processors over your typical desktop processors?

(and are all the Mac Pros using server grade processors?)

 

[bozz2006]

they're constructed to run 24/7 for a long long long time. basically, they're indestructible.

 

[rylin]

SMP capabilities, larger caches, sometimes rather conservatively clocked etc. are some of the reasons they're used as workstation procs, and yes, all MPs are using Xeon CPUs atm.

 

[wfj5444]

Also ECC memory, which helps increase stability.

 

[cmaier]

It's all about the electromigration.

 

[VirtualRain]

I believe, in the case of the quad's the Xeon's and the Core i7 desktop chips use exactly the same silicon (with the exception that they disable ECC in desktop - but I'm not sure if this is done prior to fab, post-fab, or is controlled in firmware).

From what I've read, when Intel determines if a chip is destined to be a Xeon, or a desktop chip, it goes through a qualification process. The lower power consuming chips get binned as Xeon's. This is also how they determine which clock speed a chip will be branded as... higher performing chips will get branded as higher-end models (e.g. 3.2GHz, 2.93GHz, etc.).

The dual-socket CPU's have no desktop counter parts. They also have different silicon in that they include dual QPI link components on-die.

Apple could have opted to build the quad machines with Core i7 instead of Xeon processors. The difference in cost and feature set would have been negligable. Assuming ECC is only available in Xeon parts, it makes sense that they opted for a Xeon in these products.

For the Octo's, Apple and other manufacturers have no choice. Xeon's are the only dual-socket processors available.

 

[cmaier]

I believe, in the case of the quad's the Xeon's and the Core i7 desktop chips use exactly the same silicon (with the exception that they disable ECC in desktop - but I'm not sure if this is done prior to fab, post-fab, or is controlled in firmware).

From what I've read, when Intel determines if a chip is destined to be a Xeon, or a desktop chip, it goes through a qualification process. The lower power consuming chips get binned as Xeon's. This is also how they determine which clock speed a chip will be branded as... higher performing chips will get branded as higher-end models (e.g. 3.2GHz, 2.93GHz, etc.).

The dual-socket CPU's have no desktop counter parts. They also have different silicon in that they include dual QPI link components on-die.

Apple could have opted to build the quad machines with Core i7 instead of Xeon processors. The difference in cost and feature set would have been negligable. Assuming ECC is only available in Xeon parts, it makes sense that they opted for a Xeon in these products.

For the Octo's, Apple and other manufacturers have no choice. Xeon's are the only dual-socket processors available.

As a former x86-64 microprocessor designer, I'll add that they probably also bin on physical characteristics, like dielectric layer thickness, metal dimensions, etc. (as detectable using test structures). This lets them predict which processors will likely have the longest lifespan, among other things.

 

[alanlindsay]

They also use less power. i7 2.66 = 130W, Xeon 2.66 = 95W

 

[Salavat23]

they're constructed to run 24/7 for a long long long time. basically, they're indestructible.

SMP capabilities, larger caches, sometimes rather conservatively clocked etc. are some of the reasons they're used as workstation procs, and yes, all MPs are using Xeon CPUs atm.

Also ECC memory, which helps increase stability.

It's all about the electromigration.

Those are some of the most stupid comments I've read so far, especially the first one.

The E55xx are identical to the i7 consumer CPUs. Only difference is that they have a second QuickPath interconnect and slightly different instructions.

They are just as reliable as i7s.

And ECC does not help increase stability at all. Its just a safety net in case something goes wrong, which happens very rarely.

I believe, in the case of the quad's the Xeon's and the Core i7 desktop chips use exactly the same silicon (with the exception that they disable ECC in desktop - but I'm not sure if this is done prior to fab, post-fab, or is controlled in firmware).

From what I've read, when Intel determines if a chip is destined to be a Xeon, or a desktop chip, it goes through a qualification process. The lower power consuming chips get binned as Xeon's. This is also how they determine which clock speed a chip will be branded as... higher performing chips will get branded as higher-end models (e.g. 3.2GHz, 2.93GHz, etc.).

The dual-socket CPU's have no desktop counter parts. They also have different silicon in that they include dual QPI link components on-die.

Apple could have opted to build the quad machines with Core i7 instead of Xeon processors. The difference in cost and feature set would have been negligable. Assuming ECC is only available in Xeon parts, it makes sense that they opted for a Xeon in these products.

For the Octo's, Apple and other manufacturers have no choice. Xeon's are the only dual-socket processors available.

Intel does not determine which is which from inspection. They make one CPU with two QP links, one with one. And one with a certain set of instructions, the other with a different set.

They also use less power. i7 2.66 = 130W, Xeon 2.66 = 95W

i7 stepping D0 uses 95 Watts as well. Its the same CPU at its core.

 

[soldierblue]

^^Thank you.

 

[cmaier]

Those are some of the most stupid comments I've read so far ....

And ECC does not help increase stability at all. Its just a safety net in case something goes wrong, which happens very rarely.

Hey now! Electromigration is awesome! It's not stupid! It's like my favorite thing!

BTW, I was a microprocessor designer at AMD for 10 years, and was one of the original Opteron/Athlon 64 designers, so I've designed my share of workstation class chips, and also my share of consumer grade/commercial grade chips. And even though the specs are the same on paper, I can assure you that the stuff that we sold to the workstation market was designed and built to last longer.

And "ECC does not help stability... Its just a safety net...." is sort of a self-contradiction. Automatically correcting single-bit errors is a critical feature in certain markets - the last thing you want is for a data bit to flip and not get caught. If an instruction bit flips, at least you get a crash (probably), and you know something is wrong.

 

[Salavat23]

BTW, I was a microprocessor designer at AMD for 10 years, and was one of the original Opteron/Athlon 64 designers, so I've designed my share of workstation class chips, and also my share of consumer grade/commercial grade chips. And even though the specs are the same on paper, I can assure you that the stuff that we sold to the workstation market was designed and built to last longer.

AMD is different in that department. Their new 45nm processors perform very differently from the 45nm Denebs. And they run at way different TDPs as well.

Intel has just been adding an extra QP/FSB link and some minor tweaks to their Xeons since PIII.

And "ECC does not help stability... Its just a safety net...." is sort of a self-contradiction. Automatically correcting single-bit errors is a critical feature in certain markets - the last thing you want is for a data bit to flip and not get caught. If an instruction bit flips, at least you get a crash (probably), and you know something is wrong.

I think the word you are looking for is reliability, not stability.

ECC does not help everyday stability at all. If your computer is 100% stable (impossible, but a theoretical judgment), adding ECC ram won't make your PC >100% stable.

It just acts like a safety net, which is a factor of reliability, not stability.

 

[cmaier]

AMD is different in that department. Their new 45nm processors perform very differently from the 45nm Denebs. And they run at way different TDPs as well.

Intel has just been adding an extra QP/FSB link and some minor tweaks to their Xeons since PIII.



I think the word you are looking for is reliability, not stability.

ECC does not help everyday stability at all. If your computer is 100% stable (impossible, but a theoretical judgment), adding ECC ram won't make your PC >100% stable.

It just acts like a safety net, which is a factor of reliability, not stability.

I'm not sure about the semantics of this, but I look at it this way: without ECC, I take my laptop up in a 747 and a stray alpha particle hits instruction memory and my machine crashes. Feels unstable to me.

I put ECC memory in there. Now a stray alpha particle hits instruction memory, and the flipped bit is transparently repaired. No crash. More stable. Me like.

 

[VirtualRain]

Intel does not determine which is which from inspection. They make one CPU with two QP links, one with one. And one with a certain set of instructions, the other with a different set.

Either I wasn't clear, or you missed it, but I think we agree... Intel only bins single QPI parts to determine their destiny (Xeon vs. Core i7 branding) - they are identical silicon. The dual QPI parts are obviously destined for Xeon branding from the get go since Intel doesn't offer a Core i7 branded dual-QPI CPU.

I wasn't aware there was instruction set differences between Xeon's and Core i7 desktop parts... is this true? What instructions are different? Is this implemented in firmware/microcode or actual silicon?

 

[Umbongo]

Either I wasn't clear, or you missed it, but I think we agree... Intel only bins single QPI parts to determine their destiny (Xeon vs. Core i7 branding) - they are identical silicon. The dual QPI parts are obviously destined for Xeon branding from the get go since Intel doesn't offer a Core i7 branded dual-QPI CPU.

I wasn't aware there was instruction set differences between Xeon's and Core i7 desktop parts... is this true? What instructions are different? Is this implemented in firmware/microcode or actual silicon?

When the Skulltrail platform game out I recall a small discussion on the Custom PC podcast about how the socket 771 QX9775 had different prefetch routines (I think routines was the word used) to it's Xeon 5482 counterpart. They suggested that Intel claimed they would be better for gaming compared to working with databases and such.

I couldn't find any information on Intel's site on this, though they were just relaying what they had been told.

 

[Salavat23]

When the Skulltrail platform game out I recall a small discussion on the Custom PC podcast about how the socket 771 QX9775 had different prefetch routines (I think routines was the word used) to it's Xeon 5482 counterpart. They suggested that Intel claimed they would be better for gaming compared to working with databases and such.

I couldn't find any information on Intel's site on this, though they were just relaying what they had been told.

Anything that uses FB-DIMMs will be a failure in gaming compared to a similar spec'ed computer running normal RAM. That's why SkullTrail did not do well.

 

[sneezymarble]

I'm not sure about the semantics of this, but I look at it this way: without ECC, I take my laptop up in a 747 and a stray alpha particle hits instruction memory and my machine crashes. Feels unstable to me.

I put ECC memory in there. Now a stray alpha particle hits instruction memory, and the flipped bit is transparently repaired. No crash. More stable. Me like.

Haha. Great explanation! I'm glad you're around here.

 

[VirtualRain]

I'm not sure about the semantics of this, but I look at it this way: without ECC, I take my laptop up in a 747 and a stray alpha particle hits instruction memory and my machine crashes. Feels unstable to me.

I put ECC memory in there. Now a stray alpha particle hits instruction memory, and the flipped bit is transparently repaired. No crash. More stable. Me like.

+1 😀

 

[rylin]

Those are some of the most stupid comments I've read so far, especially the first one.

Eh, how is

SMP capabilities, larger caches, sometimes rather conservatively clocked etc. are some of the reasons they're used as workstation procs, and yes, all MPs are using Xeon CPUs atm.

stupid?

Intel haven't made any SMP end-user CPUs in a long time IIRC.
Some generations, Xeon brand CPUs have larger L2 caches than the end-user versions of the same architecture.
Conservatively clocked means they can be stable even at higher temperatures, and on a finishing note, all current MPs are using Xeon brand CPUs.

How is any of this stupid? 🙄

 

[wfj5444]

I think the word you are looking for is reliability, not stability.

ECC does not help everyday stability at all. If your computer is 100% stable (impossible, but a theoretical judgment), adding ECC ram won't make your PC >100% stable.

It just acts like a safety net, which is a factor of reliability, not stability.

ECC also protects against manufacturing defects in memory, not only cosmic rays (yes that's an honest problem, not a joke 😱 ).

Its about both reliability and stability. Reliable data, and stable OS. If you flip a bit, that can EASILY crash a system.

So its about both.

 

[Salavat23]

Eh, how is
stupid?

Intel haven't made any SMP end-user CPUs in a long time IIRC.
Some generations, Xeon brand CPUs have larger L2 caches than the end-user versions of the same architecture.
Conservatively clocked means they can be stable even at higher temperatures, and on a finishing note, all current MPs are using Xeon brand CPUs.

How is any of this stupid? 🙄

Show me a Core i7 without SMP? Can you show me?

Tell me which 55xx series Xeon has more cache than the i7? Can you tell me?

Prove to me that i7 965EEs running at 90 degrees that have been stable for weeks at 100% load are somehow less stable than the Xeons? Can you prove it to me?


THAT is why all of what you said is stupid.

ECC also protects against manufacturing defects in memory, not only cosmic rays (yes that's an honest problem, not a joke 😱 ).

Its about both reliability and stability. Reliable data, and stable OS. If you flip a bit, that can EASILY crash a system.

So its about both.

99% of RAM is stable at stock clocks. There is no need to use ECC.

And if their was a manufacturing defect, about 70% of those get picked up and discarded.

 

[cmaier]

Show me a Core i7 without SMP? Can you show me?

Tell me which 55xx series Xeon has more cache than the i7? Can you tell me?

Prove to me that i7 965EEs running at 90 degrees that have been stable for weeks at 100% load are somehow less stable than the Xeons? Can you prove it to me?


THAT is why all of what you said is stupid.




99% of RAM is stable at stock clocks. There is no need to use ECC.

And if their was a manufacturing defect, about 70% of those get picked up and discarded.

Dude - seriously - I'm not kidding. Unlike (I presume) you, I actually have designed memory and CPUs (for my Ph.D., for Exponential, for Sun, for AMD).

(http://www.ecse.rpi.edu/frisc/theses/MaierThesis/ , http://ieeexplore.ieee.org/Xplore/login.jsp?url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel3%2F4%2F13972%2F00641683.pdf%3Farnumber%3D641683&authDecision=-203 , etc.)

I assure you, if you really care about not crashing, and you really care if your computations give the right answer, you really truly want ECC memory.

I'm not even talking about whether memory has a defect. Just in ordinary use, charges build up. Cosmic rays hit. Stuff happens. Once a bit flips, without ECC you are hosed.

 

[d3bug]

Dude - seriously - I'm not kidding. Unlike (I presume) you, I actually have designed memory and CPUs (for my Ph.D., for Exponential, for Sun, for AMD).

(http://www.ecse.rpi.edu/frisc/theses/MaierThesis/ , http://ieeexplore.ieee.org/Xplore/l...0641683.pdf?arnumber=641683&authDecision=-203 , etc.)

I assure you, if you really care about not crashing, and you really care if your computations give the right answer, you really truly want ECC memory.

I'm not even talking about whether memory has a defect. Just in ordinary use, charges build up. Cosmic rays hit. Stuff happens. Once a bit flips, without ECC you are hosed.

don't feed the trolls 😉

 

[Salavat23]

Dude - seriously - I'm not kidding. Unlike (I presume) you, I actually have designed memory and CPUs (for my Ph.D., for Exponential, for Sun, for AMD).

(http://www.ecse.rpi.edu/frisc/theses/MaierThesis/ , http://ieeexplore.ieee.org/Xplore/l...0641683.pdf?arnumber=641683&authDecision=-203 , etc.)

I assure you, if you really care about not crashing, and you really care if your computations give the right answer, you really truly want ECC memory.

I'm not even talking about whether memory has a defect. Just in ordinary use, charges build up. Cosmic rays hit. Stuff happens. Once a bit flips, without ECC you are hosed.

That's over exaggerating the issue. You are making it seem like if you don't have ECC, your computer will go down every hour.

I can run my memory ~20% overclocked for long 100% loads at a time with F@H or Prime and I never get any issues.

From my experience, heat is the main killer of memory after proper installation. I'd rather take non-ECC ram that runs at a cool 25 C load than ECC at a presumable 55-65 C load.

I have never had any issues with non-ECC ram when maintained correctly, and IMO, you are making the issue of errors bigger than they should be.

 

[Salavat23]

don't feed the trolls 😉

Well that very hypocritical of you to mention it because you just did that very same action.