Server CPU versus Desktop CPU whats the diff

[Perryg114]

Ok what is the difference in a desktop CPU versus a Server CPU? Why use a slower desktop CPU when the server chips are faster? What defines a server CPU from a Desktop one? When it gets right down to it, its all just 1's and 0"s. I am wanting to build some machines for general use at work and they could be doing any number of things from high speed data acquisition, 3d CAD, or for video and image processing.

Perry

 

[mooseracing]

Server equip is higher quality is the easiest way to put it. It's all about stability and uptime. You use a slower desktop equiv because of the cost from the quality of server components. I personally like server and SMP machines, they last longer and are worked harder.

I'm sure others will fill in the details about ecc, ram bus, different sockets, different slots, different hardware, and other details.

 

[myocardia]

The server chips of the same speed aren't faster. They're a bit more likely to overclock a bit higher, but that's about the only difference, besides the fact that some (most) people believe that they are of slightly better quality.

 

[Syzygies]

The difference is in the motherboard, but the different socket forces a different cpu to match. There are actually three kinds of motherboards:

* Generic desktop motherboards, what grandma uses to check email, what Dell uses. Can't be overclocked. Neither can grandma.

* "Gamer" desktop motherboards. As porn financed and drove the development of home vcr's, gamers drive the market for performance motherboards, and all of us who overclock for whatever reason benefit.

* Server motherboards. Their prime function is to keep the sys admin's job as easy as possible. They don't overclock. Neither do sys admins.

With the exception of the IBM "skulltrail" motherboard, it's impossible to find a performance motherboard with more than one cpu socket. Server motherboards on the other hand have up to four cpu sockets and many memory slots.

Server memory is both more expensive and slower than enthusiast memory, which is often "registered" and/or error correcting. Only certain errors are caught, the protection is an illusion at a steep cost.

A 24-wheeler truck can't race a decent sports car. A loaded server board is like the truck, a gamer desktop rig is like the sports car. There are more reasons for this than anyone can list in a thread.

I've built two four-core desktops using the Q6600, a chip we'll remember fondly a decade from now. A buddy spent $10K on an eight core AMD server with 64 GB of memory. We're still at the stage of devising stress tests that can take down all known machines (24 hours of Prime95 is just a start), his shop took my favorite test in-house to save delivery cycles. We're just getting the hang of parallel code, our goal, but it's clear that except for problems that can actually use his 64 GB of memory, my overclocked four Intel gamer cores can whoop his eight AMD server cores any day. Even adjusting for clock speeds, his AMD chips are slower at everything. We race 'em all the time, and he's actually a bit bummed about this. There's a reason why AMD is cutting jobs.

 

[palladium]

Originally posted by: Syzygies
The difference is in the motherboard, but the different socket forces a different cpu to We're just getting the hang of parallel code, our goal, but it's clear that except for problems that can actually use his 64 GB of memory, my overclocked four Intel gamer cores can whoop his eight AMD server cores any day.

Also stuff that can use all 8 cores, like 3D rendering/F@H, and server-related tasks. Not that an average power user do those sort of stuff frequently anyway.

 

[Perryg114]

Ok I think what is confusing me is that if you go into Sandra and look at their reference processors the Xeon are twice as fast as the desktop chips of the same speed and number of cores. Am I missing something here? Is there a chart or table that tells me what chip is fastest and maybe a comparison of what the desktop equivalent is of a given Xeon processor?

Sandra is showing 116052 MIPS for the Xeon X5482 and 42784MIPS for the Intel 2 Quad QX9770. These appear to me to be equivalent chips. Can anyone explain the difference or is Sandra full of crap?

Perry

 

[CTho9305]

I've heard from unreliable sources that some features are configured slightly differently (e.g. subtle aspects of prefetcher behavior) between the Xeons and the consumer chips... it makes sense, since you want to optimize for different benchmarks. It wouldn't surprise me if there are a few other things Intel tweaks slightly. I don't know what if anything differs on the AMD side.

Sandra is virtually worthless for predicting real-world performance. Changes that have negligible real-world effect can drastically skew its results; changes with significant real-world effect can have minimal effect on the Sandra score. You also have to question the test methodology - some sites run benchmarks just once, and come up with incorrect results because of that (it was either Anandtech or THG that had some comparison of Athlon 64 rev F3 vs G1 and G2 - Windsor vs Brisbanes - where they posted results that were blatantly garbage... maybe Windows synchronized the clock during one of their benchmark runs or something).