• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

New Workstation: CPU Choice

S

s0rce

Limp Gawd
Joined
Jan 17, 2011
Messages
495
Going to building a couple new workstations for 3D visualization of scientific data and data processing with CERN Root.

I'm trying to decide between a single-CPU Intel Xeon E5-2687W and a dual-CPU AMD Opteron 6276?

The dual opteron is actually the cheaper option

Ignoring power usage, noise and heat. Only price/performance matters. I'm not sure about the parallizability of much of the code. The vendor recommended dual 6-core xeons in the past so I'm assuming it is well suited to multicores/hyperthreaded?

Any opinions/recommendations?
 
Instead of the single E5 you might as well go with a consumer i7 3930K/3960X. They are only 6 cores instead of 8 but they are a lot cheaper.

In general usage the Intel box is going to be a lot faster, hut with highly threaded code the Opterons will have an advantage. IMHO unless you are specifically going to be running highly threaded stuff I'd go with a 3930K.
 
The Xeon would have a nice upgrade path. If the code really is well threaded you could consider a quad socket AMD.
 
What others said. Price/performance, the AMD setups will kill Intel if you use highly threaded code. If you don't, then a lot of the AMD's power will not be used, so you might as well go with Intel.
 
AMD is going to hands down absolutely rape the XEON if your software is efficiently written. Xeons are nice still.
 
The Intel® Xeon® E5's are designed for a dual processor set up it is one of the reasons that they cost what they do. If you are only going to be using a sigle processor you should be look at the Intel Xeon E3's or if you think you need more memory bandwidth you can go with the X79 chipset with a our 2nd generation Intel Core i7 processors in socket 2011.
 
IntelEnthusiast said:
The Intel® Xeon® E5's are designed for a dual processor set up it is one of the reasons that they cost what they do. If you are only going to be using a sigle processor you should be look at the Intel Xeon E3's or if you think you need more memory bandwidth you can go with the X79 chipset with a our 2nd generation Intel Core i7 processors in socket 2011.
Click to expand...

I was looking at the E5 instead of the E3 as it is 8core vs 4core. The current machines we have are dual-6 core Xeon's I dont know which model.
 
The correct answer is going to depend on how well Cern-Root will perform on each machine, and we have no way of knowing that. I did a quick google search for benchmarks and failed to come up with anything meaningful. Apparently, there's a benchmark suite for the software, but I didn't find any kind of database or log of results for current day systems. It might be better to ask your colleagues who are processing datasets similar to yours what kind of systems they run and what kind of performance they get out of them, and whether they've been satisfied with their results.

It's really important to have this information before making a decision. Just to give you an example, let me tell you a little bit about the processing requirements of Prime95.

The software runs Lucas-Lehmer primality tests on mersenne numbers. The test is basically a series of millions of large floating-point multiplications that must be done sequentially. The test can be parallelized across multiple cores or processors, but with diminishing returns. With four cores, for example, more efficient to test four different numbers, one on each core, than it is to test one number using all four cores simultaneously. So, from a value standpoint for this application, one is better off purchasing two 1-cpu systems than a single 2-cpu system.

In AMD's new Bulldozer architecture, each chip has n processing cores, but only n/2 floating point processing units. So for every two cores, there is only only one FPU to share between them. This is ok for some applications, but not at all suitable for Prime95 which relies almost exclusively on floating point calculations. Even outside of Bulldozer, the current generation of Intel processors is simply much faster at floating point than anything AMD has to offer. So if your goal were to buy hardware to run Prime95, you would be best off purchasing as many barebones quad-core, high clock-rate Intel systems as you could afford.

None of this is pertinent to your application, but hopefully it will help you understand what kind of data you need to have in order to make the best purchasing decision. Over in the Distributed Computing section, there are quite a few folks who run 4p AMD Opteron systems for Folding@Home. Something you might be able to do is run the benchmarks on your current 2p Xeon system, and ask a folder with a 4p system to install your software and run the same benchmarks, so you can start to get some data to work with.
 
You must log in or register to reply here.
Back
Top