Folding Forum

I have put together a new server which I plan to use for folding. The specification of the key components is below:

2 x XEON E5 2630 CPU's total 12 physical cores & 12 hyper threaded cores
2 x Radeon 7970 MSI Lightning Graphics cards (3 GB VRAM per card)
32 GB Registered DDR3 RAM
Windows 2012 R2 SP1 O/S

I'm already folding on an I7 with a GTX580 graphics card which produces a fair number of PPD but I want to be able to wring every possible point out of the new server.

Please could some of you more advanced folders advise how I configure the client to achieve the maximum potential & what should the expected PPD be.

Thanks for any help.

Howdy AAM20007,
Welcome to the f@h support forum.

Setup your rig to run SMP 22 (threads) and give each Radeon a slot in v7.

Your rig may be powerful enough to run BigAdv w/native Linux (or maybe virtual) according to comparisons of successful Bigadv machines in performance databases:
https://docs.google.com/spreadsheet/ccc ... DcHc#gid=0
https://docs.google.com/spreadsheet/ccc ... ERFE#gid=0
http://www.cpubenchmark.net/cpu_list.php

Thanks for your contributions to the project.

Running windows hurts you as I don't believe bigadv is avaliable.

You can run a virtualbox ubuntu setup and do bigadv that way, should be around 130k ppd or so. Native linux would be buttloads more.

Unfortunately those CPUs run @ 2.3Ghz and that's not fast enough to complete some bigadv projects before the preferred deadline even with 24 threads in native Linux. You would have to run regular SMP where your PPD should be somewhere between 30-50K.

I guess the question would be whether native linux with SMP:24 would outperform Windows SMP:22 w/2x GPU and the only real way to answer that would be to dual-boot and try both, no?

Rolo wrote:I guess the question would be whether native linux with SMP:24 would outperform Windows SMP:22 w/2x GPU and the only real way to answer that would be to dual-boot and try both, no?

With AMD GPUs, the Linux 24 probably wins, until GPU QRB changes the answer again a few weeks/months from now.

bollix47 wrote:Unfortunately those CPUs run @ 2.3Ghz and that's not fast enough to complete some bigadv projects before the preferred deadline even with 24 threads in native Linux. You would have to run regular SMP where your PPD should be somewhere between 30-50K.

Not so sure... turbo 2630 is 2.8GHZ and his machine is faster than these listed, which have successfully completed 8101s:

2p 5670@2.93GHz
4p 6128@2GHz

Rolo's got it- only way to know is try it and see.

Yha, it should be fast enough, I be;ieve 2.3ghz is the 8101 cut off for that generation of xeon. (this is a native linux speed btw)

You can always try but I know that you need dual hex xeons that run at over 2.66Ghz. I have a dedicated dual hex xeon 2.66Ghz setup running natively in linux and it's borderline when it comes to P8101 ... i.e. some WUs finish just before the preferred deadline and others don't make it. Running bigadv on that system would slow down the project so it is running regular SMP.

Turbo will never "kick in" if all cores are running at 100%. An internet search will lead to articles that explain there are a number of conditions that have to be met before Turbo boost happens and one of them is using less than the total number of cores.

It's all academic if, as patonb said earlier, there are no bigadv projects for Windows, but if there are bigadv projects that run on Windows the OP should be okay since P8101 is linux-only.

bollix47 wrote:You can always try but I know that you need dual hex xeons that run at over 2.66Ghz. I have a dedicated dual hex xeon 2.66Ghz setup running natively in linux and it's borderline when it comes to P8101 ... i.e. some WUs finish just before the preferred deadline and others don't make it. Running bigadv on that system would slow down the project so it is running regular SMP.

Turbo will never "kick in" if all cores are running at 100%. An internet search will lead to articles that explain there are a number of conditions that have to be met before Turbo boost happens and one of them is using less than the total number of cores.

It's all academic if, as patonb said earlier, there are no bigadv projects for Windows, but if there are bigadv projects that run on Windows the OP should be okay since P8101 is linux-only.

Maybe so, but you're talking different generations of Xeons. You're running, what- dual 5650s? His twin 2630s are faster- they're faster than 5670s which are reported to complete 8101s.
All he can do at this point is boot into Linux and try it. If successful w/8101s and the like, that would max out his points.
Good points about turbo, by the way.

they're faster than 5670s

I'm not understanding how 24 threads running at 2.3Ghz can be faster than 24 threads running at 2.93Ghz even with the newer technology.

Do you have a source for that statement?

Perhaps my source is incorrect or I'm misreading it.

We still haven't heard from the OP as to whether he would even consider switching to linux so we may be getting off topic. Let's hear from AAM20007 so we know whether or not our discussion is germane to the topic.

If Windows is the only option then running the client with SMP:24 and the bigadv client-type is fine but there's no guarantee the projects downloaded will in fact be bigadv.

@Bollix47- same basic source, only read the multi- cpu setups: http://www.cpubenchmark.net/cpu_lookup. ... 1215&multi

Thanks alan, that works for me. As already stated the only course of action is to try it and see unless someone with the same system comes in and reports some figures.

The Sandy Bridge has a larger cache and bigger QPI bandwidth than the Westmere; this may make F@H perform better or maybe not.
It would be fun to test the two configs.

Rolo wrote:The Sandy Bridge has a larger cache and bigger QPI bandwidth than the Westmere; this may make F@H perform better or maybe not.
It would be fun to test the two configs.

Gimme a couple of weeks an i will try to post some numbers, mine will be octo's vs hex's though at a restricted 2.4-2.5Ghz as oppoosed to the 3.2ghz i get at turbo out of my 5670's