Cores Vs speeeeed PPD?

[ubersonic]

I didn't want to jack another thread but this post about hex core Xeons stood out to me as its something I have been wondering about:

Not worth buying unless you're going to get 2 of them, and even then, it's just 12 slow Sandy cores, but can be done on the cheap I guess. Can't help but think you'd be better off getting a 3930k and taking it higher, 6 cores at 4.4GHZ are going to beat 12 at 2GHZ surely?

What is better for PPD speed or cores?

Lets compare these two:

Xeon E5 2650L - 8C16T, 1.8GHz (8 core turbo is 2GHz), 20MB.

i7 3820 - 4C8T, 3.6Ghz (4 core turbo is 3.7Ghz), 10MB.

Both are socket 2011, both have a 2.5MB cache per core ratio, both use Sandy Bridge architecture. The only differences are that the Xeon has twice the cores and half the speed (more than half under turbo clocks).

To make it simpler lets assume the i7 is clocked at 4GHz as any chip can do that and it means the only difference is the Xeon has half the speed and twice the cores. Which would get the better PPD? or would it be virtually even?

Thanks

 

[kerrgreg]

Personally I think it would be pretty even, however Linky shows that the cpu computation is about 10% more on the Xeon.

On the plus with the Xeon, if you can afford a board like the asus dual socket board, you can run one xeon until you can afford the second and increase your folding by a factor of 2x (if it scales)

My next purchase might be a rig like that.

 

[Biffa]

It depends on how many cores.

4 x 12 Core Opterons allows you to get into the bigadv crunching, I think at a push you can do it with 4 x 18 core Opterons as well but you won't score as highly.

Even then you have to overclock slower/cheaper chips to get a decent return. But they aren't going to run at Xeon speeds.

It used to be that you could get a decent return on Socket 1366 Xeons in an SR-2 and you still can get pretty ok performance (plus you can overclock the suckers) but you can't get bigadv on them anymore.

So the answer to your question is more cores are better than faster, but there are points at which how much it costs to go down the mega core route compared to the multiple cheaper machine route, power costs also come into play, when considering multiple machines and/or graphics cards to compete with a single high core count Opteron system.

 

[paradigm]

Bit of an old thread bump, but should I be running -bigadv on a dual Xeon E5-2680 platform?

16 Cores (32 threads). I am running -bigadv at the moment, but I'm not sure if that's a good idea or not?

 

[kerrgreg]

Hmm not sure I can answer your question but does big adv detect and turn on automatically?

 

[paradigm]

Well, it's a tick box in the setup of FAH GPU Tracker, if there is anything more complex at stanford's end then I've no idea.

EDIT: Supposedly PPD in FAH GPU Tracker includes any points that will be gained through -bigadv bonus, which would make my PPD at least sensible.

 

[Biffa]

bigadv only works in linux

 

[kerrgreg]

I didnt think that was the case, i know they moved to A3 but i thought it was enabled for windows also?, granted the threads I can find about the subject are dated 2010

 

[paradigm]

The machine in that screenshot has -bigadv ticked in the config of "FAH GPU Tracker", and is running fahcore_a3.exe as it's worker process.

Does that mean it is or isn't running bigadv?

 

[MGP]

Just because a flag is checked in a client config, doesn't mean that it will be utilised.

 

[paradigm]

Just because a flag is checked in a client config, doesn't mean that it will be utilised.

Whilst true, the only way for that machine to make 118,000+ PPD is for bonuses to be applied.

I was under the impression that the only bonuses that large were the bigadv ones?

 

[Biffa]

What WU is it running thats the only way to tell.

bigadv WU's are:

6901
6903
6904
8101
8102
8103

They only run on Linux and use the A5 core.

Here is a good relatively recent thread about it: http://foldingforum.org/viewtopic.php?f=85&t=23245

You could always run virtualbox, thats what I used to do on weebeastie.

It sounds about right, I used to get 30K PPD on that type of SMP WU with 10 threads, so 100K or so with 32 threads sort of makes sense.

 

[paradigm]

Unfortunately, I can't run virtualbox on them, as these are only being used to fold for an initial stress test before I hand them over to put into production in our hyper-v based VDI environment.

Hence the reason in the "new hardware" thread to get a couple of quad-socket opteron machines up and running.

Oh well, 36~ hours to go before I have to stop running FAH on them

 

[Biffa]

Oh.. for "stress testing" I used to use a bootable USB with Linux on it

 

[paradigm]

Probably would have been a good idea, but I hand them over first thing on Monday morning, and I'm not on site until then, so there's nothing I can do

Oh well, they've still netted me almost 1,000,000 points so far since I started on the 28th.

 

[MGP]

This would explain why you've suddenly appeared on my radar.

 

[paradigm]

Yeah, I wouldn't worry about it though, my production will drop down to about 50,000 PPD once my new cruncher has arrived, and between tomorrow and then, I'll be down to about 20,000

 

[Biffa]

Ah the ebbs and flows of distributed computing fortunes... I know how that feels.

 

[paradigm]

Well, that's it. All the boxes are now removed from folding duties (well, one is still running it's last WU, 71% at the moment), my production is about to die a death

I'm stuck with a dual-socket Xeon E5462 (2.8GHz, 8C8T) machine, a dual-socket Opteron 2356 (2.3GHz, 8C8T) machine (well, caseless desk build ), and soon to be a dual-socket Opteron 275 (4C4T) within the next hour or so. Hopefully my quad-socket opteron machine will be up and running by the weekend (if the bits arrive in time).

On the plus side, I've managed 1.4 Million points in the short time I had the E5's up and running, and placed myself at rank 122, not bad considering I only started a week ago.

 

[ozaudio]

the good thing is, your now hooked LOL