2p intel or 4p amd?
- Thread starter ridemx133
- Start date
mihapiha
New member
Then it depends on the budget mainly.
The thing is that a 10% difference in performance doesn't mean 10% more PPD. Due to the bonus point system the PPD differences are far greater.
I use a quad socket AMD platform with 4x AMD Opteron 6180 SE CPUs overclocked to 2.75 GHz. If I now compare that to a dual socket Intel E5-2687W system, it means that my system is roughly 30% faster in computing performance, yet produces 60% more PPD. If a system is now even faster than mine the just PPD skyrocket. Without the (10%) overclock I loose about 100k PPD.
At this level, minor differences have just huge effects on the PPD.
But unless you elaborate what you precisely wish to know, I will never be able to answer your question. You intend on building one of those setups, or are you just curious? If you intend on building one of these setups, what would be your budget?
The thing is that a 10% difference in performance doesn't mean 10% more PPD. Due to the bonus point system the PPD differences are far greater.
I use a quad socket AMD platform with 4x AMD Opteron 6180 SE CPUs overclocked to 2.75 GHz. If I now compare that to a dual socket Intel E5-2687W system, it means that my system is roughly 30% faster in computing performance, yet produces 60% more PPD. If a system is now even faster than mine the just PPD skyrocket. Without the (10%) overclock I loose about 100k PPD.
At this level, minor differences have just huge effects on the PPD.
But unless you elaborate what you precisely wish to know, I will never be able to answer your question. You intend on building one of those setups, or are you just curious? If you intend on building one of these setups, what would be your budget?
Last edited:
ok my goal is to get my gaming rig back and to build a dedicated folding rig.
my budget is $4000. I would like the most ppd out of a single system.
NeverBackDown
AMD Enthusiast
Folding is not my area of expertise but i am assuming a dual 2011 mobo will be faster than a 4p amd system depending on cpus. If the Amd is using the 16 core Opteron cpus then i would be pretty sure to say it would beat any dual 8 core cpus from intel.
mihapiha
New member
This assumption of your's is completely wrong, I'm afraid to say. I have a hand full of friends using both dual Intel as well as quad AMD to fold and quad AMD is the king of the hill.
The PPD to being expected out of a 4P system is about 400 - 600k PPD depending on the CPUs; 2P Intel is behind with about 200 - 350k PPD depending on the CPUs.
But honestly, I'm not sure if this is still the way to go. An AMD 7970 can get these days 100k PPD. It's just a matter of time before a couple of 8970 beat the quad socket system.
4P is still ideal for folding, that's for sure. But it may not be in a few months time. And I'm not sure whether it's worth investing $4000 into that system.
The Opteron 6100 series CPUs can be overclocked on SuperMicro motherboards, and you can get the 6180 SE rather cheap on ebay. So for the $4000 you should be able to build a system very much similar to mine (check out signature).
Alternatively you could get a motherboard with 4 PCIe x16 slots and a 8 core AMD CPU and fold using GPUs. Especially if you don't care about wattage consumption.
Unfortunately GPU and SMP folding doesn't work together well. GPU folding works better with Windows, SMP with Ubuntu Linux...
But I'd not get the 4P system anymore to be honest. There's not much you can do with such a system besides fold. I've been trying to sell mine and I had no luck what so ever... The system is far too big of a niche product and far too expensive, so the resell value is really quite bad. The quad GPU system is updateable too...
Last edited:
NeverBackDown
AMD Enthusiast
If you honestly feel that in a few months time things will change enough to where your $4k becomes more of a paper weight than it already is than wait those few months.
mihapiha
New member
My point basically was that a quad socket system has no real other purpose. You can't really use that type of a system for hardly anything else. 32 cores/threads and 2 CPUs are basically the specs a "normal" OS and today's software are able to handle. The extra performance on top of that often just gets wasted.
Although the performance of the dual socket Intel system is behind, I'd probably lean in that direction. But if I'd have to make a suggestion, I'd recommend getting a GPU based folding farm as the GPUs are not that far behind.
A system with lets say 4x 7970s can be used for other things besides F@H, if for some reason F@H isn't the main focus anymore. It's much harder to get rid of server hardware. It's just something I'd keep in mind if I'd be purchasing a folding farm these days.
If you spend $4000 you want most out of a system.
Although the performance of the dual socket Intel system is behind, I'd probably lean in that direction. But if I'd have to make a suggestion, I'd recommend getting a GPU based folding farm as the GPUs are not that far behind.
A system with lets say 4x 7970s can be used for other things besides F@H, if for some reason F@H isn't the main focus anymore. It's much harder to get rid of server hardware. It's just something I'd keep in mind if I'd be purchasing a folding farm these days.
If you spend $4000 you want most out of a system.
Last edited:
My point basically was that a quad socket system has no real other purpose. You can't really use that type of a system for hardly anything else. 32 cores/threads and 2 CPUs are basically the specs a "normal" OS and today's software are able to handle. The extra performance on top of that often just gets wasted.
Although the performance of the dual socket Intel system is behind, I'd probably lean in that direction. But if I'd have to make a suggestion, I'd recommend getting a GPU based folding farm as the GPUs are not that far behind.
A system with lets say 4x 7970s can be used for other things besides F@H, if for some reason F@H isn't the main focus anymore. It's much harder to get rid of server hardware. It's just something I'd keep in mind if I'd be purchasing a folding farm these days.
If you spend $4000 you want most out of a system.
let me ask you this, is the gpu as versatile as the cpu for f@h?
mihapiha
New member
you have to excuse me. English is my third language. I'm not sure entirely what you wish to know.
In terms of computing power, the GPU is always ahead of a CPU. So naturally the GPU is better. However, GPUs are not as stable as the CPU and can have problems with errors in the calculations. So the bigger the project is, the better the CPU could compete with it. Especially if stability is the main concern. ECC memory can boost stability quite a bit on server platforms. A CPU is build to run 24/7 under full load; the GPU can have problems with that after a few months...
Therefore the initial idea by Stanford was to give small WUs to CPU and GPU, mid-sized WUs to GPU and big-WUs to multi-socket CPU systems. At least that was the idea one or two years ago when the new "big-WUs" exclusively for 16+ core system for F@H were released. Since the CPU (even a multi-socket system) is much slower than the GPU, CPU WU were given bonus points. The quicker you finish a SMP WU the more bonus points you'd get. The idea was to make SMP-folding interesting. Otherwise the GPU would just dominate, and some of the WUs wouldn't be folded regularly.
Now fast forward to today. SMP-WUs still get bonus points and hence score phenomenally in terms of PPD although the computing performance is not there. The really major thing in the last few months was Stanford's attempt to really get OpenGL properly supported for folding. AMD/ATI cards now FINALLY got the proper support and finally can fold with proper results. Just a few months ago a 7970 which in terms of computing performance beats pretty much any nVidia card, was getting only a few thousand PPD (which is like a GTX 285 or something nVidia had a hand full of generations ago).
Due to the proper OpenGL support, nVidia is no longer the major favorite for F@H. A 7970 will now get up to 100k PPD while a GTX 690 will get "only" 40k PPD; The Titan is @ 50k PPD;
Just due to the OpenGL support things have changed on a major scale over the last few weeks. Will those PPD maintain? I don't know. AMD WUs are still in the test-phase so there might be some adjustments in terms of PPD;
The CPU PPD and WUs won't change much I hear. So I should be able to give you pretty precise numbers for pretty much any 2P or 4P system you're thinking of getting. I'm pretty confident that I won't miss by much.
AMD-GPU PPD in 2 months could be different. I really don't know if or when they will change anything there. nVidia GPUs are just not worth investing in anymore. At least from todays standpoint. Whether they will do some adjustments in terms of nVidia WU PPD, I don't know... They did release some beta-WUs for nVidia which had bonus points too. Either way it doesn't look that good for multi-socket folding@home systems. They now still have the best PPD/Wattage consumtion ratio. How long this will stay like that nobody really knows...
A multi GPU based system is a decent alternative. There is not much use outside of F@H you can find for a 4P system. (if you can think of something, do tell me. I'd be happy to give it a try)
If you get the GPU-based folding farm the PPD/wattage consumption will be worse (that's for sure) but you can easily find another occupation for a multi-GPU system compared to a multi-CPU system if you wish to change things up. And keep in mind that updating GPUs is much easier than updating CPUs...
I'm not sure if that answered your question. I hope it did.
In terms of computing power, the GPU is always ahead of a CPU. So naturally the GPU is better. However, GPUs are not as stable as the CPU and can have problems with errors in the calculations. So the bigger the project is, the better the CPU could compete with it. Especially if stability is the main concern. ECC memory can boost stability quite a bit on server platforms. A CPU is build to run 24/7 under full load; the GPU can have problems with that after a few months...
Therefore the initial idea by Stanford was to give small WUs to CPU and GPU, mid-sized WUs to GPU and big-WUs to multi-socket CPU systems. At least that was the idea one or two years ago when the new "big-WUs" exclusively for 16+ core system for F@H were released. Since the CPU (even a multi-socket system) is much slower than the GPU, CPU WU were given bonus points. The quicker you finish a SMP WU the more bonus points you'd get. The idea was to make SMP-folding interesting. Otherwise the GPU would just dominate, and some of the WUs wouldn't be folded regularly.
Now fast forward to today. SMP-WUs still get bonus points and hence score phenomenally in terms of PPD although the computing performance is not there. The really major thing in the last few months was Stanford's attempt to really get OpenGL properly supported for folding. AMD/ATI cards now FINALLY got the proper support and finally can fold with proper results. Just a few months ago a 7970 which in terms of computing performance beats pretty much any nVidia card, was getting only a few thousand PPD (which is like a GTX 285 or something nVidia had a hand full of generations ago).
Due to the proper OpenGL support, nVidia is no longer the major favorite for F@H. A 7970 will now get up to 100k PPD while a GTX 690 will get "only" 40k PPD; The Titan is @ 50k PPD;
Just due to the OpenGL support things have changed on a major scale over the last few weeks. Will those PPD maintain? I don't know. AMD WUs are still in the test-phase so there might be some adjustments in terms of PPD;
The CPU PPD and WUs won't change much I hear. So I should be able to give you pretty precise numbers for pretty much any 2P or 4P system you're thinking of getting. I'm pretty confident that I won't miss by much.
AMD-GPU PPD in 2 months could be different. I really don't know if or when they will change anything there. nVidia GPUs are just not worth investing in anymore. At least from todays standpoint. Whether they will do some adjustments in terms of nVidia WU PPD, I don't know... They did release some beta-WUs for nVidia which had bonus points too. Either way it doesn't look that good for multi-socket folding@home systems. They now still have the best PPD/Wattage consumtion ratio. How long this will stay like that nobody really knows...
A multi GPU based system is a decent alternative. There is not much use outside of F@H you can find for a 4P system. (if you can think of something, do tell me. I'd be happy to give it a try)
If you get the GPU-based folding farm the PPD/wattage consumption will be worse (that's for sure) but you can easily find another occupation for a multi-GPU system compared to a multi-CPU system if you wish to change things up. And keep in mind that updating GPUs is much easier than updating CPUs...
I'm not sure if that answered your question. I hope it did.
Last edited:
NeverBackDown
AMD Enthusiast
Dual Xenon and 2 7970s should compete with a 4p Amd PPD if what you say is true then?
mihapiha
New member
Yes, if it's two different systems. Remember I said that SMP WUs (or CPU WUs) have to be folded on Linux. The big-WU (meaning the 16+ core/thread WUs) are not available on Windows. At least not to my knowledge.
The difference between Linux and Windows is unreal. Let me give an example with my rig. Big-WUs on my server means 575k PPD on average (with the WUs out now); small SMP WUs would mean I'd fall to about 220k PPD-ish. And Windows folding and I'm even under 200k PPD!
So (especially) with 4P AMD, Linux is a must. I hear the difference is smaller with Intel; but I don't think that the precise numbers are now so important that I need to phone up my buddy, unless ridemx133 really decides to go down this road. If he chooses that pathway I'll call my friend and ask him. He's using the dual E5-2687W CPUs for folding... He tested on both OSs I'm sure...
I have not yet set up an AMD card since they changed it. So I'm not sure if it will work just as well in Linux. I'd have to google, but to my knowledge they barely got nVidia working in Linux. So if I'm not mistaken AMD GPU folding would mean Windows.
So you could get the to the 500k PPD neighborhood with two 7970s and a 2P Intel system. But it would mean two computers due to the OS restrictions...
Personally I'd not mix SMP and GPU folding on the same computer. I think you only run into issues. Especially since every GPU really needs one Core (not necessarily only a thread) to properly fold. It depends on the WU obviously.
If you choose running both at the same time on the same rig, I'd recommend a bit of testing what is ideal. Hard to tell with the recent changes. I've not folded with the GPU and CPU at the same time on the same computer in more than a year. Back then it was best to turn off all GPUs and only fold with the CPU... then they changed some GPU WUs and it was best to run all GPUs as they required one thread each, before that they demanded one full core or they'd be at 60% load ...
These things change as the WUs change. And since you can't pick the WUs you're folding, you have to adapt..
Last edited:
not so
I have a 3930k and 2 gtx660 combined they get about 72,000ppd
I get about 35k-50k out of the 3930k with the 2 gpus.
without the gpus I can get 45k-70k with the cpu only
mihapiha
New member
Like I thought. It just doesn't work well mixing SMP and GPU folding... Best to have one of the two properly set up
And these numbers are without taking wattage consumption into the account! :huh:
I have the weird feeling that SMP folding on that computer might be the ideal solution long-term...
Edit:
If you want I can help you set up the 3930k in Linux. If you get it to 4.2 GHz (or more) you can fold big WUs with it. It will fold one WU 3 days straight but in the end the overall PPD average should be in the 90 - 110K neighborhood....
This would mean though, that you cannot use the rig no more. The deadline is so short that you need to let it fold two to three days straight (depending on the WU) for it to finish within the deadline. Otherwise you'll get 20k points instead of 300k...
Last edited:
Like I thought. It just doesn't work well mixing SMP and GPU folding... Best to have one of the two properly set up
And these numbers are without taking wattage consumption into the account! :huh:
I have the weird feeling that SMP folding on that computer might be the ideal solution long-term...
Edit:
If you want I can help you set up the 3930k in Linux. If you get it to 4.2 GHz (or more) you can fold big WUs with it. It will fold one WU 3 days straight but in the end the overall PPD average should be in the 90 - 110K neighborhood....
This would mean though, that you cannot use the rig no more. The deadline is so short that you need to let it fold two to three days straight (depending on the WU) for it to finish within the deadline. Otherwise you'll get 20k points instead of 300k...
its running at 4.7 at the moment
I would run Ubuntu but I play games on this one. that's why I want to build a rig for just folding
mihapiha
New member
that's ok. I just wanted to let you know.
You've not replied to any of my previous posts though. So I'm still unsure if I answered that question of your's:
I somehow cannot find a translation which would make sense to me. The word "versatile" just confuses me and I cannot find a translation which would explain what you meant in context.
You've not replied to any of my previous posts though. So I'm still unsure if I answered that question of your's:
I somehow cannot find a translation which would make sense to me. The word "versatile" just confuses me and I cannot find a translation which would explain what you meant in context.
mihapiha
New member
There isn't really hardware which would be better to the entire F@H project. They have to put together the projects they send out. It's not like you finish an entire simulation on your computer. A couple hundred people share the load. That's why works so well. And a full project usually consists out of many GPU and CPU WUs. And the simulation isn't over till all the results of all the WUs come in.
So if for argument sake we all were using GPUs, many projects wouldn't get finished. So they try to balance out the usage of GPUs and CPU through those points, by awarding specific hardware specific amounts of points. Nobody would run a 4P AMD server for F@H if you'd get only 20k PPD with it
. This is not a ball park figure, but rather the points I'd get if they wouldn't give bonus points. In performance I'd be competing with a GTX 580 or a GTX 660 Ti. Which is quite realistic I think. But they need them big WUs too and GPUs could have problems folding those. Maybe it's their fear that the WUs wouldn't get finished or something, I don't know...If the numbers I heard are correct for the 7970 then it's rather easy to calculate that it must be around 100k PPD on about 350W usage. Maybe even less if you get like really low voltage CPUs, and under voltage them.
Basically now a single 7970 can nearly keep up with a 2P AMD, which is like totally absurd. The 2P AMD also consumes around 350-ish and crunches about 100 - 150k PPD depending on the CPUs...
With one move a lot of people's servers became outdated basically. I guess that's why they brought new CPU WUs. P8101 was a standard and I used to get that for over 2 months straight. 430k PPD was that WU, and now there is like a sh*tload of P8102 - P8105 WUs which crack my PPD up to 690-700k PPD; So I remained at about 575k PPD average over two months now...
But you can really look forward to 8970 results. I can see two or 3 of those cracking my server's PPD. Wattage wise they will getting closer and closer as well as the time progresses.
My half million PPD is impressive now, but I can see in a half year everybody will be pushing 100k PPD easily...
Last edited:
