Page 2 of 2
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Sun Apr 13, 2008 8:32 pm
by 41.60l23
Also:
Some people on the folding forum are having success running two or more GPU clients at a time. It's not multithreaded, so you can theoretically double your PPD if your GPU has the headroom. This especially comes in handy for those people who are using 30% or less of their GPU on a Pentium D.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Mon Apr 14, 2008 1:09 pm
by Haggis!
Right. I haven't done anything to try to improve PPD. Just loaded it up and let it do it's thing to see what it'd do.
The 3870 at home is running on an E3110 @ 3.6GHz.
At the office, I have another 3870 on a
[email protected]. So I tossed it on there this morning. FahMon is reporting that that GPU is doing around 560 PPD. Again, totally not optimized.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Mon Apr 14, 2008 2:42 pm
by ChronoReverse
Testing stock A64 X2 3800+ along with a stock HD3400 and getting 266PPD. Fully eats up one CPU core and the GPU utilization reads 99% as well.
I wonder what would be the quickest way to run it on my Q6600 and 3850. Perhaps 1 core for the GPU client and three cores for SMP client?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Mon Apr 14, 2008 3:32 pm
by Gerbil Jedidiah
ChronoReverse wrote:Testing stock A64 X2 3800+ along with a stock HD3400 and getting 266PPD. Fully eats up one CPU core and the GPU utilization reads 99% as well.
I wonder what would be the quickest way to run it on my Q6600 and 3850. Perhaps 1 core for the GPU client and three cores for SMP client?
If you're wanting to max out your PPD on the Q6600 I think running two SMP clients in virtual linux environments is the answer. People are reporting 4,000 PPD with this. Each SMP WU is worth 1760, sure you probably know that, but wanted to throw it out there before mentioning that my Q6600 does at least two WUs per day, and then every now and then it does three
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Mon Apr 14, 2008 5:43 pm
by 7im
I wouldn't get too hung up on the PPD just yet, or make any final conclusions. This is a very early look at the initial release of a BETA client. The WUs out there right now are very small test WUs, only 97 points each. You can expect changes to follow in successive releases of the GPU2 beta, and changes in bigger WUs.
And Vijay just posted some News along the lines of what I was saying... Don't make any final judgements just yet, improvements are coming.
http://folding.typepad.com/news/2008/04/how-the-gpu2-co.htmlVijay Pande wrote:There have been some misunderstandings on how the GPU2 core works. In particular, for small proteins like villin on GPU's with large number of stream processors (SP's) like the 3850 or 3870, the protein is too small to use a larger number of SP's unless the CPU is very fast. Some people have guessed that there is some internal SP limit. This is incorrect; the problem is that small proteins can't be parallelized amongst a large number of SP's.
We are working to release larger proteins (about 2x the number of atoms) as they are more interesting scientifically and use the GPU's (even the high end ones) much closer to 100%. The exciting part for us is that the larger proteins run at almost the same speed as the slower ones on GPU's (whereas on CPU's, they're 4x slower); this is where the GPU2 code should shine. In parallel, Mike Houston at AMD is working to optimize CAL such that it has lower CPU overhead.
For now, we're pushing out villin WU's as a test (good to know that the code is working well), but we expect the larger WU's to be going out soon (say a week or two, pending internal testing).
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Mon Apr 14, 2008 6:18 pm
by Gerbil Jedidiah
That's good news indeed. Thanks for clarifying
I don't own an AMD GPU, alas, but may consider this in the near future as a folding upgrade.
What videocards seem to be the best bang for the buck with respect to PPD. Heat, and power consumption?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Mon Apr 14, 2008 7:08 pm
by ChronoReverse
Just started it up on my 3850 and I'm getting ~1600PPD. 92% utilization with a 3GHz Q6600. In the interest of beta testing and science I'll let this run =)
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Tue Apr 15, 2008 11:25 am
by jeffry55
ChronoReverse wrote:Just started it up on my 3850 and I'm getting ~1600PPD. 92% utilization with a 3GHz Q6600. In the interest of beta testing and science I'll let this run =)
And you are "testing" with UGN as the username ............................ right!?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Tue Apr 15, 2008 11:40 am
by ChronoReverse
Haha, I'm using my own name but I've put myself under 2630 =)
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Tue Apr 15, 2008 2:29 pm
by jeffry55
ChronoReverse wrote:Haha, I'm using my own name but I've put myself under 2630 =)
Well, I'll settle for that for now.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Tue Apr 15, 2008 11:59 pm
by Usacomp2k3
According to a recent post, GPU1 is no longer available. I didn't think v2 supported all the cards?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 10:27 am
by Haggis!
You are correct, GPU2 has to be run on 2000 or 3000 series Radeons. I can verify first hand that you can't "round up" the model numbers and run GPU2 on the x1950.
I hope it doesn't seem like I'm freaked out about points. At this point, I understand it's early in GPU2's life. But I am trying to wrap my arms around how GPU2 is doing on my systems, and I would like it to be returning WU's at an optimum level.
On the XP system at work (P4D 3.2, 3870), the CCC almost always show my GPU clock at 300MHz (it blips up occassionally, but then settles back down to 300) and my GPU activity bounces around between 40% and 70%, and that appears to fluctuate in relation to the reported GPU clock. At 300MHz., it's in the 70% area. When the GPU clock jumps back up, the GPU activity percentage is at its low point. If I look at CPU usage, the two CPU cores are bouncing around between 45% and 60%, I'm never pegging the either core at 100% usage, so it doesn't seem like the GPU throttling up and down is related to fluctuations in CPU cycle availability.
It's odd. Just need time to dig into it.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 11:03 am
by ChronoReverse
It's single threaded so it's really just using one core (Windows scheduling just happens to make it appear to be spread across the cores).
I've settled for running the Windows SMP client on three cores for about ~1700PPD and the GPU2 client on the remaining core for 1650PPD. This might not be the fastest setup for my Q6600 but I want to run the GPU2 client for beta testing purposes.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 12:44 pm
by Gerbil Jedidiah
ChronoReverse wrote:It's single threaded so it's really just using one core (Windows scheduling just happens to make it appear to be spread across the cores).
I've settled for running the Windows SMP client on three cores for about ~1700PPD and the GPU2 client on the remaining core for 1650PPD. This might not be the fastest setup for my Q6600 but I want to run the GPU2 client for beta testing purposes.
That's more than enough PPD. I think you have the right idea.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 9:02 pm
by Haggis!
One the machine at home, I've overclocked the 3870 to 832 core, 1141 memory and FahMon is showing 1600 ppd on the GPU. Like the P4D, on the E3110 the GPU activity bops around between 40% and 80%, and the core speed on the GPU bounces back and forth, but between ~700MHz. and 832MHz.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 9:15 pm
by 2_tyma
Haggis! wrote:One the machine at home, I've overclocked the 3870 to 832 core, 1141 memory and FahMon is showing 1600 ppd on the GPU. Like the P4D, on the E3110 the GPU activity bops around between 40% and 80%, and the core speed on the GPU bounces back and forth, but between ~700MHz. and 832MHz.
thats weird that your gpu core speed jumps like that, mine only jumped when i didnt overclock it (300 to 775 and back)
now i got my core at 877 it stays locked at 877 no matter what i do.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 10:08 pm
by ChronoReverse
Well, I'm getting ~85% utilization now with Project 4701. Mind you, my 3850 is running 669MHz and the single Q6600 core I allocated for it is at 3GHz. Project 2799 under the same conditions gets about 93% utilization.
I bet when bigger WU's are available, we'll easily see 99% utilization.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 10:24 pm
by Gerbil Jedidiah
ChronoReverse wrote:Well, I'm getting ~85% utilization now with Project 4701. Mind you, my 3850 is running 669MHz and the single Q6600 core I allocated for it is at 3GHz. Project 2799 under the same conditions gets about 93% utilization.
I bet when bigger WU's are available, we'll easily see 99% utilization.
Are you worried 99% util will kill your videocard? Any info on how safe it is to run a card at that util 24/7?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 10:41 pm
by 2_tyma
david00214 wrote:ChronoReverse wrote:Well, I'm getting ~85% utilization now with Project 4701. Mind you, my 3850 is running 669MHz and the single Q6600 core I allocated for it is at 3GHz. Project 2799 under the same conditions gets about 93% utilization.
I bet when bigger WU's are available, we'll easily see 99% utilization.
Are you worried 99% util will kill your videocard? Any info on how safe it is to run a card at that util 24/7?
im curious about this too, i dont really think itd harm it as long as temps are under control however, but dont quote me on that, jus a guess
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 17, 2008 11:45 pm
by ChronoReverse
Considering I sometime play 3D games (which fully use the GPU) for half a day at the time, I don't think it's really an issue as long as the card doesn't overheat. If a video card burns itself out simply being used when it's not even overclocked, there's something serious wrong (I'd think anyway).
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Wed Apr 23, 2008 2:14 pm
by 7im
david00214 wrote:
Are you worried 99% util will kill your videocard? Any info on how safe it is to run a card at that util 24/7?
No. The GPU2 folding client is blessed by ATI who helps to write the code. If your card dies before the warranty period, simply RMA it back to ATI.
My X1900XTX has been running the GPU1 client, and I started that almost 2 years ago, and with an overclocked core. YMMV.
The question you have to answer is can you afford to own a race car and drive it like a race car, or are you more of a soccer Mom who only drives the car (slowly and carefully) to the store every Saturday for groceries, so that it will last a very long time, well passed its normal and useful expected lifespan. (pardon the exaggggeration, most people fall somewhere in the middle)
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Wed Apr 23, 2008 10:48 pm
by leor
will GPU2 work on an x1950?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Wed Apr 23, 2008 11:29 pm
by Haggis!
No. GPU2 runs on 2000 and 3000 series cards.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 24, 2008 6:20 am
by Gerbil Jedidiah
Haggis! wrote:No. GPU2 runs on 2000 and 3000 series cards.
And the upcoming 4000 I hope? Yes?
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Thu Apr 24, 2008 7:26 am
by 41.60l23
I don't see why not. The only differences between the 3000 and 4000 series are the speed and number of the core, RAM, and stream processors. The X1000 series and the X2000 series represented a jump from vertex/shader to a unified shader architecture.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Tue May 06, 2008 6:35 pm
by themattman
Quick question. If I paired up an E2180 with a 3650, would I be limited by the CPU? Both those would cost around $140 total. Would it better to spend the $140 on a better processor and forget the ATI graphics card? I want to get the most points possible with light overclocking of the CPU and graphics card.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Fri May 09, 2008 10:49 pm
by Flying Fox
themattman wrote:Quick question. If I paired up an E2180 with a 3650, would I be limited by the CPU? Both those would cost around $140 total. Would it better to spend the $140 on a better processor and forget the ATI graphics card? I want to get the most points possible with light overclocking of the CPU and graphics card.
If things haven't changed much, the GPU client should eat a CPU core but only for polling, so the speed of the CPU itself does not mean a lot. If you are running an SMP client at the same time there will be issues.
If you are strictly after points then yes just run the SMP client with a better CPU (get the E4K series with the larger cache).
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Fri May 09, 2008 10:59 pm
by Heiwashin
themattman wrote:Quick question. If I paired up an E2180 with a 3650, would I be limited by the CPU? Both those would cost around $140 total. Would it better to spend the $140 on a better processor and forget the ATI graphics card? I want to get the most points possible with light overclocking of the CPU and graphics card.
put that 300 towards another ps3 to fold if you want a pure folding machine.
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Fri May 09, 2008 11:31 pm
by Tarx
note: some of the latest GPU2 projects now fully load a fast GPU like the 3850 even with a slower CPU (e.g. 2GHz AMD64).
Right now it is pretty hard to judge what works best for PPD and the PPD aren't even finalized for this client.
On a dual core CPU, it doesn't make sense to run an SMP client with a GPU2 client.
If just running the GPU2 client (and a regular client for the other core), then can get a cheap dual core CPU (e.g. an E2180 for about $70). I'm not sure if there are any inexpensive motherboards that can run a second GPU2 client (would want 8x PCIe minimum)
For Windows SMP (i.e. no Linux SMP under VMware), the E7200 looks like a great CPU for under $140.
For Linux SMP under VMware, the quad core Q6600 at under $230 looks to be the best value (running two Linux SMP projects under VMware = aheckofalot of points!) until the Q9450 prices get reasonable.
Not discussed yet included is the power use and electricity cost factor. (An E7200 is very thrifty, Q6600 not so much).
Re: GPU2 CLIENT IS HERE!!!! THIS IS NOT A JOKE!!
Posted: Fri Jun 20, 2008 8:48 pm
by Mr.Lif
allright so I'm going to bump this topic as it's relevant to my conundrum.
After hearing about the 1500 ppd from a 3850, I find it as quite an advancement from my current folding farm which consists of a ps3 and my main computer with a 3870 and an x2 4400 on socket 939. Now looking back, I get less than 800 ppd from the processor.
Anyway I fire up the gpu2 client and it never seems to work for me, the log file says something along the lines of this:
[01:33:05] CoreStatus = FFFFFFFF (-1)
[01:33:05] Client-core communications error: ERROR 0xffffffff
The GPU2 faq says it may be a problem with OEM drivers. I don't see how that's possible as the ones I'm using are catalyst 8.6.
I've tried 8.5 as well and neither show any change.
Edit: uh... nevermind. I just reinstalled it with a fresh download from the website and. Hah! it works! I only downloaded the previous gpu2 client not three days ago. Sometimes I just don't get it.