GPU2 open beta

Testing the Windows GPU client.

Re: GPU2 open beta

Postby Tobit » Mon Apr 28, 2008 3:06 pm

mdk777 wrote:my opinion is that the vast majority of the EUE seen with the new client were caused by this cooling issue, on many models beyond the 38xx.

Is this opinion of yours based on EUE's in general with the new client or just the new 2794/5/6 WUs? Because it isn't the case with me. My stock fan and stock fan settings work fine onmy MSI HD3870 and I was still getting EUE's on SOME of the new 2794/5/6 WUs at 50C GPU core temp. Not all of the new WUs EUE'd for me but about 50% of them did. I have never EUE'd on any of the other WU's (47xx and 2799) nor on any of the CPU WU's.
User avatar
Tobit
 
Posts: 620
Joined: Thu Apr 17, 2008 2:35 pm
Location: Manchester, NH USA

Re: GPU2 open beta

Postby RAH » Mon Apr 28, 2008 3:10 pm

Yes its a shame. But I don't think Pande can do anything about it.
The cards have problems just gaming. I believed its mainly due to the controller
that everyone is using. It should be fixed in later cards, but Oh so many are already
out there.
I don't lay the blame on any one, its something that happened. The blame goes to the ones
that don't fix it. One of them will always say its the other ones fault.
Normal business practice you know. :lol:
Image
User avatar
RAH
 
Posts: 614
Joined: Sun Dec 02, 2007 6:29 am
Location: Florida

Re: GPU2 open beta

Postby AussieFX » Mon Apr 28, 2008 3:48 pm

I have the Sapphire 3870 "blue" card which is a single slot cooler and uses the vapour change HSF. I haven't had any temperature problems nor have I lost a WU.

I wonder if the problem is with the dual slot coolers.

If you are having problems with heat and EUE's can you specify what card/cooler combination you have. That should help others who might run into problems and we can catch any potential problems that might arise.
AussieFX
 
Posts: 114
Joined: Thu Mar 27, 2008 11:26 am

Re: GPU2 open beta

Postby Tobit » Mon Apr 28, 2008 3:52 pm

AussieFX wrote:If you are having problems with heat and EUE's can you specify what card/cooler combination you have.

No heat problems here but a 50% EUE rate ONLY WITH 2794/5/6 projects - flawless with 47XX and 2799. GPU runs between 90% and 95% utilization and 50C to 55C core temp as observed with RivaTuner 2.08.

MSI HD3870 with 512MB DDR4
User avatar
Tobit
 
Posts: 620
Joined: Thu Apr 17, 2008 2:35 pm
Location: Manchester, NH USA

Re: GPU2 open beta

Postby AussieFX » Mon Apr 28, 2008 3:55 pm

I should change that to heat related EUE's. :D
AussieFX
 
Posts: 114
Joined: Thu Mar 27, 2008 11:26 am

Re: GPU2 open beta

Postby mdk777 » Mon Apr 28, 2008 4:04 pm

Tobit,

Yes, I meant the new 2794/5/6 WU. There are of course other factors. When you go from 45-80% GPU utilization, to 100% continuous load, you also significantly change the load factor on your power supply.
I have 4 gig of memory, if I run at 2.075 volt my system never crashes, but when I had it set at 2.00 volt, running the SMP client caused a hard lock(requiring a jump reset of the bios)
Similarly, when I first ran the new client, I had no EUE. However, when I ran two clients on one card, (to get the GPU utilization to approach 100%, I did experience a few EUE after reaching 100%.) Since I made many changes, to fan speed, OC, voltages, and bus speeds subsequently, I cannot tell you cooling was the only problem.

These clients stress available amps, bus chip sets, memory, and cooling. GPU memory can overheat from inadequate cooling when the GPU CORE does not.

Since I already was maxing out my GPU to 100% load running two clients on one card prior to the new 2794/5/6 WU, I did not have any EUE when they came out(and I went to one client per card{at 100% utilization})

What was I doing differently from you? The only clear thing was setting my fans to 100% and leaving them.

My opinion is of course limited to my very few sample of cards, CPU, power supplies and OS.
Since we don't have access to the WU anymore, there is no way to test, but if you were getting 50% failure, setting the fans to 100% would have given you an immediate read if this was the problem.
Transparency and Accountability, the necessary foundation of any great endeavor!
mdk777
 
Posts: 813
Joined: Fri Dec 21, 2007 4:12 am

Re: GPU2 open beta

Postby Tobit » Mon Apr 28, 2008 4:17 pm

mdk777 wrote:Since we don't have access to the WU anymore, there is no way to test, but if you were getting 50% failure, setting the fans to 100% would have given you an immediate read if this was the problem.

Possibly but I'm not entirely convinced it's a cooling issue yet. Some of them failed at 1% to 4% of completion while others made it to 98%. I could see it being a heat issue if they were all EUE towards the middle to end but not right at the start of the WU. I wish we still had access to these projects and feel Vijay was a bit premature with pulling the plug. I was merely reporting my findings as each one EUE'd and didn't mean to cause such a panic.

I'm 99.9% sure it isn't a power supply problem either. It's a 700W SeaSonic M12 that was purchased not very long ago.
User avatar
Tobit
 
Posts: 620
Joined: Thu Apr 17, 2008 2:35 pm
Location: Manchester, NH USA

Re: GPU2 open beta

Postby X1900AIW » Mon Apr 28, 2008 4:56 pm

No heating problems, no EUEs, everything is running, but no gaming, just FAH running at the card: Sapphire HD3850/512MB @796/999 and @stock cooling, 68 degrees controlled by RivaTuner: 55% fan speed (can´t hear it).

Q6600@3.0GHz is running 3 BOINC-WUs since days (24/7-system), no difference in stability between GPU2-projects, but GPU activity is low at project 4707.

Image
User avatar
X1900AIW
 
Posts: 151
Joined: Thu Apr 17, 2008 8:50 am
Location: Germany

Re: GPU2 open beta

Postby codysluder » Tue Apr 29, 2008 6:11 am

mdk777 wrote:I flunked out of engineering 25 years ago, but I think an industry that produces 500 GIG flop, 666 million transistor cards, could mange to write a fuzzy logic program to control the Temp. of their cards a little better.


As a current engineer, it doesn't need fuzzy logic, but it does need a better control algorithm. You're seeing a fan control that is Proportional. With a controller that has a PID control algorithm (Proportional plus Integral plus Derivative) it would be able to anticipate the changes (by calculating the rate of increase (Derivative) and adding a factor for that. To maintain an absolute maximum, the Integral factor can add a factor whenever the temperature exceeds a maximum value. If it stays a fractional degree above that limit, the longer it stays there, the higher the fat rate will go until it actually cools off (or it could shut it down or exercise some other control to prevent damage.
codysluder
 
Posts: 2128
Joined: Sun Dec 02, 2007 12:43 pm

Previous

Return to Windows GPU2 (ATI HD 2xxx/3xxx and beyond)

Who is online

Users browsing this forum: No registered users and 1 guest

cron