Gpu2 Client Is Here!!!! This Is Not A Joke!!

[ghost101]

I get 2095.20 PPD for the 2799 wu with the specs below. Also, according to rivatuner, my card is correctly using 3d clocks while crunching.

Also, people may want to know that using my card HD3850 864/999 with the cpu at 3.0ghz i got ~1611 PPD. With the cpu at 4.05ghz i got the ppd above.

This is all using a single gpu2 client.

[rcofell]

I crunch on wcg on my q6600's but both my rigs have radeon 3XXX series cards, 2 x3870's and one 3850, how will running this affect my WCG points as it is so cpu dependent?

Well, you could either set Boinc to use only 3 cores, or have it try to share that 4th one and see what happens. As Boinc just runs a separate work unit on each core it might not be so bad, at worst you'd probably only receive 3/4 of your normal on WCG, but at the same time you'd be getting F@H points. I'd give it a try

[Marvin_The_Martian]

Well, you could either set Boinc to use only 3 cores, or have it try to share that 4th one and see what happens. As Boinc just runs a separate work unit on each core it might not be so bad, at worst you'd probably only receive 3/4 of your normal on WCG, but at the same time you'd be getting F@H points. I'd give it a try

Each gpu client needs one core, he has 3 cards and two quads equal eight cores. That leaves 5/8 off his normal wcg output.

[rcofell]

Yeah, I guess I was only thinking in terms of one video card per machine, brain fart

[Marvin_The_Martian]

It's pretty close, 5/8 and 3/4 or 6/8 but I thought I should clarify. Runing boink on 5 cores should leave the 3 spare one's dedicated to gpu folding so it could be worth it if he's willing to let WCG slip abit ( his call lol ).

Still pretty strange, ps3 and gpu folding seems to be quite on par at the moment it's really only the smp client which get's the highest ppd. Was hoping with the new approach gpu client would settle somewhere inbetween the ps3 and smp with points.

[SocketMan]

http://folding.typepad.com/news/2008...e-gpu2-co.html
Quote:

"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)."
End quote

[Mechromancer]

I get the same PPD at 2D clocks as well and it saves A TON of power on this 2900XT of mine. Basically on my slow dual core (Opty 185), I have improved my PPD by 300 points while decreasing my power usage by 20watts compared to the SMP client. The GPU client is economical in my particular case.

Unfortunately for any 65 or 45nm Intel quad core user, it is not economical. A quad core puts out more PPD and uses less power than a GPU. I put a post up on the F@H forums asking for an adjustment on the GPU work unit points. 97 points per WU isn't worth it for most users so they go back to using the SMP client; However, the GPU WUs are very important for the F@H project. Because of this, I think we should get more points per GPU work unit. Id' like to see 150 to 200 points per GPU WU.

[Marvin_The_Martian]

I get the same PPD at 2D clocks as well and it saves A TON of power on this 2900XT of mine. Basically on my slow dual core (Opty 185), I have improved my PPD by 300 points while decreasing my power usage by 20watts compared to the SMP client. The GPU client is economical in my particular case.

Unfortunately for any 65 or 45nm Intel quad core user, it is not economical. A quad core puts out more PPD and uses less power than a GPU. I put a post up on the F@H forums asking for an adjustment on the GPU work unit points. 97 points per WU isn't worth it for most users so they go back to using the SMP client; However, the GPU WUs are very important for the F@H project. Because of this, I think we should get more points per GPU work unit. Id' like to see 150 to 200 points per GPU WU.

Didn't you read the post above yours? I think your whish will come true ( but still crossing my fingers when I say that.. should say; I hope with the newer protiens the awarded points will change as well even without the mentioned speed increase these bigger proteins will have due to increased paralellization ).

[KTE]

AFAIK you need to dedicate at least 30% CPU to the GPU client in 3D Mode or it may error and/or produce low results, it'll use around 25-28% of 4-cores with stock 2.4G Q6600 or 2.5G 9850BE [tested]. My PPD for 13 hours with only a 2600XT is 986 ATM. GPU fan is 45-7% load at 64C and low CPU/RAM clocks. 96-99% GPU stream utilization though.



Better hardware and clocks should get higher. I'm still testing low end yet. After 24 hours will turn CPU/MEM clocks up to the stable settings of >3700/550 and see what it gives, then run the SMP client in combo.

[Marvin_The_Martian]

AFAIK you need to dedicate at least 30% CPU to the GPU client in 3D Mode or it may error and/or produce low results, it'll use around 25-28% of 4-cores with stock 2.4G Q6600 or 2.5G 9850BE [tested]. My PPD for 13 hours with only a 2600XT is 986 ATM. GPU fan is 45-7% load at 64C and low CPU/RAM clocks. 96-99% GPU stream utilization though.



Better hardware and clocks should get higher. I'm still testing low end yet. After 24 hours will turn CPU/MEM clocks up to the stable settings of >3700/550 and see what it gives, then run the SMP client in combo.

Erhm that's still ONE DEDICATED CPU CORE... which has been said throughout the thread.

[KTE]

Erhm that's still ONE DEDICATED CPU CORE... which has been said throughout the thread.

LOL .. yeah I know, that's why I didn't state otherwise

[Marvin_The_Martian]

LOL .. yeah I know, that's why I didn't state otherwise

Why use confusing terms like 30% off the cpu in 3d mode or 25-28% of a quad.

First of all it can not use more then 25% since it's a single thread which occupies one core so all your numbers are off anyway.

Way to post something which is entirely not usefull..

If you have both a q6600 and the 9850be why didn't you post comparison shots between them? Much more interesting

[KTE]

Why use confusing terms like 30% off the cpu in 3d mode or 25-28% of a quad.

Task Manager CPU usage is not confusing, they are a common reference with home systems.

First of all it can not use more then 25% since it's a single thread which occupies one core so all your numbers are off anyway.

You're wrong. If I set it to one core PPD falls to 972.

EDIT: 1-core affinity set, max CPU usage seen as 25% consistent, PPD has been consistent 986 before now but within 5mins fell to 974

Way to post something which is entirely not usefull..

This is exactly what you're doing here. Maybe you didn't understand the purpose of the post, in which case it would be better for you to ask before going off on a tangent. It was to show what a 2600XT can do with it. A few pages back someone suggests only 29xx or above cards are supported which was also not true.

If you have both a q6600 and the 9850be why didn't you post comparison shots between them? Much more interesting

If that's what you wanted then you can easily ask for it in a decent manner. But that's useless in a thread for a GPU client since I implied the PPD and the core usage is the same for both Q6600 and 9850BE at stock.

As for "not loading GPU" possibilities, that is also not true. AC watt system, DC watt tool as well as the P-Tuner DC measurements all show the same power draw frm the system as gathered if I run ATi Tool at full GPU load and the CPU core usage is exactly the same as what I get if I run the ATi Tool GPU test.

[Marvin_The_Martian]

Task Manager CPU usage is not confusing, they are a common reference with home systems.

You're wrong. If I set it to one core PPD falls to 972.

EDIT: 1-core affinity set, max CPU usage seen as 25% consistent, PPD has been consistent 986 before now but within 5mins fell to 974

Where am I wrong then? Please I just said it can't be more then 25% I never referenced ppd besides asking you to post comparison from the phenom and in c2q. You still missed the entire point trying to show single screen shots claiming something. It's a single thread, period. I know you're better then this, you know a single thread can only use one core so what are you trying to do?

This is exactly what you're doing here. Maybe you didn't understand the purpose of the post, in which case it would be better for you to ask before going off on a tangent. It was to show what a 2600XT can do with it. A few pages back someone suggests only 29xx or above cards are supported which was also not true.

If that's what you wanted then you can easily ask for it in a decent manner. But that's useless in a thread for a GPU client since I implied the PPD and the core usage is the same for both Q6600 and 9850BE at stock.

Please, you're the one withholding valid information ( like screenshots off phenom and c2q direct comparisons ) and instead spreading things like over 25% cpu utilization which just is impossible.

And 'the meaning of the post' got pretty lost in wrongly presented utilization numbers. About the 2600xt folding capabilities, current wu's are off small proteins which aren't parallized over all available sp's something the coming wu's will. At that time the 2600 will fall drasticly behind the cards with more streaming processors.

Instead of implying, show screenshots. Pretty please, with sugar on top?

As for "not loading GPU" possibilities, that is also not true. AC watt system, DC watt tool as well as the P-Tuner DC measurements all show the same power draw frm the system as gathered if I run ATi Tool at full GPU load and the CPU core usage is exactly the same as what I get if I run the ATi Tool GPU test.

Don't think this was aimed at me so I won't respond.

Edit:

Got a pm asking me why I seem to be picking a fight. Infact I'm not, I just don't like posts like yours where you're so unclear and leave so much open for interpretation. It's not really usefull information then, but that doesn't give me the right to 'demand' you post other things so I apologize for my attitude, and I edited this post accordingly. I appriciate you tinkering with your hardware and putting some effort into the folding project, just would like you to post your findings in a clearer manner.

[[XC] riptide]

KTE. Your X2600 behaves a little differently than others.My X2900 XT I know loads up rather slightly with these WU's. However, as the Alien from Netherlands said, once the bigger WU's come out we'll see high end cards getting rather warm.

[MikeB12]

once the bigger WU's come out we'll see high end cards getting rather warm.

that's why I've held off on testing this client after my 3 runs posted on earlier in this thread.. the behavior of the 97point beta wu's probably don't reflect an accurate data set of what is to come when real production wu's get here.. these beta units are nice to play around with, but I wouldn't put too much emphasis on getting nitty gritty with them..

7im from fah posted a couple pages back on this..

[SocketMan]

I get 2095.20 PPD for the 2799 wu with the specs below. Also, according to rivatuner, my card is correctly using 3d clocks while crunching.

Also, people may want to know that using my card HD3850 864/999 with the cpu at 3.0ghz i got ~1611 PPD. With the cpu at 4.05ghz i got the ppd above.

This is all using a single gpu2 client.

What's the load % on the 3850 ?
If possible, can you set the affinity to just 1 core for a little bit please - to see if there is a big jump in ppd? THX.

[ghost101]

What's the load % on the 3850 ?
If possible, can you set the affinity to just 1 core for a little bit please - to see if there is a big jump in ppd? THX.

74% load @ 3ghz amd 95% @ 4.05ghz. I have no other crunching programs running so, setting affinity shouldn't make a difference. These ppds are purely through a single gpu2 client.

I don't usually fold and i only used this to see how my hd3850 would perform. Because my hd3850 is overclocked and voltmodded and at the brink of instability, i dont think putting it under 24/7 load would be very good for it even though temps don't go above 55C.

[[XC] riptide]

Ghost! How do you tell what load is on teh GFX?

[[XC] DragonOrta]

Ghost! How do you tell what load is on teh GFX?

Rivatuner will tell you.

[the]

just installed my hd3870, first numbers are here, running gpu2 client only

[01:22:47] Completed 1%
[01:24:07] Completed 2%
[01:25:26] Completed 3%
[01:26:46] Completed 4%
[01:28:03] Completed 5%
[01:29:21] Completed 6%
[01:30:39] Completed 7%
[01:31:57] Completed 8%
[01:33:14] Completed 9%
[01:34:32] Completed 10%
[01:35:51] Completed 11%

ppd is 1074



this is the dual opteron in my sig, only its back to all stock, i let the vapo chill go more than a year ago, and my dual stager i sold today. i just had my appendix pulled and need to pay fot the operation, couldnt use the 2 stager anyway, cold bug. i'll play with the clocks what little i can, see what performance we can get out of it. anyone know how much ram speed affects this client?

[[XC] riptide]

Rivatuner will tell you.

Thanks (noob)

What version of Riva will tell me? Will this monitoring function work with a X2900XT?

[[XC] riptide]

just installed my hd3870, first numbers are here, running gpu2 client only

[01:22:47] Completed 1%
[01:24:07] Completed 2%
[01:25:26] Completed 3%
[01:26:46] Completed 4%
[01:28:03] Completed 5%
[01:29:21] Completed 6%
[01:30:39] Completed 7%
[01:31:57] Completed 8%
[01:33:14] Completed 9%
[01:34:32] Completed 10%
[01:35:51] Completed 11%

ppd is 1074



this is the dual opteron in my sig, only its back to all stock, i let the vapo chill go more than a year ago, and my dual stager i sold today. i just had my appendix pulled and need to pay fot the operation, couldnt use the 2 stager anyway, cold bug. i'll play with the clocks what little i can, see what performance we can get out of it. anyone know how much ram speed affects this client?

eh. Th rig in your sig is a X1800.

EDIT: Nevermind. 3870 you installed.

[ghost101]

Thanks (noob)

What version of Riva will tell me? Will this monitoring function work with a X2900XT?

I'm using rivatuner 2.07 but i think its been a feature of rivatuner for quite some time. It outputs graphs for nearly everything. Core usage, clock speeds, framerates, vram load(xp only), core temps etc.

[SocketMan]

Been running a few /gpu/ clients...
This one shows the most ppd