• A Great friend to the HardForum with a great kid that he is trying to get a scholorship to continue his schooling. Please give hime a vote! Only 24 hours left! Thanks.
    If you have an VOTE FOR KEENAN!

folding on ati cards

M

mrbigshot

Gawd
Joined
Sep 13, 2005
Messages
865
i know they were working on getting it right for quite a while, did they actually get it right now with v7 beta or is it still planned for a future release?
 
I believe it is better from what I have gathered from people experimenting with it, but so far Nvidia still seems to dominate.
 
The under the v7 client using the improved Core 16, computation is improved, using all GPU cores, but it still takes a dedicated CPU core assigned to the GPU process. I believe this to be the way AMD supports OpenCL in it's drivers. Nvidia's implementation utilizes less CPU cycles.

At least you will be getting a point value more in line with the expense you put in it.
 
mrbigshot said:
i know they were working on getting it right for quite a while, did they actually get it right now with v7 beta or is it still planned for a future release?
Click to expand...

performance is double the old core...
which means a 5870 puts out about 8000 ppd
given game performance thats a little less than half of where it should be...
given the gpgpu hardware differences between ati and nvidia... 2/9 of where it should be...

5870 vs 480 ...5870 has roughly double the compute power... 2.7Tflop vs 1.3Tflop\\
480 has 18k ppd... 5870 should have 36k theoretically

So... why doesn't it?

Bad driver from ati? (milkyway shows accurate performance difference between ati and nvidia...cuda client that was hacked to support opencl by single person)

Hard architecture to work with?

Core 16 young?

small wu?

TWIMTBP?

all possible...
 
Patriot said:
performance is double the old core...
which means a 5870 puts out about 8000 ppd
given game performance thats a little less than half of where it should be...
given the gpgpu hardware differences between ati and nvidia... 2/9 of where it should be...

5870 vs 480 ...5870 has roughly double the compute power... 2.7Tflop vs 1.3Tflop\\
480 has 18k ppd... 5870 should have 36k theoretically

So... why doesn't it?
Click to expand...

It was mentioned over on FF that the current ATI units are small test unit to verify that the client is folding correctly. When the much larger 'real' work units start dropping it is said that CPU usage will drop significantly, and PPD rise accordingly.

H.
 
Patriot said:
performance is double the old core...
which means a 5870 puts out about 8000 ppd
given game performance thats a little less than half of where it should be...
given the gpgpu hardware differences between ati and nvidia... 2/9 of where it should be...

5870 vs 480 ...5870 has roughly double the compute power... 2.7Tflop vs 1.3Tflop\\
480 has 18k ppd... 5870 should have 36k theoretically

So... why doesn't it?

Bad driver from ati? (milkyway shows accurate performance difference between ati and nvidia...cuda client that was hacked to support opencl by single person)

Hard architecture to work with?

Core 16 young?

small wu?

TWIMTBP?

all possible...
Click to expand...

I'm afraid of the possibility of TWIMTBP......

Haitch said:
It was mentioned over on FF that the current ATI units are small test unit to verify that the client is folding correctly. When the much larger 'real' work units start dropping it is said that CPU usage will drop significantly, and PPD rise accordingly.

H.
Click to expand...

I hope that this is correct. I don't fold on my GPU's, but would be willing to fire my my 5850's in the gaming rig if they would actually be worth it. I did run my GPU's for a day and got roughly 7700-7800ppd per GPU for a total of 15-16k ppd total.
 
That is good to hear. I would like to see AMD GPU folding improve. They are generally cheaper, lower power, cool, and made by AMD.
 
I've been running the v7 client on my 6850 since it was released and am getting close to 7000 ppd (running the GPU at 875 MHz). This is about double what the old WUs did. Hopefully when they do release new units they'll load up more of the shader cores.

As far as the client goes it's been pretty stable for me so far. So I think it's worth running on ATI/AMD now. Maybe someone should post something about this in the AMD video cards section?
 
Haitch said:
It was mentioned over on FF that the current ATI units are small test unit to verify that the client is folding correctly. When the much larger 'real' work units start dropping it is said that CPU usage will drop significantly, and PPD rise accordingly.

H.
Click to expand...

it was supposed to happen last core as well...

I don't understand the logic behind larger wu less cpu...
but yes larger wu would actually use more of the cards potential...
unfortunately... they have never had such a large gpu unit...

current test unit is 600atoms... largest previous unit that failed to max 320sp was 1200atom... basically we need something 8-12k atoms...
 
so basically they have no clue why the small ones eat cpu cycles... just that they do...
 
Patriot said:
so basically they have no clue why the small ones eat cpu cycles... just that they do...
Click to expand...

There was another thread, can't find it just now, where it was said/discussed that with the small WU's the CPU is flat out feeding data for the GPU to crunch, and dealing with replies back from the GPU, whereas with the bigger the GPU spends more time working on the data, and there is significantly less effort on the part of the CPU to keep the GPU serviced. I thought it was a post from Mike Houston, but until I find it ........

H.
 
i believe the reason ati's dominate in milky way is the advantage they have in advanced math. comparing that to cuda in that application is apples to oranges.
 
I'm getting better ppd after turning the GPUs off and folding bigadv still

tis a shame, agree with the post above, I was getting around 7KPPD/card but it brought SMP folding to a pathetic crawl...also got sick of the box crashing every time somebody accidentally clicked on a flash element.
 
sorry bout the thread necro but its been about 5 months, anything in the way of improvements?
 
v7 is pretty decent on 5850's and above IIRC, ask EngrChris, he has some of the most experience i think..
 
If anyone is running ATI hardware, what are they producing these days on average? When the V7 client first came out, I know that 5850's were getting around 7500ppd each. I really can't run my GPU's because it would spike my electric bill, but it really wouldn't be worth it as well. With winter coming up, I can see myself running the 5850's to produce heat so long as the production is worth it.
 
You must log in or register to reply here.
Back
Top