Folding on AMD GPUs

[theAnimal]

Don't think I've seen this posted here although I have not been around much this past year.

With the new Core 17 WUs, AMD HD6xxx & 7xxx cards are the new PPD leaders. (Edit: Looks like GTX780 has taken the lead, but 7950 is much higher PPD/$)

My 7970 at stock gets over 80k ppd (ran a few units to test but not running atm, may do for the race).

You will need to request beta units by adding variable client-type with value beta in the appropriate GPU slot.

 

[brownstone]

I just ran across this news a week or so ago and was pretty excited to hear about it. Despite the quoted numbers I've been seeing at various sites, I've only been averaging ~44000 ppd. Any ideas anybody? I've got a slight OC on my 7970 and the beta flag is in place.

 

[theAnimal]

I just ran across this news a week or so ago and was pretty excited to hear about it. Despite the quoted numbers I've been seeing at various sites, I've only been averaging ~44000 ppd. Any ideas anybody? I've got a slight OC on my 7970 and the beta flag is in place.

Is it crunching core 17 WUs, is it running full time?

 

[brownstone]

Is it crunching core 17 WUs, is it running full time?

It is crunching the core 17's and runs 24/7. The only strange thing I've noticed is that it peaks out at 97-98% gpu utilization. It doesn't seem to be cpu limited or temp. throttling...I'm at a loss.

 

[ZipSpeed]

Try sliding the Power Limit to 20% under Catalyst Control Center.

 

[brownstone]

Found that and gave it the slide, but alas, no change.

 

[brownstone]

Alright, I found that for whatever reason my gpu is running at x8 rather than x16. Could this be the issue?

 

[theAnimal]

Alright, I found that for whatever reason my gpu is running at x8 rather than x16. Could this be the issue?

I know that did not used to matter, not sure if that has changed with newer WU but would think not.

 

[Bradtech519]

...

 

[Bradtech519]

sorry posted in wrong thread

 

[Sellyme]

Alright, I found that for whatever reason my gpu is running at x8 rather than x16. Could this be the issue?

Yes, that is the issue. 7970s bottleneck considerably at x8. My 7970s run between 70k and 140k each on Core 17 WUs (averaging out at about 90k on 7810s/7811s, and 105k on 8900s) on x16, but on x8 they would average similar numbers to what you've encountered.

 

[brownstone]

Yes, that is the issue. 7970s bottleneck considerably at x8. My 7970s run between 70k and 140k each on Core 17 WUs (averaging out at about 90k on 7810s/7811s, and 105k on 8900s) on x16, but on x8 they would average similar numbers to what you've encountered.

Interesting. I'll have to figure out a way to experiment with that. In the meantime, I updated the bios on my motherboard, flashed a new bios to my gpu, dropped the OC on my cpu and fiddled with my gpu fan profile a bit and it has helped bump up my points some...not sure which item actually played the biggest role, but I'm happy with the results.

 

[Assimilator1]

What did ppd did you get it upto?

Don't think I've seen this posted here although I have not been around much this past year.

With the new Core 17 WUs, AMD HD6xxx & 7xxx cards are the new PPD leaders. (Edit: Looks like GTX780 has taken the lead, but 7950 is much higher PPD/$)

My 7970 at stock gets over 80k ppd (ran a few units to test but not running atm, may do for the race).

You will need to request beta units by adding variable client-type with value beta in the appropriate GPU slot.

Seeing as the HD 5870 performs very closely to the HD 6950 in games, does it do well with the core 17 WUs?
[update]
I found a partial answer here https://foldingforum.org/viewtopic.php?f=38&t=24225 , although no one's posted a score for a 6xxx yet. But there is a 7870 score which is twice as fast as the only 5870 there (by the same guy ironically).

 

[theAnimal]

What did ppd did you get it upto?

I OCed my 7970 to 1050 which gave me about 104k ppd.

 

[Plimogz]

^

7950@1100Mhz gives me roughly that same PPD.

 

[Assimilator1]

Awesome! , any idea what a 6950, 5870 or a 5850 would do?

 

[Plimogz]

My 6950 w/ unlocked shaders and @900Mhz core (so pretty much a stock 6970) gets a bit less than 20K PPD. It is not at all an efficient F@H cruncher.

Oh, and the 6950 is virtually useless with anything other than 8900 (probably core 17 in general) WUs. Though I guess that the same could be said about AMD cards across the board. Fortunately, Fahcore 17 is the future.

 

[Sunny129]

Yes, that is the issue. 7970s bottleneck considerably at x8.

i can confirm this on the Einstein@Home project as well, though i imagine the difference between PCIe x8 and x16 bandwidth (at least with regard to the PCIe 3.0 standard) has more of an effect on some projects than others. until recently, i was running Einstein@Home on 2 7970s in PCIe 3.0 x16 slots and 1 7970 in a PCIe 2.0 x16 slot. as you're all probably aware, the PCIe 3.0 standard had double the bandwidth of the PCIe 2.0 standard, and so running the single 7970 in the PCIe 2.0 x16 slot was the equivalent of running it in a PCIe 3.0 x8 slot. that 7970 would take close to ~120 minutes to complete 4 concurrently running E@H tasks, whereas the 2 7970s in PCIe 3.0 x16 slots would only take ~70 minutes to complete 4 concurrently running E@H tasks. so while doubling the PCIe bandwidth didn't quite double my performance in Einstein@Home, it increased it substantially (my E@H GPU task run times are only 58% of what they used to be).

 

[VirtualLarry]

My E@H tasks are only using up to 4% GPU load on my 7790. Does this sound right? It seems kind of low to me. CPU is an i3-3240, 8GB DDR3-1600. Win7 64-bit.

 

[Ken g6]

E@H has a "GPU utilization factor" option in its preferences. I guess that's why: So you can run as many E@H WUs on one GPU as your GPU RAM will allow.

But how did this thread get an extra line at the bottom of the "F" in "F@H"?

 

[Sunny129]

Larry, YGPM so as not to derail this F@H thread w/ E@H material...

 

[Assimilator1]

My 6950 w/ unlocked shaders and @900Mhz core (so pretty much a stock 6970) gets a bit less than 20K PPD. It is not at all an efficient F@H cruncher.

Oh, and the 6950 is virtually useless with anything other than 8900 (probably core 17 in general) WUs. Though I guess that the same could be said about AMD cards across the board. Fortunately, Fahcore 17 is the future.

I remember a few years ago that ATI cards were a long way behind Nvidia ones in F@H (can't remember why now), so after GPU crunching F@H for a race I then started Milkyway@Home on my HD 4830, ATIs kicked arse on MW , I think they still do.
Glad to hear that F@H is finally about to bring ATI/AMD cards upto speed.
Btw what's your suedo 6970 ppd on non core 17 WUs? 20k or did that include core 17s?

Oh I've got a HD 5850 now, any idea what ppd that would do? (although I've no plans to switch it from MW for sometime, it's nearly doubling my 4870s output )

 

[Markfw]

Don't think I've seen this posted here although I have not been around much this past year.

With the new Core 17 WUs, AMD HD6xxx & 7xxx cards are the new PPD leaders. (Edit: Looks like GTX780 has taken the lead, but 7950 is much higher PPD/$)

My 7970 at stock gets over 80k ppd (ran a few units to test but not running atm, may do for the race).

You will need to request beta units by adding variable client-type with value beta in the appropriate GPU slot.

OK, the 7950@$400 get 80k ppd, and my GTX780@$500 get 160kppd. How is the 7950 the winner PPD/$ ? or am I missing something.

 

[Sunny129]

OK, the 7950@$400 get 80k ppd, and my GTX780@$500 get 160kppd. How is the 7950 the winner PPD/$ ? or am I missing something.

in all fairness, theAnimal posted that info 3 months ago now, just as AMD GPU prices started to skyrocket due to the GPU mining craze. if the average price of a 7950 was say ~$250 back then, then that puts the 7950 on equal footing w/ the GTX 780 as far as PPD/$ is concerned, w/ the slight advantage to the 7950 if you consider that the GTX 780 consumes more power than the 7950. and if you consider that the average price i paid for each of my 7970s was $225, and that i get 95K PPD out of just one of them while Folding, that's a substantial PPD/$ advantage over the GTX 780...of course by today's AMD GPU pricing however, you'd be right in saying that the GTX 780 has the substantial PPD/$ advantage over either the 7950 or the 7970.

 

[Bubbleawsome]

I've been waiting for the prices to go down since December, and now I think only the 800 series might do it.