Strange PG AMD GPU Results

[FourOh]

I recently picked up an R9 280X, which I have put to use on Primegrid in place of an HD 7950 in a dual Xeon machine. In an AMD FM1 rig I have an HD 7970 also running Primegrid.

The 7970 is a reference card, clocked at 975/1375. The 280X is an MSI Gaming card clocked at 1020/1500. The 7950 is a Sapphire Vapor-X clocked at 950/1250.

I anticipated the 280X would outperform the 7970 by a small but noticeable margin, but in fact the 280X is 30%-40% slower on PPS Sieve and 10% slower on GFN World Record. In fact the 280X is barely outproducing the 7950 it replaced.

The two cards are on the same, most recent, drivers; both on PCIe 2.0 x16 slots; both showing 98-99% GPU usage when running 2 PPS units or 1 GFN-WR unit. I have 2 threads (one core) dedicated to GPU work on the 280X and one core on the 7970.

Any thoughts on why my 7970 is beating the crap out of my 280X?

Edit: I suppose in the interest of science I should swap the 280X and 7970 to see if the platform is the issue; then isolate variables from there. But, I was hoping someone could just tell me :p

[denim]

Over the years, it just seems like no matter what, PG should be run on NVidia cards unless there is no other choice.

[FourOh]

Over the years, it just seems like no matter what, PG should be run on NVidia cards unless there is no other choice.

There's no doubt Nvidia cards are much more efficient on PG, Maxwell cards in particular. But to get to 100M, I need all my GPUs including the Tahitis. They are probably best suited to the GFN World Record work units, but PPS still pays better by far.

Just for fun, here are my results for PPS by GPU:



GPU
Avg Daily Credits
Watts TDP
Credits/Watt/Day


HD 7750
45,000
55
812


HD 7790
88,000
85
1,035


HD 7950
242,000
225
1,076


HD 7970
402,000
250
1,608


R9 280X
290,000
250
1,160


Quadro K2000M
99,000
55
1,801


GTX 750 Ti
144,000
60
2,400


GTX 970
474,000
175
2,709



I think the items of particular note are 1) my K2000M laptop card outproduces a 7790 Bonaire by a considerable margin and 2) the GTX 970 blows the Tahiti cards away using 70% of the power.

[Shandia]

Just a note. My GTX 580 is doing between 63,440 to 66,813 a day on Genefer. I'll have to run PPS again since I did have it running on my HD5770 and didn't bother to look at the results. ;p

[FourOh]

Just a note. My GTX 580 is doing between 63,440 to 66,813 a day on Genefer. I'll have to run PPS again since I did have it running on my HD5770 and didn't bother to look at the results. ;p

GTX 580 should be strong on PPS - 370k per day +/-.

[zombie67]

Wait, which PG sub-project(s) are being discussed here? Some require DP, some do not. And the various AMD and nVidia cards are better/worse at it.

[FourOh]

Wait, which PG sub-project(s) are being discussed here? Some require DP, some do not. And the various AMD and nVidia cards are better/worse at it.

Specifically I was asking about PPS Sieve and that, oddly, a 7970 is outproducing a 280x by 40%.

[denim]

Man, all this PG talk just makes me want a big horse of a desktop.

[Shandia]

Man, all this PG talk just makes me want a big horse of a desktop.

Better make one with some good gpus. We really need to work on that aspect with the Pentathlon coming up. I saw an article on the next gen of AMD R9s today. Not sure when they will be ready to be out the door......but from FourOh's comparison, we may need to do a comparison of existing hardware on the team. I'm just a little floored to see that his 7970's are doing so good. Manufacturers also need to be added in on the comparison list since I'm in the market for a few for my next build. I know that not all gpus are the same and some of a particular manufacturer seem to be better than others. I'm most likely going to invest in at least one of the EVGA GeForce GTX 960 4GB FTW ACX 2.0+ P/N: 04G-P4-3968-KR just to see what it will do.

[FourOh]

Better make one with some good gpus. We really need to work on that aspect with the Pentathlon coming up. I saw an article on the next gen of AMD R9s today. Not sure when they will be ready to be out the door......but from FourOh's comparison, we may need to do a comparison of existing hardware on the team. I'm just a little floored to see that his 7970's are doing so good. Manufacturers also need to be added in on the comparison list since I'm in the market for a few for my next build. I know that not all gpus are the same and some of a particular manufacturer seem to be better than others. I'm most likely going to invest in at least one of the EVGA GeForce GTX 960 4GB FTW ACX 2.0+ P/N: 04G-P4-3968-KR just to see what it will do.

I'm starting to think my 7970 is somehow special... I'm wondering if DrPop performed some performance-enhancement on it before sending it my way :D I don't see any other Tahiti cards in the PG top hosts earning anything close to 400k per day.

A serious & efficient PrimeGrid machine could have dual (or triple :cool:) GTX 970s - they provide WAY more bang for the buck than the 980 on PG. For some reason, unlike GPUGrid, performance does not scale from the 970 to the 980. The GTX 960 should be a great card for PG or GPUGrid... a little less than double the production of a 750 Ti with similar credits/watt.