Almost a waste to use a good GPU like a 2080 for sieving when speed isn't really required. It really doesn't matter at all if you get a "1st" for finishing a sieve compared to finding a prime. It nice to do a lot of work very quickly anyways though.

Now if credit if what you are after, the 2080 tears PPS Sieve up pretty nicely. Ran it for about 12 hrs and averaged 224 seconds to do 2 simultaneous tasks. That's 1min 52 sec per task. Each task yields 3,371 credits. So that's equal to ~1,805 credits per minute. That's 108,300 credits per hour.

Uses about the max power just slightly below 240watts, and hangs around the 98-99% range for GPU usage.

It's a RAC booster beast !
____________

"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment.

Likely because PPS sieve is a CUDA application (rather than OCL), older cards do relatively well compared to newer architectures. For example, a GTX 1080 runs them around 300 seconds. Even an old GTX 660 Ti runs them around 750 sec each. So the new RTX cards have greater performance gaps on some of the GFN projects and the AP27 project than here...but if credit is all one is after, the payout is still here.

Also, I think the RTX cards are bottlenecked a good bit on PPS Sieve. For example, my RTX 2070 had average times of 229 seconds (almost the same as your 2080). I would guess that running 2 PPS Sieves at a time on these cards will yield noticeable credit yields.

How many threads you need to assign to fully utilize GTX 2080 in GFN21?

Thanks. It's looks great.

Penguin wrote:

Now if credit if what you are after, the 2080 tears PPS Sieve up pretty nicely. Ran it for about 12 hrs and averaged 224 seconds to do 2 simultaneous tasks. That's 1min 52 sec per task. Each task yields 3,371 credits. So that's equal to ~1,805 credits per minute. That's 108,300 credits per hour.

If credit's your game, then the best choice for RTX cards seems to be manual Genefer sieving. I don't know what it is about them, but Turing absolutely destroys there. It's a huge jump forward compared to Pascal; even a 2070 is at least twice as fast as a 1080 Ti. My 2080 yields over 3.5 million per day on GFN21.

I run two instances in tandem to keep utilization at 100% (each at around 420 P/day, the benefit compared to one instance is 20-30 P/day), each with the w1 parameter and size B13. Together they use about a single thread, according to the task manager. So I guess one per instance is more than enough in this case. I got 26 of my 32 threads occupied with other BOINC tasks, the rest is kept free.

I was unable to get PPS Sieve running on both my GPUs at the same time. I could get two tasks running using .5 GPUs to run on device 0 which is the 2080 but I couldn't get it to use device 1 no matter what. That's the 1070. It will run two unique projects fine like seti on the 1070 and PPS Sieve with two tasks on the 2080 but disabling all other projects freeing up device 1 it still will not run it side by side with device 0.

I've had other Primegrid sub projects running on both cards like GFN21 for the challenge. I just can't finagle it to work with PPS Sieve. I didn't try just sieving with the 1070 but I will try now just for the heck of it for comparison speed. Thought I could throw both of em at it for a few days and get that next badge.

I was unable to get PPS Sieve running on both my GPUs at the same time. I could get two tasks running using .5 GPUs to run on device 0 which is the 2080 but I couldn't get it to use device 1 no matter what. That's the 1070. It will run two unique projects fine like seti on the 1070 and PPS Sieve with two tasks on the 2080 but disabling all other projects freeing up device 1 it still will not run it side by side with device 0.

I've had other Primegrid sub projects running on both cards like GFN21 for the challenge. I just can't finagle it to work with PPS Sieve. I didn't try just sieving with the 1070 but I will try now just for the heck of it for comparison speed. Thought I could throw both of em at it for a few days and get that next badge.

Managed to shave off 10 sec for 2 tasks down from 224 sec to 214 sec now. So 1 min 47 sec per task on average.

Azmodes suggested lowering the memory clock and upping the GPU clock which works better for sieving. It helped. Memory clock is running at 2040 now stable with 6800 memory stock speed. Temps are lower.

Opening my outdated poor cooling case helped also. 2 GPUs on top of one another needs some good ventilation. So like many have done I simply took off the side of the case and put a fan blowing directly into it on the cards :) 5-6c temp difference from that alone. New case for the new build will be much better with airflow.

Also, I think the RTX cards are bottlenecked a good bit on PPS Sieve. For example, my RTX 2070 had average times of 229 seconds (almost the same as your 2080). I would guess that running 2 PPS Sieves at a time on these cards will yield noticeable credit yields.

Also, I think the RTX cards are bottlenecked a good bit on PPS Sieve. For example, my RTX 2070 had average times of 229 seconds (almost the same as your 2080). I would guess that running 2 PPS Sieves at a time on these cards will yield noticeable credit yields.

Is this a time for 1 wu at a time or 2 wu's?

And if for 1, did you try 2 simultaneous tasks?