I've been running a lot of GFN18 sieving tasks. Here are some findings on my current machine. Gigabyte Geforce 1060 6gb, AMD Fx-8350, 32gb ram, win 10. Both GPU and CPU overclocked.



The gfvsvcuda app is at least half the speed of the gfnsvocl app on single tasks even with a B of 13. It runs about 40P per day.

On the ocl app with B13 on a single task it runs around 90P per day.

To improve upon this, I found running 2 simultaneous tasks using B10 they each run around 54P per task giving it 108P per day total.

A third task yields no speed improvement.

Yes using B13 and 2 tasks at B10 all slow the display time down so don't run them at those levels if the machine is being used.

A couple more findings with the gfnsvcuda app...


The most performance I could get was with 4 apps running with B10. They averaged 17.1P each or 68.4P total per day.

With 3 tasks at B13 it ran 22.7P each or 68.1P per day.

So still the OCL performs better. The GPU usage according to GPU-Z is also more steadily peaked with the OCL apps vs a single or a couple CUDA apps.



Anyone else have similar findings with their configurations?

I think my CPU is a limiting factor too... haven't quite reached that single 104... two instances gets that while the single app averages around 89-94.

And I didn't even think trying to run 6 at once with the cuda app. As you reported, the ocl is still faster overall by P per day.


as for the 1070 TI... that's going to be nice. I'm not sure of speed improvement scale, but do be sure to let us know.

So almost a double improvement in performance when running 3 tasks on the 1070 ti vs the 2 tasks on a 1060. 200P vs 108P. Interesting find about the CPU. It's a little newer than my fx 8350 but I bet it's a slight bottleneck as mine is.

Curious to see more results and findings.

So far the majority of my manual sieving efforts have been with GFN18, now with the others open I will see what sort of performance I get with GFN17 and the GFN21 and GFN22 projects.

A quick test of a P or a couple P showed GFN17 to be quite a bit slower and GFN22 to be a lot faster. Exact numbers to come in a few days.

This makes perfect sense based on the P credit differences for those sieving projects.

GFN22 sieve test results

With GFN22 sieving on my current setup it seems there is little difference in performance by running 2 simultaneous sieves with B10 vs running a single sieve with B13. Also a single task with B10 is only marginally slower.

B10 single task - around 102P-104P/day
B10 two tasks - 53P/day each, 106P/day total
B13 single task - 108P/day (but with significant screen lag)


Default (B7) single task - 81P/day
Default (B7) two tasks - 42P/day each, 84P total
B9 single task - 98P/day
B9 two tasks - 49/day each, 98P/day total
B11 single task - 105P/day
B12 single task - 106P-107P/day


So it seems the optimal for GFN22 is to run a single task and and save a CPU core. Why waste a whole extra core just for a max 1-2P/day improvement?

It is also interesting to note that this is about the same P/day as GFN18 tasks but with lower credit per P.

Will check GFN17 next, but not today.

Freeing up one cpu core completely instead of letting even a low priority process use it, it is now sieving 112-114P/day for GFN22 and B13. And with B10 105-07P/dat.

So a little extra processing time helps it as it figures. My CPU is an AMD FX 8350. Single core performance is a little low even with hyperthreading.

Updates now on the GFN17 sieving.

From what I am finding the lower the number, the more cpu power is needed to achieve the same P/day.

This held true for the GFN21 and GFN22 sieve tasks compared to the GFN18 sieve tasks and now the GFN17 really supports this as well. With my current AMD 8350 the CPU single core performance is the bottleneck. That is why running multiple instances of the sieves are needed to max out the GPU. With GFN22 and from what I've seen with GFN21, running one instance achieves around 105-108P/day using the lowest amount of CPU. It would take 2 GFN18 sieving instances to achieve that same 105-108P.

Now with GFN17, with 2 tasks, at around 44-45P/day each, it still is not reaching peak performance. So I added a third instance. Now with 3 tasks running at around 36P/day each I am reaching max performance of 108P. All tasks are being tested mainly with B10 which i have found to be a good balance for my configuration.

So we can see why there is more credit per P for these tasks, more CPU power/time is required, thus more work, but not necessarily by the 1060 GPU I am running.



I hope these results help everyone run their sieving at peak efficiency.

GFN21 single instance sieving at 111P/day now with slight overclocking on the OC version of the Gigabyte 1060 6GB card. Little difference in temperature, slight increase in power consumption.

Updates now on the GFN17 sieving.

From what I am finding the lower the number, the more cpu power is needed to achieve the same P/day.

This held true for the GFN21 and GFN22 sieve tasks compared to the GFN18 sieve tasks and now the GFN17 really supports this as well. With my current AMD 8350 the CPU single core performance is the bottleneck. That is why running multiple instances of the sieves are needed to max out the GPU. With GFN22 and from what I've seen with GFN21, running one instance achieves around 105-108P/day using the lowest amount of CPU. It would take 2 GFN18 sieving instances to achieve that same 105-108P.

Now with GFN17, with 2 tasks, at around 44-45P/day each, it still is not reaching peak performance. So I added a third instance. Now with 3 tasks running at around 36P/day each I am reaching max performance of 108P. All tasks are being tested mainly with B10 which i have found to be a good balance for my configuration.

So we can see why there is more credit per P for these tasks, more CPU power/time is required, thus more work, but not necessarily by the 1060 GPU I am running.

I hope these results help everyone run their sieving at peak efficiency.

Updates now on the GFN17 sieving.

From what I am finding the lower the number, the more cpu power is needed to achieve the same P/day.

This held true for the GFN21 and GFN22 sieve tasks compared to the GFN18 sieve tasks and now the GFN17 really supports this as well. With my current AMD 8350 the CPU single core performance is the bottleneck. That is why running multiple instances of the sieves are needed to max out the GPU. With GFN22 and from what I've seen with GFN21, running one instance achieves around 105-108P/day using the lowest amount of CPU. It would take 2 GFN18 sieving instances to achieve that same 105-108P.

Now with GFN17, with 2 tasks, at around 44-45P/day each, it still is not reaching peak performance. So I added a third instance. Now with 3 tasks running at around 36P/day each I am reaching max performance of 108P. All tasks are being tested mainly with B10 which i have found to be a good balance for my configuration.

So we can see why there is more credit per P for these tasks, more CPU power/time is required, thus more work, but not necessarily by the 1060 GPU I am running.

I hope these results help everyone run their sieving at peak efficiency.

So do you open 3 separate Admin CMD boxes to run 3 workunits at once? I have a 1080Ti and run one GFN 17 wu at a time but would love to try running at least one more at the same time. I don't OC or anything like that but what you are doing sounds interesting.

A note on overclocking...

while it has slightly improved sieving performance, benchmark performance, and a little game performance, I have been having issues running the GFN non sieving GPU tasks. So I'm still fiddling with it to not have it mess those up. Not sure if it's the memory clock or the GPU clock increases. Those tasks show much more memory controller usage than the sieving tasks. So tweaking is needed. The sieving difference is only a few P per day so it's not all that major to restore it to the lower settings. Primgrid is an excellent test to see if your cards can run stable or not :) It always has been.

edit - seems the voltage increase was causing issues, so back to a slightly lower config now and no more problems :) Remember they do say "overclocking not recommend." and there is a good reason for it. The apps are very efficient and really push your GPU to the max!. Love PrimeGrid!

Updates now on the GFN17 sieving.

From what I am finding the lower the number, the more cpu power is needed to achieve the same P/day.

This held true for the GFN21 and GFN22 sieve tasks compared to the GFN18 sieve tasks and now the GFN17 really supports this as well. With my current AMD 8350 the CPU single core performance is the bottleneck. That is why running multiple instances of the sieves are needed to max out the GPU. With GFN22 and from what I've seen with GFN21, running one instance achieves around 105-108P/day using the lowest amount of CPU. It would take 2 GFN18 sieving instances to achieve that same 105-108P.

Now with GFN17, with 2 tasks, at around 44-45P/day each, it still is not reaching peak performance. So I added a third instance. Now with 3 tasks running at around 36P/day each I am reaching max performance of 108P. All tasks are being tested mainly with B10 which i have found to be a good balance for my configuration.

So we can see why there is more credit per P for these tasks, more CPU power/time is required, thus more work, but not necessarily by the 1060 GPU I am running.

I hope these results help everyone run their sieving at peak efficiency.

So do you open 3 separate Admin CMD boxes to run 3 workunits at once? I have a 1080Ti and run one GFN 17 wu at a time but would love to try running at least one more at the same time. I don't OC or anything like that but what you are doing sounds interesting.

Yes precisely that. 3 separate CMD boxes, 3 separate instances. That is because those CMD windows utilize one CPU core/thread and on my AMD 8350 it is the bottleneck with low single core CPU performance. In order to get the max utilization out of the GPU on my setup, it takes 3 instances of running GFN17.

For the GFN21 and GFN22 tasks it runs at peak efficiency with only a single instance and one CMD window used.

Thank you very much, that won't work for me right now then as I'm only using a dual core AMD 5600 to run the 1080Ti. I may have to move things around again...hmmmm.

Thank you very much, that won't work for me right now then as I'm only using a dual core AMD 5600 to run the 1080Ti. I may have to move things around again...hmmmm.

You are very welcome Mikey. I would think your CPU would be a bottleneck also. I didn't even see it on the list for cpu-z benchmarks. But you can download the app for windows - https://www.cpuid.com/softwares/cpu-z.html and run it yourself and see what sort of single core/thread performance you get and compare it to other cpu's listed. My AMD 8350 as it's configured scores around 208 single core, and 1600 with 8 threads. There are some CPU's listed in the app, but more can be found for comparison online.

Good luck Mikey. Happy sieving! :) That 1080Ti should tear it up! If you can't achieve max performance with the GFN17 sieves, try one of the others I tried like GFN21 or GFN22 and see if you get a higher P/day.

edit: I found it listed here - https://valid.x86.fr/bench/2 it only scores a 199 for 2 threads/cores. That's definately going to be a limiting factor for your 1080ti sieving.

Thank you very much, that won't work for me right now then as I'm only using a dual core AMD 5600 to run the 1080Ti. I may have to move things around again...hmmmm.

You are very welcome Mikey. I would think your CPU would be a bottleneck also. I didn't even see it on the list for cpu-z benchmarks. But you can download the app for windows - https://www.cpuid.com/softwares/cpu-z.html and run it yourself and see what sort of single core/thread performance you get and compare it to other cpu's listed. My AMD 8350 as it's configured scores around 208 single core, and 1600 with 8 threads. There are some CPU's listed in the app, but more can be found for comparison online.

Good luck Mikey. Happy sieving! :) That 1080Ti should tear it up! If you can't achieve max performance with the GFN17 sieves, try one of the others I tried like GFN21 or GFN22 and see if you get a higher P/day.

edit: I found it listed here - https://valid.x86.fr/bench/2 it only scores a 199 for 2 threads/cores. That's definately going to be a limiting factor for your 1080ti sieving.

Thank you very much, I may have to rethink using the 1080Ti and use one of my 1060's like you have instead, they are both in I7 4770 pc's. Don't get me wrong the 1080Ti has done very well by getting me to the Ruby badge, once my pending credits are given to me, and I've only been doing it since the 21st of June 2018.

So do you open 3 separate Admin CMD boxes to run 3 workunits at once? I have a 1080Ti and run one GFN 17 wu at a time but would love to try running at least one more at the same time. I don't OC or anything like that but what you are doing sounds interesting.

I should be getting my mittens on a GTX 1070 Ti soon, which I intend to put into a computer solely dedicated to sieve GFN18 to completion 24/7. You guys think I can expect ~200 P/day or does it not scale like that?

So do you open 3 separate Admin CMD boxes to run 3 workunits at once? I have a 1080Ti and run one GFN 17 wu at a time but would love to try running at least one more at the same time. I don't OC or anything like that but what you are doing sounds interesting.

How many p/day were you achieving with just that one cmd window on that AMD 5600 machine with the 1080ti GPU?

How many p/day were you achieving with just that one cmd window on that AMD 5600 machine with the 1080ti GPU?

It says anywhere from 36 to 38.5 as it crunches along. I don't run really long units, about 5 to 6 hours each during the day and then longer ones overnight, but I do have down times when I'm snoozing or running errands etc. I'm getting better at predicting those, I'm retired so it's easier, but it doesn't crunch 24/7. The only thing the machine does is the PSA badge for PG.

As far as moving it I have to find a box to put it in that can handle the heat, I had it another box and the pc kept rebooting due to the heat in the case. I already keep the sides off the cases and the room is a constant 82F but when running at 100% gpu's can crank out the heat.

How many p/day were you achieving with just that one cmd window on that AMD 5600 machine with the 1080ti GPU?

It says anywhere from 36 to 38.5 as it crunches along. I don't run really long units, about 5 to 6 hours each during the day and then longer ones overnight, but I do have down times when I'm snoozing or running errands etc. I'm getting better at predicting those, I'm retired so it's easier, but it doesn't crunch 24/7. The only thing the machine does is the PSA badge for PG.

As far as moving it I have to find a box to put it in that can handle the heat, I had it another box and the pc kept rebooting due to the heat in the case. I already keep the sides off the cases and the room is a constant 82F but when running at 100% gpu's can crank out the heat.

Yeah that 36-38 is way way way lower than what you should be getting. That's only 1.5P/hr. With the 1060 as you can see I average around 100p/day. I think JimB posted in another thread that he was getting 200P/day with his 1080tTi. You should find away to utilize it to its full potential. Use a different slower card in that box :) (love that term, can tell you are computer guy).

As for the heat, and 82F ambient temp is rough to work with. My current computer room stays around 70-75F and the GPU never gets hotter than 76-78C but that is a whopping 168-172F!

I do have water cooling on the CPU which keeps things in the case cooler overall and a big side case fan. See what you can figure out. I had an issue with a lot of heat at one point, and had that side of the case off as well, and just put a big box fan next to it blowing directly inside. CPUs don't like the heat, AMDs especially from what I have seen with my own setup. And that's why it's rebooting for sure. Two interesting utilities for windows are Core Temp and Speedfan which show the temps affecting the CPU. Not sure if there is a LINUX version of them.

Your CPU specs are here - http://www.cpu-world.com/CPUs/K8/AMD-Athlon%2064%20X2%205600+%20-%20ADA5600IAA6CZ%20(ADA5600CZBOX).html It lists the safe operating temp at 55-70C. So you need to keep it within that range.

With these sieving tasks it uses both the CPU and the GPU and will max them out and generate some serious heat.

I have a computer room that is 7 feet wide by 12 feet long and 10 feet high with 16 desktop pc's in it, so yes it stays warm. Getting it below 82F isn't the problem, paying for the electricity IS, last month my bill was $700 and it wasn't as hot as it is right now. I can only cool it down by shutting down pc's and that's not what I want to do right now. Oh and I'm a Windows guy, I have run Linux but don't right now.

In comparison to the 1060, a laptop which is a little old, is sieving at 12.4P/day using a NVIDIA Geforce GT 555M. Uses very little power, but almost 10x slower. It does have superior single core performance and runs one GFN17 task the same as one GFN21 or GFN22 task so for more credit per P I am sieving GFN17 with this machine. 25P takes about 2 days.

In comparison to the 1060, a laptop which is a little old, is sieving at 12.4P/day using a NVIDIA Geforce GT 555M. Uses very little power, but almost 10x slower. It does have superior single core performance and runs one GFN17 task the same as one GFN21 or GFN22 task so for more credit per P I am sieving GFN17 with this machine. 25P takes about 2 days.

I have a computer room that is 7 feet wide by 12 feet long and 10 feet high with 16 desktop pc's in it, so yes it stays warm. Getting it below 82F isn't the problem, paying for the electricity IS, last month my bill was $700 and it wasn't as hot as it is right now. I can only cool it down by shutting down pc's and that's not what I want to do right now. Oh and I'm a Windows guy, I have run Linux but don't right now.

OMG you are making me jealous of the computing power :) the bill not so much. I understand now. I have a similar issue over the summer with the PC stuff generating heat and then using ACs to cool it down. Two opposites duking it out and twice the electric lol I'm sure whatever benefit you are providing with those PCs is appreciated by whatever sites or people are utilizing them. Happy crunching Mikey, also love your icon/avatar :)

I just discovered the W parameter. W1 is a work around for NVIDIA cards and high CPU usage. I am sieving GFN22 now at the same P per day or close to it still with B10 at a rate of 101P/day. But, it is only using 2.5% of the total CPUs compared with 12-13% previously. Interesting. I'm curious how GFN17 will do, even with 3 sieves at once, it would save CPU usage possibly.

I just discovered the W parameter. W1 is a work around for NVIDIA cards and high CPU usage. I am sieving GFN22 now at the same P per day or close to it still with B10 at a rate of 101P/day. But, it is only using 2.5% of the total CPUs compared with 12-13% previously. Interesting. I'm curious how GFN17 will do, even with 3 sieves at once, it would save CPU usage possibly.

I just discovered the W parameter. W1 is a work around for NVIDIA cards and high CPU usage. I am sieving GFN22 now at the same P per day or close to it still with B10 at a rate of 101P/day. But, it is only using 2.5% of the total CPUs compared with 12-13% previously. Interesting. I'm curious how GFN17 will do, even with 3 sieves at once, it would save CPU usage possibly.

With no increase in crunching time?

I just discovered the W parameter. W1 is a work around for NVIDIA cards and high CPU usage. I am sieving GFN22 now at the same P per day or close to it still with B10 at a rate of 101P/day. But, it is only using 2.5% of the total CPUs compared with 12-13% previously. Interesting. I'm curious how GFN17 will do, even with 3 sieves at once, it would save CPU usage possibly.

With no increase in crunching time?

Exactly mikey! Same P/day and using less CPU resources. I'm going to try some GFN17 sieves and see the outcome but for GFN22, it ran 101-105P/day same as when it was using an entire CPU core/thread.

Success! On the laptop, the W1 parameter does reduce CPU usage. (windows 7 task mgr) 13% without using W1, and anywhere bouncing around from 5%, 7%, 8%, 10% with W1.

I bet on a faster CPU it would be even more significant.

And again, no loss in crunching power or increase in time. Achieving the same P/day with or without W1.

Success! On the laptop, the W1 parameter does reduce CPU usage. (windows 7 task mgr) 13% without using W1, and anywhere bouncing around from 5%, 7%, 8%, 10% with W1.

I bet on a faster CPU it would be even more significant.

And again, no loss in crunching power or increase in time. Achieving the same P/day with or without W1.

Success! On the laptop, the W1 parameter does reduce CPU usage. (windows 7 task mgr) 13% without using W1, and anywhere bouncing around from 5%, 7%, 8%, 10% with W1.

I bet on a faster CPU it would be even more significant.

And again, no loss in crunching power or increase in time. Achieving the same P/day with or without W1.

I will have to try it then on my next workunit, thank you!!