Just a quick post, with an AMD RX 580, on Kubuntu 18.04 and OpenCL 18.30 AMDGPUPRO driver and a Ryzen 1700, everything at default settings, it takes 9:12 for a LATeah1042L wu to be completed.
Just a quick post, with an AMD RX 580, on Kubuntu 18.04 and OpenCL 18.30 AMDGPUPRO driver and a Ryzen 1700, everything at default settings, it takes 9:12 for a LATeah1042L wu to be completed.
Just a quick post, with an AMD RX 570, on Lubuntu 18.04 and OpenCL 18.3 AMDGPU Mesa driver and a AMD Phenom II X4, everything at default settings, it takes 10:37 for a LATeah1042L wu to be completed. Same host under Win7, default settings, similar wu takes 11:07.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Yes, the AMD cards can benefit a lot from undervolting the GPU, and not just Vegas. My RX 570 (Window 7 64-bit) can operate on Milky Way down to 0.900 volts, and runs quite cool. And on Folding, it can run on 1.050 volts before errors occurs (on both, I can even increase the GPU clock from 1244 to 1348 MHz).
But here in Einstein, I have to keep the voltage at 1.150 and the clock at 1244 MHz (both the defaults for this card), or else invalids result (not errors). So give it a try, but keep an eye on the results.
I have same kind of experiences here in Einstein. I made a short test drive with undervolting RX 580. -8 mV seems to work OK with that card but going to -10 mV offset (which is still extremely SMALL change) produced quickly four fresh results with "validation error". With stock voltage and using only a power limiting slider there hadn't been a single validation error until that. I've tried undervolting also 390´s in the past and they won't tolerate much at all. As small as -5 mV offset is working OK for them but errors started to accumulate with larger offset. Again, using the power limiting slider doesn't cause problems with those cards either.
Over at Seti, one of our crunchers keeps an updated GPU FLOPS vs Reality graph constantly updated as new hardware enters the crunching phalanx. It shows which cards are most productive for credit per hour and most efficient at credit per watt-hour.
Thank you for your reply, that is a great piece of information.
Now do the same 1660ti results (or GPUs in general) from that graph apply to einstein, is SETI NVidia leaning? Or is it better for my powerhungry 780ti to do some other project where NVidia beats AMD energywise and leave AMD cards to do Einstein? I would really like to know how many WU/24h a single 1660ti crunches on Einstein@Home specifically, before even maybe purchase one.
I haven't been crunching for a long time and almost all my past little crunches were E@H, my favorite project. I came back a few days ago to spend some KWh on E@H and with that the low TDP 1660ti drew my attention as potential replacement for the 780ti. In the beginning of me BOINCing I have done a little crunch (testing) with SETI to give the aliens a small computing ear, but personally I find E@H a more interesting project.
It shows which cards are most productive for credit per hour and most efficient at credit per watt-hour.
If you are going to promote that data, you really need to add a prominent disclaimer that the figures apply solely to Seti and may or may not translate to other projects. Otherwise, people will look at what's at the top of the charts and (quite wrongly) blindly assume that it will be just the same for everywhere. For the case of Einstein, the picture is quite different.
Classic Example: At Seti, a GTX 750Ti and an AMD R7 370 are shown right next to each other in the credit/hr chart. They have virtually an identical output over there. I have both types that have been crunching long term here. The R7 370 kills the 750Ti by a factor of 3x. My 750Ti can crunch 23-24 tasks/day whereas my R7 370s can do over 70.
Hwpecker commented that his GTX 780Ti does "46+WU/24hr". There isn't an entry for the 780Ti at Seti but the entry there for a GTX 780 shows around 600 on the credit/hr scale. We could safely assume that a 780Ti should at least perform the same and more likely a bit better. For comparison, look at the entry for the AMD RX 570, lower down in the chart. It shows as around 500 on the same scale. So at Seti, a 780Ti is a lot 'better' than an RX 570.
At Einstein, the picture is quite different. My RX 570s can complete 140 tasks per day. Compared to Hwpecker's 780Ti at 46+, that's again, a factor of 3 better for a supposedly inferior GPU according to the Seti data.
The data/charts over at Seti are an absolutely great resource for people wanting to crunch there. It's really commendable that a volunteer is prepared to devote the time and energy to maintain it. We need to be careful to point out to people looking for a quick recommendation that using that information for where it doesn't apply might lead to poor decisions.
Just a quick post, with an
)
Just a quick post, with an AMD RX 580, on Kubuntu 18.04 and OpenCL 18.30 AMDGPUPRO driver and a Ryzen 1700, everything at default settings, it takes 9:12 for a LATeah1042L wu to be completed.
Chooka wrote:How does the R9
)
I only use my R9 280x during colder months to run FP64 projects like MW/Collatz. The RX series run E@H quite well for the power required.
Millenium wrote:Just a quick
)
Just a quick post, with an AMD RX 570, on Lubuntu 18.04 and OpenCL 18.3 AMDGPU Mesa driver and a AMD Phenom II X4, everything at default settings, it takes 10:37 for a LATeah1042L wu to be completed. Same host under Win7, default settings, similar wu takes 11:07.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
My AMD R9 280 average 11:50
)
My AMD R9 280 average 11:50 for a LATeah1044L on Windows 7 Ultimate x64 Edition,
Quote:Yes, the AMD cards can
)
I have same kind of experiences here in Einstein. I made a short test drive with undervolting RX 580. -8 mV seems to work OK with that card but going to -10 mV offset (which is still extremely SMALL change) produced quickly four fresh results with "validation error". With stock voltage and using only a power limiting slider there hadn't been a single validation error until that. I've tried undervolting also 390´s in the past and they won't tolerate much at all. As small as -5 mV offset is working OK for them but errors started to accumulate with larger offset. Again, using the power limiting slider doesn't cause problems with those cards either.
Anyone knows the WU/24h for
)
Anyone knows the WU/24h for an 1660-Ti? (TDP 125W)
My overclocked GTX780Ti-3GB, design TDP 250W, gpu-z Power Consumption(W) ~190 crunches 1840s per WU running 3 WU simultaneously. That's 46+ WU/24h@3456 credit/WU. Stats always show half average credit...
Over at Seti, one of our
)
Over at Seti, one of our crunchers keeps an updated GPU FLOPS vs Reality graph constantly updated as new hardware enters the crunching phalanx. It shows which cards are most productive for credit per hour and most efficient at credit per watt-hour.
https://setiathome.berkeley.edu/forum_thread.php?id=81962&postid=1997116
Thank you for your reply,
)
Thank you for your reply, that is a great piece of information.
Now do the same 1660ti results (or GPUs in general) from that graph apply to einstein, is SETI NVidia leaning? Or is it better for my powerhungry 780ti to do some other project where NVidia beats AMD energywise and leave AMD cards to do Einstein? I would really like to know how many WU/24h a single 1660ti crunches on Einstein@Home specifically, before even maybe purchase one.
I haven't been crunching for a long time and almost all my past little crunches were E@H, my favorite project. I came back a few days ago to spend some KWh on E@H and with that the low TDP 1660ti drew my attention as potential replacement for the 780ti. In the beginning of me BOINCing I have done a little crunch (testing) with SETI to give the aliens a small computing ear, but personally I find E@H a more interesting project.
Greetings :D
That'll be great for SETI but
)
That'll be great for SETI but AMD cards work better in relation to NV cards here at E@H so any comparison with performance isn't going to line up.
Keith Myers wrote:It shows
)
If you are going to promote that data, you really need to add a prominent disclaimer that the figures apply solely to Seti and may or may not translate to other projects. Otherwise, people will look at what's at the top of the charts and (quite wrongly) blindly assume that it will be just the same for everywhere. For the case of Einstein, the picture is quite different.
Classic Example: At Seti, a GTX 750Ti and an AMD R7 370 are shown right next to each other in the credit/hr chart. They have virtually an identical output over there. I have both types that have been crunching long term here. The R7 370 kills the 750Ti by a factor of 3x. My 750Ti can crunch 23-24 tasks/day whereas my R7 370s can do over 70.
Hwpecker commented that his GTX 780Ti does "46+WU/24hr". There isn't an entry for the 780Ti at Seti but the entry there for a GTX 780 shows around 600 on the credit/hr scale. We could safely assume that a 780Ti should at least perform the same and more likely a bit better. For comparison, look at the entry for the AMD RX 570, lower down in the chart. It shows as around 500 on the same scale. So at Seti, a 780Ti is a lot 'better' than an RX 570.
At Einstein, the picture is quite different. My RX 570s can complete 140 tasks per day. Compared to Hwpecker's 780Ti at 46+, that's again, a factor of 3 better for a supposedly inferior GPU according to the Seti data.
The data/charts over at Seti are an absolutely great resource for people wanting to crunch there. It's really commendable that a volunteer is prepared to devote the time and energy to maintain it. We need to be careful to point out to people looking for a quick recommendation that using that information for where it doesn't apply might lead to poor decisions.
Cheers,
Gary.