I would also like to note this system is #5 on the Single GPU ranking list!
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I have changed the name to more accurately reflect the new concerns.
As you probably know OS3GW is now processing low frequency data. This has resulted in much longer processing times. And in the change over the reward for these tasks first went lower and is now higher at 20,000 vs the previous 10,000 / valid task.
As a consequence of this transition the majority of the Top 50 switched to Brp7/meerKat tasks which haven't slowed down and are paying 3,333 / valid task.
Experiments are under way to see what settings will wring the highest production per GPU out of the new situation.
If you are running Nvidia's MPS you may find 40 percent/3x to be the best setting for OS3GW.
And 70 percent/2x or 40 percent/3x for Brp7/meerKat.
Stay tuned for new developments as they arrive.
HTH.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Important to note (although not a majority of users): If using Windows, meerkat tasks are going to be more productive right now (on nvidia gpus, not sure about amd) due to the cuda version of the application.
Important to note (although not a majority of users): If using Windows, meerkat tasks are going to be more productive right now (on nvidia gpus, not sure about amd) due to the cuda version of the application.
Thank you! I also run an AMD Windows box with a tiny iGpu so that wouldn't have shown up for me at all from a Nvidia POV.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I have changed the name to more accurately reflect the new concerns.
As you probably know OS3GW is now processing low frequency data. This has resulted in much longer processing times. And in the change over the reward for these tasks first went lower and is now higher at 20,000 vs the previous 10,000 / valid task.
As a consequence of this transition the majority of the Top 50 switched to Brp7/meerKat tasks which haven't slowed down and are paying 3,333 / valid task.
Experiments are under way to see what settings will wring the highest production per GPU out of the new situation.
If you are running Nvidia's MPS you may find 40 percent/3x to be the best setting for OS3GW.
And 70 percent/2x or 40 percent/3x for Brp7/meerKat.
Stay tuned for new developments as they arrive.
HTH.
Tom M
Will your research also tell people who have ie 4gb, 6gb or 8gb gpu's whether 1, 2, 3 or even 4 tasks at a time is better? Or are you just sticking to the high end cards with 12+gb of onboard ram?
Will your research also tell people who have ie 4gb, 6gb or 8gb gpu's whether 1, 2, 3 or even 4 tasks at a time is better? Or are you just sticking to the high end cards with 12+gb of onboard ram?
I am not running any lower end video "cards" except a Ryzen 5700G iGpu at the moment. I could put a gtx 1060 6GB onto that Windows box. But I don't have any other hardware.
I am perfectly happy for ANYONE to chime in with their specific experience with a particular brand/model of gpu.
I am pretty sure I ran across a wingman of mine who was running a 6 core cpu with a gtx 970 (probably Windows?).
Anyone?
Respectfully,
Tom M
=====edit==
One thing I am certain of is for the baseline OS3GW application the faster the cpu is and if it is not running any cpu crunching tasks. The faster the OS3GW will be processed.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Will your research also tell people who have ie 4gb, 6gb or 8gb gpu's whether 1, 2, 3 or even 4 tasks at a time is better? Or are you just sticking to the high end cards with 12+gb of onboard ram?
the O3ASBu tasks use a little more than 2GB per task.
i'm only running 2 at a time on my Titan Vs. and it's working well. 3 at a time on the stronger V100, but it's only a tiny better than 2x was anyway. not a huge benefit to doing more than 2 on any card.
you'll need a 5-6GB card or more to do more than 1 at a time.
Important to note (although not a majority of users): If using Windows, meerkat tasks are going to be more productive right now (on nvidia gpus, not sure about amd) due to the cuda version of the application.
Thank you! I also run an AMD Windows box with a tiny iGpu so that wouldn't have shown up for me at all from a Nvidia POV.
I have just switched my Windows box (Ryzen 5700G) over the brp7/meerKat. Who knows, I might even make it past 100,000 Rac.
:)
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Will your research also tell people who have ie 4gb, 6gb or 8gb gpu's whether 1, 2, 3 or even 4 tasks at a time is better? Or are you just sticking to the high end cards with 12+gb of onboard ram?
the O3ASBu tasks use a little more than 2GB per task.
i'm only running 2 at a time on my Titan Vs. and it's working well. 3 at a time on the stronger V100, but it's only a tiny better than 2x was anyway. not a huge benefit to doing more than 2 on any card.
you'll need a 5-6GB card or more to do more than 1 at a time.
Definitely a difference from the higher frequency work. Having a bunch of vram is not too relevant anymore- it doesn't hurt but doesn't really matter. Maybe it could be argued that if someone has a bottleneck at the CPU or system memory, running more simultaneous tasks might be an advantage if it is really slow through the recalc step.
My question (not directly towards anyone) would be: how long does the recalc step take on the system? On this host of ours, the GPU calculation is FAST but it is stuck in the CPU recalc for 300 seconds thanks for the known bottlenecks of this CPU. So, we run 4x because a work unit is usually "stuck" in recalc at any given time. If no bottleneck, I am also leaning towards only 2 or 3 simultaneous work units but will make the transition after the winter break since we are going to be almost shut down (maybe one system running).
(Also, note: our 4090 systems only do GPU work during school hours so they are not examples of what the RAC can be if running full time).
Ian&Steve C.: how do you approach MPS with 2 tasks running in the context of active thread percentage? If I understand MPS correctly (which I might not), when one of the work units is in recalc, then there would only be one task using the GPU cores? Or, do you not use MPS or rebalance percentages when running 2x?
for the new O3ASBu tasks, they have such a short relative time in the recalc step that I'm not overcommiting stuff nearly as much as I was with O3ASHF, and CPU use isnt really slowing them down since there's such a large portion of it in GPU only stage
so for 2x tasks, running MPS at 70%
for 3x tasks, running MPS at 40%
using the 1.08/1.15 CPU app, my recalc step only takes about 1-2 minutes on my EPYC Rome systems, allowing CPU load to stay around 75% from other projects. adding only 1-2 mins on a task that runs for 2700s is hardly any impact. less than 5% of the overall runtime for me. compared with O3ASHF tasks where CPU recalc would take at least 50% of the runtime.
I'm kind of surprised that it's taking 5 mins on that Threadripper system, even if it's only a Zen1+ chip. are you running a bunch of CPU work on it too?
I would also like to note
)
I would also like to note this system is #5 on the Single GPU ranking list!
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I have changed the name to
)
I have changed the name to more accurately reflect the new concerns.
As you probably know OS3GW is now processing low frequency data. This has resulted in much longer processing times. And in the change over the reward for these tasks first went lower and is now higher at 20,000 vs the previous 10,000 / valid task.
As a consequence of this transition the majority of the Top 50 switched to Brp7/meerKat tasks which haven't slowed down and are paying 3,333 / valid task.
Experiments are under way to see what settings will wring the highest production per GPU out of the new situation.
If you are running Nvidia's MPS you may find 40 percent/3x to be the best setting for OS3GW.
And 70 percent/2x or 40 percent/3x for Brp7/meerKat.
Stay tuned for new developments as they arrive.
HTH.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Important to note (although
)
Important to note (although not a majority of users): If using Windows, meerkat tasks are going to be more productive right now (on nvidia gpus, not sure about amd) due to the cuda version of the application.
Boca Raton Community HS
)
Thank you! I also run an AMD Windows box with a tiny iGpu so that wouldn't have shown up for me at all from a Nvidia POV.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Tom M wrote: I have changed
)
Will your research also tell people who have ie 4gb, 6gb or 8gb gpu's whether 1, 2, 3 or even 4 tasks at a time is better? Or are you just sticking to the high end cards with 12+gb of onboard ram?
mikey wrote:Will your
)
I am not running any lower end video "cards" except a Ryzen 5700G iGpu at the moment. I could put a gtx 1060 6GB onto that Windows box. But I don't have any other hardware.
I am perfectly happy for ANYONE to chime in with their specific experience with a particular brand/model of gpu.
I am pretty sure I ran across a wingman of mine who was running a 6 core cpu with a gtx 970 (probably Windows?).
Anyone?
Respectfully,
Tom M
=====edit==
One thing I am certain of is for the baseline OS3GW application the faster the cpu is and if it is not running any cpu crunching tasks. The faster the OS3GW will be processed.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
mikey wrote: Will your
)
the O3ASBu tasks use a little more than 2GB per task.
i'm only running 2 at a time on my Titan Vs. and it's working well. 3 at a time on the stronger V100, but it's only a tiny better than 2x was anyway. not a huge benefit to doing more than 2 on any card.
you'll need a 5-6GB card or more to do more than 1 at a time.
_________________________________________________________________________
Tom M wrote: Boca Raton
)
I have just switched my Windows box (Ryzen 5700G) over the brp7/meerKat. Who knows, I might even make it past 100,000 Rac.
:)
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Ian&Steve C. wrote: mikey
)
Definitely a difference from the higher frequency work. Having a bunch of vram is not too relevant anymore- it doesn't hurt but doesn't really matter. Maybe it could be argued that if someone has a bottleneck at the CPU or system memory, running more simultaneous tasks might be an advantage if it is really slow through the recalc step.
My question (not directly towards anyone) would be: how long does the recalc step take on the system? On this host of ours, the GPU calculation is FAST but it is stuck in the CPU recalc for 300 seconds thanks for the known bottlenecks of this CPU. So, we run 4x because a work unit is usually "stuck" in recalc at any given time. If no bottleneck, I am also leaning towards only 2 or 3 simultaneous work units but will make the transition after the winter break since we are going to be almost shut down (maybe one system running).
(Also, note: our 4090 systems only do GPU work during school hours so they are not examples of what the RAC can be if running full time).
Ian&Steve C.: how do you approach MPS with 2 tasks running in the context of active thread percentage? If I understand MPS correctly (which I might not), when one of the work units is in recalc, then there would only be one task using the GPU cores? Or, do you not use MPS or rebalance percentages when running 2x?
for the new O3ASBu tasks,
)
for the new O3ASBu tasks, they have such a short relative time in the recalc step that I'm not overcommiting stuff nearly as much as I was with O3ASHF, and CPU use isnt really slowing them down since there's such a large portion of it in GPU only stage
so for 2x tasks, running MPS at 70%
for 3x tasks, running MPS at 40%
using the 1.08/1.15 CPU app, my recalc step only takes about 1-2 minutes on my EPYC Rome systems, allowing CPU load to stay around 75% from other projects. adding only 1-2 mins on a task that runs for 2700s is hardly any impact. less than 5% of the overall runtime for me. compared with O3ASHF tasks where CPU recalc would take at least 50% of the runtime.
I'm kind of surprised that it's taking 5 mins on that Threadripper system, even if it's only a Zen1+ chip. are you running a bunch of CPU work on it too?
_________________________________________________________________________