Since the profile is CPU only but allows GPU tasks to be run on CPU. Who knows?
I think I switched to Big Aerico(sp). But still allowing non-selected tasks.
I remember having trouble with this generation of the APU in CPU processing. I think it also has a smaller CPU cache than the CPU only chip so it still is likely to have issues.
The first change to different CPU tasks is hours away. And could be much later before all are gone.
The 5700g is due Wednesday but probably will not get installed till later.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Since the profile is CPU only but allows GPU tasks to be run on CPU. Who knows?
I think I switched to Big Aerico(sp). But still allowing non-selected tasks.
I remember having trouble with this generation of the APU in CPU processing. I think it also has a smaller CPU cache than the CPU only chip so it still is likely to have issues.
The first change to different CPU tasks is hours away. And could be much later before all are gone.
The 5700g is due Wednesday but probably will not get installed till later.
Tom M
Your pco will be faster than the one that did this task but the principle is the same:
Completed and validated
31,974
32,086
500
Binary Radio Pulsar Search (Arecibo, large) v1.33 () windows_x86_64
That's a task that finished TODAY and as you can see the credits were only 500 per task and taking @8 hours each!! The pc that did that task is a AuthenticAMD AMD Ryzen Threadripper 1920X 12-Core Processor [Family 23 Model 1 Stepping 1 and NO it's not tweaked like your Team does so yes it could run things faster but I'm not sure those tasks are the 'best bang for the buck' either way. I don't have a Linux pc running those tasks.
Binary Radio Pulsar Search (Arecibo, large) v1.33 () windows_x86_64
That's a task that finished TODAY and as you can see the credits were only 500 per task and taking @8 hours each!! The pc that did that task is a AuthenticAMD AMD Ryzen Threadripper 1920X 12-Core Processor [Family 23 Model 1 Stepping 1 and NO it's not tweaked like your Team does so yes it could run things faster but I'm not sure those tasks are the 'best bang for the buck' either way. I don't have a Linux pc running those tasks.
You can't establish the 'best bang for bucks' metric from this single result row because the number of concurrently running threads is unknown here. This CPU may ran 8...12 threads in parallel or 16 or even 24 (e.g. Threadripper 12-Core), which will greatly influence the overall runtime of this task. The 'best bang for bucks' metric is difficult to measure. One has to try different configurations (#parallel threads, thread's app mix) and do some statistics.
And even if threads are slowed down by many concurrent threads they may give more 'credits per day' if the slowdown (percentage) is less than the additionally generated credits by further parallel threads. I think, you can't guess this from runtimes of single tasks. Instead it needs some logging of runtimes for different configurations (#threads, which apps).
Binary Radio Pulsar Search (Arecibo, large) v1.33 () windows_x86_64
That's a task that finished TODAY and as you can see the credits were only 500 per task and taking @8 hours each!! The pc that did that task is a AuthenticAMD AMD Ryzen Threadripper 1920X 12-Core Processor [Family 23 Model 1 Stepping 1 and NO it's not tweaked like your Team does so yes it could run things faster but I'm not sure those tasks are the 'best bang for the buck' either way. I don't have a Linux pc running those tasks.
You can't establish the 'best bang for bucks' metric from this single result row because the number of concurrently running threads is unknown here. This CPU may ran 8...12 threads in parallel or 16 or even 24 (e.g. Threadripper 12-Core), which will greatly influence the overall runtime of this task. The 'best bang for bucks' metric is difficult to measure. One has to try different configurations (#parallel threads, thread's app mix) and do some statistics.
And even if threads are slowed down by many concurrent threads they may give more 'credits per day' if the slowdown (percentage) is less than the additionally generated credits by further parallel threads. I think, you can't guess this from runtimes of single tasks. Instead it needs some logging of runtimes for different configurations (#threads, which apps).
I didn't know you could multi-thread these tasks!! I'm running then one task at a time on each core of the pc's they are running on.
he's not talking about a single task using multiple threads. he's talking about the effect on the runtime of multiple tasks running concurrently (on different threads). more tasks = more utilization = more power = lower clocks in most cases with modern CPUs. you can also run into slowdowns due to increased cache utilization or memory bandwidth limits. but for einstein it's probably mostly the effect on clockspeed that has the largest impact.
he's not talking about a single task using multiple threads. he's talking about the effect on the runtime of multiple tasks running concurrently (on different threads). more tasks = more utilization = more power = lower clocks in most cases with modern CPUs. you can also run into slowdowns due to increased cache utilization or memory bandwidth limits. but for einstein it's probably mostly the effect on clockspeed that has the largest impact.
he's not talking about a single task using multiple threads. he's talking about the effect on the runtime of multiple tasks running concurrently (on different threads). more tasks = more utilization = more power = lower clocks in most cases with modern CPUs. you can also run into slowdowns due to increased cache utilization or memory bandwidth limits. but for einstein it's probably mostly the effect on clockspeed that has the largest impact.
Sorry, my wording wasn't optimal: threads = virtual CPU cores. e@h tasks represent one compute thread.
Uuuh and your're right. I forgot to think about this feature of most current CPUs. They apply thermal management which limits heat output instead of strictly limiting core frequency: TDP (thermal design power). If one or more cores are idle, others (can) spend more heat with higher core frequency (automatic, selective overclocking). So there are further parameters influencing the 'best bang for bucks' metrics...
And that's only the whole story for large enough desktop/server cases with large and capable CPU cooling which allows to run the CPU at maximum TDP for 24/7. If we talk about small form factor desktop boxes (small business desktops or small media boxes, notebooks) then there are often additional limits set by computer's manufacturer which allow high peak power, but not for more than e.g. 10 minutes. CPU is clocked down automaticely to prevent the system from overheating (manufacturer's specific BIOS feature).
From my empirical tests I'd assume it's not "einstein tasks" but the different e@h science apps which cause different level of CPU stress, even different heat output. (I assume: memory access pattern, total amount of mem loads/writes #bytes, mix of µOps).
Running CPU tasks on a GPU project is a big no, no. Even the fastest server CPU would get absolutely demolished by any AMD/Nvidia GPU of recent years. Your RAC would stay at what it is with or without running CPU tasks.
On a side note, I'm wondering, is it possible to crunch on your behalf from my account, by assigning your CPID as an external one?
I have GPUs to spare, so if you post your CPID here, I would give you a free boost.
Running CPU tasks on a GPU project is a big no, no. Even the fastest server CPU would get absolutely demolished by any AMD/Nvidia GPU of recent years. Your RAC would stay at what it is with or without running CPU tasks.
This thread is about how to maximize RAC using a fixed setup (the alldays work horse on your desk). We don't want to increase electric consumption (by adding GPU cards) but simply maximize RAC. That means optimal use of CPU cores (percentage of cores to use, app mix on those cores), optimal use of CPU internal iGPU.
So it's about getting the science done while saving the planet. And it's about the Olympic idea: to achieve the best performance without outside help. ;-)
Tom M wrote: I have a Ryzen
)
Oops, make that a 5700g
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Since the profile is CPU only
)
Since the profile is CPU only but allows GPU tasks to be run on CPU. Who knows?
I think I switched to Big Aerico(sp). But still allowing non-selected tasks.
I remember having trouble with this generation of the APU in CPU processing. I think it also has a smaller CPU cache than the CPU only chip so it still is likely to have issues.
The first change to different CPU tasks is hours away. And could be much later before all are gone.
The 5700g is due Wednesday but probably will not get installed till later.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Tom M wrote: Since the
)
Your pco will be faster than the one that did this task but the principle is the same:
That's a task that finished TODAY and as you can see the credits were only 500 per task and taking @8 hours each!! The pc that did that task is a AuthenticAMD AMD Ryzen Threadripper 1920X 12-Core Processor [Family 23 Model 1 Stepping 1 and NO it's not tweaked like your Team does so yes it could run things faster but I'm not sure those tasks are the 'best bang for the buck' either way. I don't have a Linux pc running those tasks.
mikey
)
You can't establish the 'best bang for bucks' metric from this single result row because the number of concurrently running threads is unknown here. This CPU may ran 8...12 threads in parallel or 16 or even 24 (e.g. Threadripper 12-Core), which will greatly influence the overall runtime of this task. The 'best bang for bucks' metric is difficult to measure. One has to try different configurations (#parallel threads, thread's app mix) and do some statistics.
And even if threads are slowed down by many concurrent threads they may give more 'credits per day' if the slowdown (percentage) is less than the additionally generated credits by further parallel threads. I think, you can't guess this from runtimes of single tasks. Instead it needs some logging of runtimes for different configurations (#threads, which apps).
Scrooge McDuck wrote: mikey
)
I didn't know you could multi-thread these tasks!! I'm running then one task at a time on each core of the pc's they are running on.
he's not talking about a
)
he's not talking about a single task using multiple threads. he's talking about the effect on the runtime of multiple tasks running concurrently (on different threads). more tasks = more utilization = more power = lower clocks in most cases with modern CPUs. you can also run into slowdowns due to increased cache utilization or memory bandwidth limits. but for einstein it's probably mostly the effect on clockspeed that has the largest impact.
_________________________________________________________________________
Ian&Steve C. wrote: he's not
)
Oh okay that makes sense, thanks
Ian&Steve C. schrieb: he's
)
Sorry, my wording wasn't optimal: threads = virtual CPU cores. e@h tasks represent one compute thread.
Uuuh and your're right. I forgot to think about this feature of most current CPUs. They apply thermal management which limits heat output instead of strictly limiting core frequency: TDP (thermal design power). If one or more cores are idle, others (can) spend more heat with higher core frequency (automatic, selective overclocking). So there are further parameters influencing the 'best bang for bucks' metrics...
And that's only the whole story for large enough desktop/server cases with large and capable CPU cooling which allows to run the CPU at maximum TDP for 24/7. If we talk about small form factor desktop boxes (small business desktops or small media boxes, notebooks) then there are often additional limits set by computer's manufacturer which allow high peak power, but not for more than e.g. 10 minutes. CPU is clocked down automaticely to prevent the system from overheating (manufacturer's specific BIOS feature).
From my empirical tests I'd assume it's not "einstein tasks" but the different e@h science apps which cause different level of CPU stress, even different heat output. (I assume: memory access pattern, total amount of mem loads/writes #bytes, mix of µOps).
@TOM M Running CPU tasks
)
@TOM M
Running CPU tasks on a GPU project is a big no, no. Even the fastest server CPU would get absolutely demolished by any AMD/Nvidia GPU of recent years. Your RAC would stay at what it is with or without running CPU tasks.
On a side note, I'm wondering, is it possible to crunch on your behalf from my account, by assigning your CPID as an external one?
I have GPUs to spare, so if you post your CPID here, I would give you a free boost.
Dido schrieb: @TOM
)
This thread is about how to maximize RAC using a fixed setup (the alldays work horse on your desk). We don't want to increase electric consumption (by adding GPU cards) but simply maximize RAC. That means optimal use of CPU cores (percentage of cores to use, app mix on those cores), optimal use of CPU internal iGPU.
So it's about getting the science done while saving the planet. And it's about the Olympic idea: to achieve the best performance without outside help. ;-)