the ATLAS info claims the newest servers have “64 cores and 512GB memory”, which matches the specs of this system. Since the CPU is a 32 core, they must be running two of them without SMT to be reporting 64 cores on the host.
So no SMT, plus a higher clock speed, plus faster memory is likely the reason for the difference.
On a side note, my 14 years old Xeon W3680 with 1066 mhz memory speed is achieving slightly faster run times per core if I don't use SMT the difference is it has only 6 cores and that thing has 32.
The old Intel has a TDP of 130, the Epyc of 225 watts. so the new hardware is much more efficient. Yet it's amazing how these old cores sometimes can keep up with new cores in terms of speed. And I remember when I got that Xeon the Gamma-ray pulsar tasks were running faster on a 4 years older Core 2 Duo system with much lower clock speeds. For some reason they seem to love old hardware.
Binary Pulsar search has very different characteristics, they always were faster on newer systems. CPU voodoo at work...
the ATLAS info claims the newest servers have “64 cores and 512GB memory”, which matches the specs of this system. Since the CPU is a 32 core, they must be running two of them without SMT to be reporting 64 cores on the host.
So no SMT, plus a higher clock speed, plus faster memory is likely the reason for the difference.
Ah.
I wonder if I could get the same speed up at Universe at Home? It would be worth it if I could get u@h tasks to run in 1/3 the time they currently are. I can throw more ram on to that box.
Oh, no. He is making system changes again! :)
Somehow I doubt it. If it was THAT easy....
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!
Binary Pulsar search has very different characteristics, they always were faster on newer systems. CPU voodoo at work...
Fade to "Black Magic Women...." tune.
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!
What would be the best CPU for einstein@home? 7800x3d or 7950x?
For CPU crunching? Or Gpu crunching?
For Gpu crunching it almost doesn't matter. For CPU crunching more cores/threads the better (usually).
I think Keith Myers is running one or more 7950x's. The issue is there a memory bandwidth bottleneck on cpu tasks?
There probably is. But for 32 threads or less it MIGHT not be visible.
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!
Tom, you're equipped to comment about the memory usage on the Einstein cpu tasks as you run them. I do not or have not.
What does the System Monitor report for memory usage for the cpu tasks? That value will enable you to figure out whether a standard cpu or a 3D V-cache cpu is best to use.
My best guess is that more cores is more beneficial than more L3 cache.
Well I'd argue that with the CPU intensive parts of the GW tasks the higher the single core performance of a CPU the better. If you do both GPU and CPU crunching you have to balance the system so that you don't get thrown out of boost clock speeds for the GPU task.
This is no issue with the BRP7 tasks, only affects the (current) GW tasks.
If you want to compare run times of CPU tasks, I have a 7800X3D system. In general however it is said that the 7700X is better for productivity because it has a higher boost clock. The 7800X3D has a 10°C lower thermal limit to protect it's 3D cache. To so they did undervolt it to throttle it - but this has the side effect that it is crazy energy efficient. Probably the most energy efficient CPU out there. So if you want to balance the energy bill with performance the 7800X3D for sure is the best choice. Wether the 3D cache has an effect on E@H or not I don't know. All the tests I know show it's great in games and has no effect on the tested productivity apps.
If you want to do GPU crunching only take a look at the 7700X it has among the highest boost clocks. If you want to do CPU+GPU crunching the 7950x for sure is the best you can get.
In general, from over 20 years of BOINC'ing . . . the higher and faster the base clock of the cpu, the faster cpu apps run. Memory has never been any limiter except for the old Rosetta tasks and some PrimeGrid tasks.
Those do in fact benefit from large L3 cache sizes. Specifically the server parts like the Xeons and Epycs which allows the entire task to live and execute within the L3 caches.
My 7950X hosts benefit from the 1 and 2 Ghz clock speed advantages over my older 5950X and Epyc hosts.
No secret sauce. Just the
)
No secret sauce. Just the normal variability of tasks. Look at the oldest tasks in its lists. 3X the times as the ones you referenced.
Most likely that it’s running
)
Most likely that it’s running without SMT.
the ATLAS info claims the newest servers have “64 cores and 512GB memory”, which matches the specs of this system. Since the CPU is a 32 core, they must be running two of them without SMT to be reporting 64 cores on the host.
So no SMT, plus a higher clock speed, plus faster memory is likely the reason for the difference.
_________________________________________________________________________
On a side note, my 14 years
)
On a side note, my 14 years old Xeon W3680 with 1066 mhz memory speed is achieving slightly faster run times per core if I don't use SMT the difference is it has only 6 cores and that thing has 32.
The old Intel has a TDP of 130, the Epyc of 225 watts. so the new hardware is much more efficient. Yet it's amazing how these old cores sometimes can keep up with new cores in terms of speed. And I remember when I got that Xeon the Gamma-ray pulsar tasks were running faster on a 4 years older Core 2 Duo system with much lower clock speeds. For some reason they seem to love old hardware.
Binary Pulsar search has very different characteristics, they always were faster on newer systems. CPU voodoo at work...
Ian&Steve C. wrote:Most
)
Ah.
I wonder if I could get the same speed up at Universe at Home? It would be worth it if I could get u@h tasks to run in 1/3 the time they currently are. I can throw more ram on to that box.
Oh, no. He is making system changes again! :)
Somehow I doubt it. If it was THAT easy....
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!
B.I.G wrote: Binary Pulsar
)
Fade to "Black Magic Women...." tune.
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!
What would be the best CPU
)
What would be the best CPU for einstein@home? 7800x3d or 7950x?
Filipe wrote: What would be
)
For CPU crunching? Or Gpu crunching?
For Gpu crunching it almost doesn't matter. For CPU crunching more cores/threads the better (usually).
I think Keith Myers is running one or more 7950x's. The issue is there a memory bandwidth bottleneck on cpu tasks?
There probably is. But for 32 threads or less it MIGHT not be visible.
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!
Tom, you're equipped to
)
Tom, you're equipped to comment about the memory usage on the Einstein cpu tasks as you run them. I do not or have not.
What does the System Monitor report for memory usage for the cpu tasks? That value will enable you to figure out whether a standard cpu or a 3D V-cache cpu is best to use.
My best guess is that more cores is more beneficial than more L3 cache.
Tom M wrote: For Gpu
)
Well I'd argue that with the CPU intensive parts of the GW tasks the higher the single core performance of a CPU the better. If you do both GPU and CPU crunching you have to balance the system so that you don't get thrown out of boost clock speeds for the GPU task.
This is no issue with the BRP7 tasks, only affects the (current) GW tasks.
If you want to compare run times of CPU tasks, I have a 7800X3D system. In general however it is said that the 7700X is better for productivity because it has a higher boost clock. The 7800X3D has a 10°C lower thermal limit to protect it's 3D cache. To so they did undervolt it to throttle it - but this has the side effect that it is crazy energy efficient. Probably the most energy efficient CPU out there. So if you want to balance the energy bill with performance the 7800X3D for sure is the best choice. Wether the 3D cache has an effect on E@H or not I don't know. All the tests I know show it's great in games and has no effect on the tested productivity apps.
If you want to do GPU crunching only take a look at the 7700X it has among the highest boost clocks. If you want to do CPU+GPU crunching the 7950x for sure is the best you can get.
In general, from over 20
)
In general, from over 20 years of BOINC'ing . . . the higher and faster the base clock of the cpu, the faster cpu apps run. Memory has never been any limiter except for the old Rosetta tasks and some PrimeGrid tasks.
Those do in fact benefit from large L3 cache sizes. Specifically the server parts like the Xeons and Epycs which allows the entire task to live and execute within the L3 caches.
My 7950X hosts benefit from the 1 and 2 Ghz clock speed advantages over my older 5950X and Epyc hosts.