Could be the issue with Windows scheduler which still has a lot of catching up to do to achieve parity with how it works on Intel hardware.
I don't know whether benchmarking the new TR host would make BOINC and the OS choose better optimization but I see that host hasn't been benchmarked yet. Still at the default BOINC values for new systems.
I have a teammate that was investigating for over a year why older Intel systems use less cpu resources compared to AMD system in identical configurations.
Turns out the science application was not configured correctly for AMD systems while it was correct for Intel cpus.
So it could be a lot of different variable that are the reason for the discrepancy.
I see the other older Intel host is also faster than the TR host though not as glaringly.
[Edit] I also wonder about the core to core latencies that are generally higher with AMD processors compared to the ring bus architecture of Intel cpus.
You might want to play around with locking a core to the specific gpu thread so that it doesn't get moved off it and around to other dies.
Thanks for the ideas- I was able to benchmark right before I left for the day so that should now be updated. I also noticed that the version of BOINC I installed is older, but I am not sure that would influence the speed of work units.
Like you said, it definitely could be compounding issues that slow the tasks down versus the Xeons we are also running on.
Is it possible in Windows to lock a core to a gpu thread? If so, this is something I am willing to explore.
I won't be back at the workstation until Monday and then we will start digging in deeper.
PS..... no rolling of eye for Windows here. Me too. Sucks for some projects though :(
Can't wait to see what that bad boy can do!
Sure! It was easier to get the TR systems approved. Dell is just coming out with their 4th gen EPYC systems and I know that receiving purchase approval for one of these server systems will require lots of "hoops" to jump through. The Precision systems are well-known by the group that I have to receive approval from. That being said, I have some LONG term projects in the works where I will pursue EPYC setups.
I just had the other two TR systems arrive today so we should see some data from these as well sometime this week (24 core/48 thread).
Ran across a listing for a Ryzen 7 5700G on eBay for under $200.
I am looking at "side grading" my Windows Daily Driver from a Ryzen 3700x + gt710 to the 5700G.
I think I would get a more robust video card out of it.
My previous experience with the previous generations of the CPU say the boinc CPU processing would be slower.
This change would free up a 3700x for backup, testing, etc.
This is the machine I work from home on and sometimes run boinc on. I would be driving two monitors.
---edit---
Just noticed the target MB only has one video port :( So will have to research which other MB I own could drive two without adding back in another video card.
Apparently, an Asus b450-f MB does have two video ports. So I have a solution to drive two video monitors without adding a video card.
-------edit-----
Comments?
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 believe you are incorrect. There is only one video interface in the AMD APU's. The only thing that your Asus b450-f MB example does is to give you two choices of connector, HDMI or DisplayPort.
Only one port can be used at a time.
And as far as running both cpu and gpu tasks on a APU, that is normally not recommended as both types of tasks are slowed down too greatly because of the limited RAM and slow access to the video buffer.
Even though I had claims that I could run multiple monitors I just low bidded another gtx 1060 6gb and another 3700x for a little less money total.
In both cases known results.
Thank you for weighing in.
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!
Does floating point operations per second map on to RAC when you hold number of CPU threads constant?
Take a look at the top 20 list at u@h for instance.
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!
Boinc benchmark listing on the e@h project system details page.
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!
Keith Myers wrote: Could be
)
Thanks for the ideas- I was able to benchmark right before I left for the day so that should now be updated. I also noticed that the version of BOINC I installed is older, but I am not sure that would influence the speed of work units.
Like you said, it definitely could be compounding issues that slow the tasks down versus the Xeons we are also running on.
Is it possible in Windows to lock a core to a gpu thread? If so, this is something I am willing to explore.
I won't be back at the workstation until Monday and then we will start digging in deeper.
Thanks!
Couple of different tools.
)
Couple of different tools. Built-in Process Explorer in Windows SysInternals tools. Replaces Task Manager.
Aftermarket Process Lasso Free Edition which I really liked when I was running Windows.
Holy Moly that's an expensive
)
Holy Moly that's an expensive CPU!!!!
Wow. AUD$10K.
May I ask why you chose a 5995WX over a EPYC?
PS..... no rolling of eye for Windows here. Me too. Sucks for some projects though :(
Can't wait to see what that bad boy can do!
Chooka wrote: Holy Moly
)
Sure! It was easier to get the TR systems approved. Dell is just coming out with their 4th gen EPYC systems and I know that receiving purchase approval for one of these server systems will require lots of "hoops" to jump through. The Precision systems are well-known by the group that I have to receive approval from. That being said, I have some LONG term projects in the works where I will pursue EPYC setups.
I just had the other two TR systems arrive today so we should see some data from these as well sometime this week (24 core/48 thread).
Ran across a listing for a
)
Ran across a listing for a Ryzen 7 5700G on eBay for under $200.
I am looking at "side grading" my Windows Daily Driver from a Ryzen 3700x + gt710 to the 5700G.
I think I would get a more robust video card out of it.
My previous experience with the previous generations of the CPU say the boinc CPU processing would be slower.
This change would free up a 3700x for backup, testing, etc.
This is the machine I work from home on and sometimes run boinc on. I would be driving two monitors.
---edit---
Just noticed the target MB only has one video port :( So will have to research which other MB I own could drive two without adding back in another video card.
Apparently, an Asus b450-f MB does have two video ports. So I have a solution to drive two video monitors without adding a video card.
-------edit-----
Comments?
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 believe you are incorrect.
)
I believe you are incorrect. There is only one video interface in the AMD APU's. The only thing that your Asus b450-f MB example does is to give you two choices of connector, HDMI or DisplayPort.
Only one port can be used at a time.
And as far as running both cpu and gpu tasks on a APU, that is normally not recommended as both types of tasks are slowed down too greatly because of the limited RAM and slow access to the video buffer.
Even though I had claims that
)
Even though I had claims that I could run multiple monitors I just low bidded another gtx 1060 6gb and another 3700x for a little less money total.
In both cases known results.
Thank you for weighing in.
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!
Does floating point
)
Does floating point operations per second map on to RAC when you hold number of CPU threads constant?
Take a look at the top 20 list at u@h for instance.
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!
Where are you getting your
)
Where are you getting your FLOPS value from? Theoretical published? BOINC Benchmarks?
That value is highly dependent on the actual test used. You don't know which test produced the value.
Also I have noticed wildly inaccurate stratospheric FLOPS values for low powered SoC devices which are the slowest computing devices in BOINC.
I believe the same device running under different OS produces very different results also.
So I would say, no FLOPS does not map to RAC.
Boinc benchmark listing on
)
Boinc benchmark listing on the e@h project system details page.
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!