Generic CPU discussion

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6453
Credit: 9580039840
RAC: 7392379

Ian&Steve C. wrote:from

Ian&Steve C. wrote:

from what I've read, Determinism = Power/Performance, and the cTDP at it's max should be the best setting.

I set it to "Performance" with cTDP on "auto" and it just started registering 2.7+ MHz.  That might be all I can get out of it.

---edit---

It just regressed to 2.5-2.6 MHz :(

---edit---

The PDF on HPC claims I should also turn off the Virtual Machine and Memory crypto stuff.  What do you think?

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!

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18742420636
RAC: 6999037

Try Power for determinism. 

Try Power for determinism.  That way at least a couple of cores will boost.

cTDP to 170W

But your boost is being held back by your cooling.

 

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3956
Credit: 46958482642
RAC: 64664293

also keep in mind that you

also keep in mind that you wont see the same clock speeds or power draw that Keith and I see on our Rome series CPUs. your CPU is spec'd for only 2.8GHz all-core clock, and that will likely depend on temperature also.

what temps are you seeing for all-core clocks, Tdie/Tctl and Tccd's

like Keith said, do "Power"

 

and I never messed with any memory settings. honestly I never really changed anything except the determinism and cTDP settings.

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6453
Credit: 9580039840
RAC: 7392379

Tom M wrote: Ian&Steve C.

Tom M wrote:

Ian&Steve C. wrote:

from what I've read, Determinism = Power/Performance, and the cTDP at it's max should be the best setting.

I set it to "Performance" with cTDP on "auto" and it just started registering 2.7+ MHz.  That might be all I can get out of it.

---edit---

It just regressed to 2.5-2.6 MHz :(

---edit---

The PDF on HPC claims I should also turn off the Virtual Machine and Memory crypto stuff.  What do you think?

While the MHz is "down" (back to 2.5-2.6) theU@H seem to be processing "faster" so lets see if that result holds up.

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!

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3956
Credit: 46958482642
RAC: 64664293

Do you have a way to measure

Do you have a way to measure temperatures? I like the psensors software.

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6453
Credit: 9580039840
RAC: 7392379

Ian&Steve C. wrote: what

Ian&Steve C. wrote:

what temps are you seeing for all-core clocks, Tdie/Tctl and Tccd's

On performance psensor is reporting about 50C for all of them.

Will toggle "Power" while leaving cDTP on "auto" and see if MHz changes. 

I had lockup problems last time I tried cDTP at 170.

The doc I am looking at claims "Performance" causes "less variability" on the nodes.

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!

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3956
Credit: 46958482642
RAC: 64664293

Tom, give this thread a

Tom, give this thread a read.

https://forums.servethehome.com/index.php?threads/amd-epyc-7401p-centos-7-4-benchmarks-cpu-soft-lockups.18831/#post-182764

this person seems to see the same 2.5-2.6GHz clocks. and even mentiones some soft lockup issues as well.

 

might be just the nature of this CPU that it doesnt reach full clocks under boost in some situations.

 

you might try setting cTDP again, since you have a fresh OS install now. to check if the lockups are actually caused by that setting, or caused by something in your old OS.

 

if it does lockup again with cTDP at 170, try 165.

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6453
Credit: 9580039840
RAC: 7392379

Tom M wrote: Ian&Steve C.

Tom M wrote:

Ian&Steve C. wrote:

what temps are you seeing for all-core clocks, Tdie/Tctl and Tccd's

On performance psensor is reporting about 50C for all of them.

Will toggle "Power" while leaving cDTP on "auto" and see if MHz changes. 

I had lockup problems last time I tried cDTP at 170.

The doc I am looking at claims "Performance" causes "less variability" on the nodes.

"Power" on "Auto" is currently running at 2.7+ MHz with a Psensor of 53C-54C.

I think this is as good as I was getting with a manual setting of 170 and nothing has locked up (yet).

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!

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3956
Credit: 46958482642
RAC: 64664293

Tom M wrote: Tom M

Tom M wrote:

Tom M wrote:

Ian&Steve C. wrote:

what temps are you seeing for all-core clocks, Tdie/Tctl and Tccd's

On performance psensor is reporting about 50C for all of them.

Will toggle "Power" while leaving cDTP on "auto" and see if MHz changes. 

I had lockup problems last time I tried cDTP at 170.

The doc I am looking at claims "Performance" causes "less variability" on the nodes.

"Power" on "Auto" is currently running at 2.7+ MHz with a Psensor of 53C-54C.

I think this is as good as I was getting with a manual setting of 170 and nothing has locked up (yet).

Tom M

yeah if you can hold 2.7+, i wouldnt worry about it anymore. 2.8 is the max you'll ever see and something like 2.75 wont have a perceptible difference in performance in my opinion. not worth the frustration trying to squeeze out that extra 50MHz

_________________________________________________________________________

Exard3k
Exard3k
Joined: 25 Jul 21
Posts: 66
Credit: 56155179
RAC: 0

I sadly had to disable the

I sadly had to disable the O2MD CPU workunits because they need so much RAM (welcome to scientific computing!). But as soon as my new Threadripper system is shipping (128GB ECC), I want to get back to those CPU tasks...like 40-50 at a time. Until then FGRP and hunting for pulsars is my computers job.

 

>2GB RAM per logical core is very high considering the increase in core counts on desktop PCs in recent times. I don't see today's standard PCs with >1GB per logical core in the vast majority of systems. I doubt the majority of contributing PCs can handle them very good.

 

I know Folding@home has bigger workunits for larger systems. Is this something Einstein@home will do at some point? is this useful? I certainly can't utilize my future 64 threads all the time, so much of CPU is open for E@H usage.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.