I installed this and it took my Indigo benchmark from 2.0 to 3.6 (2990wx) and an 8 for the car. And Cinebench is 5500. Awesome!

Thanks for the tool. Now I just need to save up a few months so I can afford a 2990wx :-p
This is starting to be spread around, popped up in a few facebook posts. Everyone is starting to question existing benchmarks. Could be that Threadripper was under-performing and it took a while before anyone noticed.

I suppose it goes without saying, but It is hard to recommend the 2990wx. It is more an expensive novelty that few people have a good use for (and would choose it over an EPYC). The 2950x is what I'd go with. it has better per-thread performance, no performance quirks, and is half the cost.

So... What about CPUs that *don't* have more cores than memory channels; Would Coreprio have any benefit there?

Also, I'd look into experimenting with OpenACC and OpenMP compilers. There's a lot of untapped potential in the realm of GPU computing which has been opened up by those utilities.

There are possibly additional use cases. This will be more true when process-specific "core prioritization" is added. These "soft" process CPU affinities are less restrictive than normal static process CPU affinities. For instance, a user may want to prioritize the fastest core on a die.

Is there going to be update or anything? is it still useful for the Intel 12th series with Efficency cores?

I'm still using a 2990wx. I've got Win 10 Pro x64, v. 21H1, build 19043.1023. Since the release of Coreprio, AMD has launched the 3xxx series, all the way up to their most recent 3995wx, which still uses NUMA. I'm wondering if AMD or Microsoft have done any work on the issue at all or if Coreprio is still necessary. What's the current state of affairs with the scheduling problem?

I tried the software and it really speed up my 2990wx .It was rendering the scene 2:07sec and now 1:25sec. Thank you for the software

Will the Dynamic Mode or NUMA discc work for TR4 1950x ?? which one should I use ? Only dynamic or only Numa Discc ? Also , do I need to use Distributed instead Local in the bios ?

At this time there is no known performance gain by use of either of the two current Coreprio features on the 1950x. It doesn't suffer the same complexities of the 4 die processors. I can advise use of Process Lasso though!

We run a game server (ARK) and noticed a similar issue that is fixed by doing the processor affinity in task manager. We have a TR 1950X. With our game server - ARK - what would happen prior to doing the affinity fix is that there would be periodic world freezes lasting about 200ms - long enough to cause lag. I think the problem that you fixed isn't just for high end 2990WX processors. It think that it is across any large NUMA node set processors such as TR.

Holding off until it is integrated with Process Lasso, but I am really looking forward to seeing the results on my 2990wx!! Thanks for the hard work, gents!

Will this functionality eventually be integrated into Process Lasso? If not already there, it would be nice to run one program to fix all these issues.

Yes, it will. It may be a week or two, but expect it!

Awesome. I had bought it and was wondering if it will be added or if both programs can run concurrently without affecting each other (I'm on a 1950X and like Jefferson said, I also have that stuttering randomly. It can happen to Firefox while using other browsers or windows in my other screen, or also games)

Until it is added (soon), I should note that you can run both programs together without complications. Coreprio ignores processes with a non-default CPU affinity mask, so any customized CPU affinities set with Process Lasso will still work fine.

speaking of fixes..
Your fix is going to be huge if AMD's rumor of
having 16 cores in AM4 socket and 32 cores by default on threadripper
and beyond (up to 128)?.
Or would the Chiplet design make this fix obsolete?

Since they are retaining socket compatibility (AM4 and TR4), DLM and NUMA Dissociater will presumably be very important - depending on what Microsoft fixes I suppose.

Could this also benefit 2950x ?

Unclear. Presumably not since it only has 2 NUMA nodes and the problem seems most apparent on systems with >2 NUMA nodes. Give the Indigo benchmark a try /w and w/o the fix applied (be sure to restart the process between tests) to confirm.

This is on a 2950x?

Could this also benefit amd fx-8350 users?

Presumably not. The Windows CPU scheduler was tuned to properly support Bulldozer / Piledriver CPU modules (so that paired cores are not treated as physical cores) and no further deficiency is known to exist.

Process lasso works for intel ? toi be honest im trying to try it but cant see any big change, i really wanna know how it works for i9 11900k the core for gameing and better fps

What's the refresh rate setting does? It takes a lot of CPU Usage when it's 200 Default. can I change it to higher number?

Would process Lasso be enough to automatically tell it to use NODE0 all the time? because CorePrio works perfect, I never had lagging or FPS issues when I started using CorePrio, would Process Lasso be enough? How can I set it to do the same thing as CorePrio for 1950x?

I just don't want manually change profiles for every game or thing. CorePrio does it automatically, the problem is it uses CPU USAGE with the Refresh HZ.

Could you please explain everything I asked? it's really important.

The refresh rate is how often the threads are evaluated to rebalanced. Too high a value will cause a delay in migrating threads to optimal CPU cores, and too low a value will increase CPU use. Certainly you can safely increase it beyond 200ms.

Process Lasso's doesn't presently offer 'soft' CPU affinities like Coreprio, because there has been very little (if any) use case outside the 2990wx. However, you can of course use persistent CPU affinity rules with Process Lasso to keep processes on Node0.

if I want to enable HT or amd SMT. which cores do I need to select in the Node0 and node1 and which I need to prioritized in the thread priority? so it will still use only node0 with the cores and smt from node0. also if I just want to have smt enable but only use it when it's needed. process lasso didn't work well, no matter what I did I got lower fps when smt enabled in bios and disabled in windows

Because then I have 31 instead 16 Cores to choose from. so which ones should I select if I don't want SMT but only for the programs I choose specifically?

Select every other core. It doesn't matter which you start with, so long as you get one from each physical CPU core.

Hi !

With Indigo Benchmark (2990WX) I can't see any difference with core prio on or off.
On the bedroom I score around 4.4... Is windows 21H1 nailed it ?
I came here because, recently I got many CPU crashes with my 3d apps (black screen with EZ CPU led red)... Solved it by disabling the global c-state in Bios. Something changed but I don't know what ;-)

Excellent program, it also has a positive (and similar to what is described above) on my dual Xeon E5 system, for instance when using multiple projects in BOINC. A suggestion: replace the "refresh rate" label with "refresh interval (ms)". Before reading here that it is in "ms" I thought it represented a frequency of update (in Hz).

Great Job

this site was blocked in www.facebook.com whats dangerous on this site.

Can you post a screenshot of that, or email to support@bitsum.com ?

If I had to guess, it is taking issue with one or more downloads on this site - whether a false positive or simply 'uncommon download'.

Thanks!

do i need Coreprio for ryzen 3900x

No. Try Process Lasso

whats the change log in version 0.0.6.8 ?

"+ Changed 'software thread count' caption to 'threads to prioritize', for clarity"

I found some users were inappropriately adjusting this field to match the hw thread count of the CPU. I attempted to clarify the caption.

https://community.bitsum.co...

Is this program currently helpful with Windows 10 August 2019
latest updates?

The NUMA Dissociater is still applicable. Our configurable DLM can also still be helpful, though is less necessary than before.

I have a Threadripper 2990wx 32 core. After some testing this are the results:
1- When "Bdlm" and "Numa dissociater" are selected I can only see half of the threads working at full throttle with either vray, unreal engine light bake, unity lighbake and corona render.
2-When only "Numa dissociater" is selected, I can see all processors working at 100% - 3.4 ghz speed, but somehow the rendering times for all testings take longer and it becomes impossible to do any multitasking.
??

For BDLM: That doesn't sound right. Have you changed the 'thread count' field by chance?

On NUMA Dissociater, it shouldn't ever do any harm. Are you certain of those results? There can be some 'natural' variance in performance that makes it difficult to draw conclusions without averaging several runs.

p.s. Process Lasso's ProBalance can do a lot to help with the ability to multi-task during heavy CPU loads.

Sorry for the late reply.. Yes I've changed the software thread count from 32 to 64. Don't know if that's the cause here...

Thanks for the Help

Ok.... with Pro Balance enabled I just saw and increase from 65% CPU use to 90% but goes up and down from 54% to 90%... is not constant. With my old intel i7 it was 100% from start to finish while rendering...

Most likely your rendering software (or other load) is not parallelized enough to take full advantage of all 64 CPU cores, hence you are seeing the excess capacity. Not every operation can be split up to that degree.

Ok I get that, but this is Vray for 3ds max... I also have an HP proliant with 40 xeon cores and that same program takes all those 80 threads to the fullest....
I've also tested coreprio with BDLM not enabled while NUMA disassociater enabled and get the almost the same rendering times with a 5 sec difference in favor of this second test on a 9 minute total time.

Maybe uninstall it then. Not every load will matter.

The thread count is the number of software (not hardware) threads that should be prioritized. Setting it to 64 causes it to try to cram up to 64 threads onto the first 32 logical cores, so that explains exactly what you saw in #1.

I just downloaded and test Pro Balance and CPU Eater. With the eater I can finally see all cores working at 100%. But then I reset the PC tried all over again and the results are the same. Only half of the cores are working when rendering, or lightbaking or whatever.

I've just made some testing with the thread count to 32 and it's still the same...