• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

One compute unit per module

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Infinite66

Member
Joined
Jun 27, 2012
I had a few questions regarding this function in my BIOS. Would making my fx9370 a "quad core" make the voltage amount for overclocking less? I'm at 1.325v now for all 8 cores to be at 4.4 ghz. Would this same voltage let me overclock higher with only 4 compute unit as opposed to 8? I ask because I've seen on most games only a max of 3-4 cores being utilized. And as we know with AMD cpus, single core performance isn't amazing so the higher clock the better. I've been using coretemp to monitor core load usage. Unless I'm missing something, this seems like it would work for the better. Help from the AMD gurus Is appreciated
 
With intel, disabling HT can allow one to use lower voltage... You can give it a try on AMD, Im not sure personally.

People commonly did this to AMD CPUs as they would run faster... there is a ton of info on this on the web (as far as people disabling and why).
 
With intel, disabling HT can allow one to use lower voltage... You can give it a try on AMD, Im not sure personally.

People commonly did this to AMD CPUs as they would run faster... there is a ton of info on this on the web (as far as people disabling and why).

Yeah I googled it. It seems with an fx 8350, it gets close to 20 fps more I metro last light benchmark with 4 cores disabled, clocked .6 ghz more than all 8 cores enabled. There is of course the decrease in performance when running heavily multi threaded applications. If I have time this weekend, I'll play around with it. I can get 5.0ghz on all 8 cores at 1.55v but it creeps to 1.58 to stay stable. I might be able to get the same speed on 4 cores with less voltage. Maybe more. Who knows.
 
For AMD, Yes it is fact .......benching anyhow. Less cores = less heat, which allows you more headroom on voltage for greater OC........or less voltage at the same clockrate. I do this all the time pending the bench I'm running.
 
I googled a bit as well, and it seems that allowing only one core per module gives a higher per core performance.

Could you run Cinebench 1 thread with 8 cores and then with only one core per module and report the results, please?

That could be interesting, as some guys on some forum said that the perf/core raise like 15%@same clock.

Edit: replace "core" by "compute unit" in all of the above please... getting tired... lol!
 
Last edited:
I googled a bit as well, and it seems that allowing only one core per module gives a higher per core performance.

Could you run Cinebench 1 thread with 8 cores and then with only one core per module and report the results, please?

That could be interesting, as some guys on some forum said that the perf/core raise like 15%@same clock.

I'll take all this into consideration when I have time this weekend
 
Could you run Cinebench 1 thread with 8 cores and then with only one core per module and report the results, please?
How would one shut off the individual compute modules though? I know with my Crosshair V formula you have the option to disable "Cores" when doing so you are shutting down a complete module. The 4 modules are 1 and 2, 3 and 4, 5 and 6, 7 and 8 how could you shut down say 1 and not 2 or 3 and not 4?
 
My motherboard (asrock 990fx ext9) has the option for 1 compute unit per module. I.e. turns off 1 "core" per module allowing the full use of the L2 cache allocated to 1 core instead of shared. Ok so here are some images. I got impatient and did it tonight instead of waiting to the weekend.

This is 8 threads 4.4ghz 1.35v
8 thread single core perf.PNG

This is 4 threads 4.4ghz 1.35v
4 threads single core pef..PNG

This is 4 threads 5.0ghz 1.45v (I did some quick and dirty overclocking. I could probably get the same speed for less juice. Lastly, even with 4 cores off, CPU will not got beyond 5.2ghz even with 1.55v. Decided 5.0 at 1.45v is better.
4 thread 5 ghz single core perf..PNG

This is Heaven Extreme preset with only 4xAA 8 threads. (i forgot this run had an overclocked GPU profile. 1150/1600)
heaven 8 threads 4AA.png

This is Heaven again same as before with 4 threads, 5.0ghz, stock GPU clock 1040/1350. Results are very good still.
heaven 4 threads 5ghz.PNG

My CPU used to get up to 64C at 5ghz for 8 threads under water on a 360mm UT60 rad with 6 2200rpm fans at full tilt.
With 4 threads, 2 of the same fans, but at a low RPM, and on an air heatsink (frio extreme), it reached a max of 57C. Thats with all fans in the case at less than 1k rpm.
 
with fx you are better of turning off what you are not using, less heat, less wattage and many times less vcore.
I have seen on boards, with weak power sections large decreases in vcore.
 
Thanks a lot for all the info Infinite.

It seems that the guy was wrong regarding the wrong regarding the perf increase per core with one comput eunit per module.
 
Well...consider that Heaven is a benchmark that is heavily GPU dependent so I do not expect that to show any difference and is a bad test for this

Its seems to me that Cinebench isn't using the other compute cores as there was no change at all from '8' to 4. I would imagine it is FPU heavy which there are only 4 cores for that in the first place so disabling the half of the interger cores wouldn't matter.

This change was really meant for gaming IIRC and should be tested on game benchmarks to show a difference I would imagine.

EDIT: Here is a great read (this has been mentioned and tested in years past): http://www.overclock.net/t/1493092/fx-8320-disabling-cores-increased-single-threaded-performance
 
Last edited:
When I gamed on my 8350 , I had one title ( Prototype II) that would give weird issues with more than 4 cores running. So I shut 4 off, cranked it to 5.0 and gamed like that with most everything. It does help in most of the titles I played. 8 cores just aren't necessary in for most gaming.
 
You can select the CPU tests only.

And there is a one month free trial. Which means you can use it every time you build a new rig or do a new install ;)
 
I stay away from it... its not the WEI, but, it doesn't hold much merit for a lot of people so I jumped on the bandwagon, LOL!
 
Back