SOLVED AMD FX - Throttling (not temp related)

[salsoul]

Ive got some intersting findings about the FX throttling. its nothing to do with temperature it seems to be task specific. I went back to stock setting and ran small FFTs with prime 95.. immediatly within 2 seconds of starting prime, the cores were throttling. all temps were normal 33core 42socket.. I stopped benching and changed to blend. a few minutes in its starts throttling again.. temps 40core 48socket.

I swiched to cinebench. some throttling but not much. again temps normal. remeber this is at stock. it was underclocking to 3.3ghz 16.5 multi. nothing to do with heat.

From my findings ( and I am by no means an expert in this field and this is all speculation)

tasks with High FPU calculations get throttled by the cpu. I dont know why ? possibly because of the lower FPU count. Im hoping an expert can verify what im saying. here is a screenshot of what im on about.. I had been running prime95 for 15mins small ffts. the setting are slightly above stock multi 21.0 200FSB = 4200Mhz speed. voltage 1.3875v LLC full calibration.

As you can see all temps are normal but throttling is occuring ? so it cannot be temp related.

I also ran cinebench at 4.62ghz with just a multi overclock. the results were

Pass 1 with 4.6ghz via Multiplyer : 23.0 results = 7.05 points @4.62Ghz.
---------------------------------------------------------------------

For the next test I OC via FSB to 4.62 set fsb to 279 x16.5 = 4.62Ghz

Pass 2 with 4.6Ghz via FSB OC 279x16.5 results = 7.58 points @ 4.62Ghz

Even when the temps are ok the FX processor is throttling on specific tasks. which is also holding back performance on cinebench as you can see from below when compared to multi OC and FSB OC.



So its throttling seems to be task specific and the only way around it is to OC via FSB set 16.5 multi and off you go. but be warned this also increases heat from the cpu as its not throttling.

Can anyone else verify my findings ?

If you would like to - run prime small fft's. post a screenshot to see if this is affecting everyone

even @ stock.

here are the screenshots I was on about.

 

[Woomack]

Try to turn off APM in BIOS (I'm not sure if you have it on your board ). That throttling is because you are passing some max cpu power that is ~125W and then APM is lowering core speed.

 

[salsoul]

Try to turn off APM in BIOS (I'm not sure if you have it on your board ). That throttling is because you are passing some max cpu power that is ~125W and then APM is lowering core speed.

If that were true why is it not happening when I OC via FSB >? same power output same wattage ??
APM/EUP/NB-OCP/CPU-OCP/C6/C1 are all disabled.

 

[Woomack]

Did you try to use AOD ? Simply turn on and then turn off turbo core and check how it's working. Maybe APM isn't working in your bios, hard to say. I just have no other idea what can it be as it doesn't fit to anything else.

 

[salsoul]

Did you try to use AOD ? Simply turn on and then turn off turbo core and check how it's working. Maybe APM isn't working in your bios, hard to say. I just have no other idea what can it be as it doesn't fit to anything else.

I am equally perplexed as you are over this. there is no reason for it.

I installed AOD and disabled APM via that as well as APM in the BIOS.

If APM was on then it would still throttle when I OC via FSB but it does not throttle at all.

 

[RGone]

You are the not the only one to see it by any means. There are threads in other forums back last year. Most do not see it unless they bench at various speeds and steps of speed. I suggested OpenHardwareMonitor to user Caddi Daddi in the forum here and he caught it with the individual core speeds similar to CoreTemp.

The only workaround so far is turning off APM with AOD and it must be done at every boot so I hear. Others have set the multi to 16.5 and of course then you have to up HT Ref speed to get more cpu speed.

I have not been able to find a decent white paper to see if there is an answer by what the cpu internal management is able to do.

Some have even found that some of the 6core thubans will do it.

Straightforward answer? Noone has got one to date that I have been able to uncover.

 

[RGone]

Wish I could find a 'white-paper' with less color and more black and white. You know that text stuff.


See Page #20 thru #23. Sounds like too many what ifs. Too much "dynamic" stuff.

Sounds like a lot of laws that are passed; which have UN-intended consequences.

 

[salsoul]

You are the not the only one to see it by any means. There are threads in other forums back last year. Most do not see it unless they bench at various speeds and steps of speed. I suggested OpenHardwareMonitor to user Caddi Daddi in the forum here and he caught it with the individual core speeds similar to CoreTemp.

The only workaround so far is turning off APM with AOD and it must be done at every boot so I hear. Others have set the multi to 16.5 and of course then you have to up HT Ref speed to get more cpu speed.

I have not been able to find a decent white paper to see if there is an answer by what the cpu internal management is able to do.

Some have even found that some of the 6core thubans will do it.

Straightforward answer? Noone has got one to date that I have been able to uncover.

Im glad im not the only one.. it does from my findings seem be very task specific. which would further allow me to draw the conclusion that this is hardcoded into the CPU.

Allow me to elaborate. If it was APM then it would throttle full stop regardless of OC technique. once wattage/votalge/heat is upto a certain point I would expect it to throttle regardless.

the fact that it throttles back to 16.5 and no lower.
the fact that if you set your multi to 16.5 and Oc there is no throttling at all. even when heat is the factor. Ive had my core temp as close to 62c without going over and socket temp was 75c NO Throttling at all.

this must be hardcoded(Not sure if that is the correct phrase) into the CPU.

its not software / it cant be APM as it would kick in when tjmax is hit.

it throttles @ stock settings with small FFT's in Prime95

The only thing left is a FPU bottleneck. the cpu cant handle 8 threads all calculating Floating Points. so it throttles modules down in order to get the data out. ( pure speculation). I dont have any proff, I am no where near intelligent enough to speculate further or prove/disporve my ow theory.

But something is not quite right..

Im not suffering performance drop outs or BSOD/freezeups. I am mearly curious as to why an unlocked CPU decides to take control away from the end user without explanation and throttle itself without just cause.

Not much point in having an UNLOCKED CPU if it gets too hairy the CPU locks itself to 16.5

 

[RGone]

I posted again just before you did and it is from a white-paper with too much color.

 

[salsoul]

I posted again just before you did and it is from a white-paper with too much color.

Thank you, still none the wiser..

 

[Dolk]

Didn't we have this same problem with Deneb in the beginning. I can't remember what the P-state was called that effected Deneb to down clock like crazy.

 

[salsoul]

Didn't we have this same problem with Deneb in the beginning. I can't remember what the P-state was called that effected Deneb to down clock like crazy.

Im not sure I never had a Deneb.My thuban never did this and I threw alot of abuse at it.. If it was a P- State would it not happen when fsb Ocing ?

 

[Dolk]

It was activated regardless of the overclocking method. Its a feature in the CPU itself but can be controlled by software just like in this case. A BIOS update will probably come out to help with this problem, but thats if people try to push for it.

 

[salsoul]

It was activated regardless of the overclocking method. Its a feature in the CPU itself but can be controlled by software just like in this case. A BIOS update will probably come out to help with this problem, but thats if people try to push for it.

If it was activated regardless of the overclocking method I dont think it can apply to this, I appear to not get throttled when setting the multi to 16.5 and Oc via FSB, no matter how hot it gets.

 

[Dolk]

Sounds like a Bios bug rather than something with the CPU.

 

[salsoul]

Sounds like a Bios bug rather than something with the CPU.

Definitely can't rule that one out, the BIOS is still in beta 3027 with no sign of coming out of it either.

 

[salsoul]

just to add a note. im begining to think the BIOS on the CHIVE is at fault here.

I remeber reading a while back about HT & NB on the thubans.

the golden rule was the HT link could not be higher than the NB as it would clock down. equally you could not have the NB lower than the HT as it would clock up.

Now with the BD, even AMD's own tweak guide says that the NB (default) is 2200 and the HT (default) is 2600.

When I set this up in the BIOS and boot into windows and check with AIDA the HT link has clocked down to the NB speed automatically. when I go back into the BIOS its set @ 2200NB 2600HT link. I dont undertand.

can someone clairfy for me what is going on ?

regards

Sal.

 

[caddi daddi]

yes , the be ht MUST equal to or lower than nb, fx/thuban/Deneb.
i just roasted my complete fx8210 system and the only way I could get around the throttling was a clean reformat, and not connect to the internet.
i got to a clean, smooth 5.5ghz, here is the tail of toasted stuff.

I am A noob, I was using aod #404 4.5ghz, no problem, moved to #405 beta, no problem, moved to #405 post beta, throttling appeared.
clean reformat, clocks set via bios, set to 5.5ghz, no throttling, wife said do this, and when i came back 5 mins later, still running and then 5.5ghz/80c...... poof........lights out.


amd has slipped us a mickey in aod #405 post beta. weeeeeeeeell, thats what I think.

RGone is schooling me on motherboards right now and at this point I am leaning intels way. A few things I do know at this point is that it will have an 8 phase vrm with a heatsink with lots lots of fins, 6 ply board with at least 1 oz of copper in 3 plys, metal mosfets, cuz silicone was made for boobs.

 

[salsoul]

yes , the ht MUST equal to or lower than nb, fx/thuban/denab.
i just roasted my complete fx8210 system and the only way I could get around the throttleing was a clean reformat, and not connect to the internet.
i got to a clean, smooth 5.5ghz, here is the tail of toasted stuff.

I am A noob, I was useing aod #404 4.5ghz, no problem, moved to #405 beta, no problem, moved to #405 post beta, throttleing appeared.
clean reformat, clocks set via bios, set to 5.5ghz, no throttleing, wife said do this, and when i came back 5 mins later, still running and then 5.5ghz/80c...... poof........lights out.


amd has slipped us a mickey in aod #405 post beta. weeeeeeeeell, thats what I think.

Oh my word ! Im so sorry to hear about your h/ware

Yes I completely agree about AOD, ive been using it for 2 days to see if the apm trick sorted it out but no avail. Instead AOD~~#4.05 decides all by iteself to set the multi to 39.5 while im already at 4.8... 250x39.5= 9.875Ghz. the screen went black instantly no BSOD, nothing, fans 100% all of them.. there was a moment my heart sank. I tentativly pressed the reset button and hoped for a glimmer of life, fans started and I get the usual asus message about my Oc failing. I go into the BIOS and see the Multi set to 39.5 x 250.... Never again will that trashy software grace my pc. I revo'd its untidy *** and removed all traces. Thankfully no harm that I can see was done.

whats your plan for your BD rig ?

 

[caddi daddi]

I had all ready hurt the poor thing so i new it was trash so i'm not to sad
i had removed aod and still had throttleing till i reformated, if i turnd core boost off most of the time i would kill the issue.