12900kf, typical voltages and temps for various ratios?

[markm75]

Yes adaptive offset at +0.015(~1.235) , 5.3 takes +0.09, and 5.4 takes +0.16 (~1.38v) running at XMP timings at 6200Mhz since touching anything causes instability on this sammy kit. Biggest latency decrease seems to be upping the TREFI but the kit is impossible to stabilize. The other timings make almost no difference. I think the raptor lake IMCs are going to be better with this. SA volts at 1.2v

Ram is the GSkill 6000 @ 36-36-36-95 @1.35v, Samsung ICs.

Ah interesting, i hadnt considered trying the + offset mode instead of negative.

I had thought maybe i should drop the ram to 6200mhz too. Are you setting sa to 1.2 or is that what it reports, is that cpu sa i assume?
I need to see which IC i have in my gskill.. i know they are rated at 1.40 out of the box.

So far i still have no memory errors on the memory test, maybe i'd be better off with y cruncher instead of this 9 hour memory test. Still puzzled that p95 blend 10 hour was fine and msfs reported that error unless it really wasnt a ram issue but maybe paging or other error.

 

[phanbuey]

Ah interesting, i hadnt considered trying the + offset mode instead of negative.

I had thought maybe i should drop the ram to 6200mhz too. Are you setting sa to 1.2 or is that what it reports, is that cpu sa i assume?
I need to see which IC i have in my gskill.. i know they are rated at 1.40 out of the box.

So far i still have no memory errors on the memory test, maybe i'd be better off with y cruncher instead of this 9 hour memory test. Still puzzled that p95 blend 10 hour was fine and msfs reported that error unless it really wasnt a ram issue but maybe paging or other error.

The current stress test tools do not test DDR5 stability well. It doesn't fail consistently and it takes exponentially longer to figure out if it's stable. Also, the current GSkill kits are already pretty topped out so an additional few hundred MHz is probably the best case, tons of people smashing their heads into a wall with those online just to end up right back at XMP.

Some guy posted a 20 hour stability test then BSOD when doing random desktop stuff.

 

[ir_cow]

@markm75 listen to what @phanbuey is saying. Or really anyone here trying to help....

 

[markm75]

i gave 1.23~ + 0.015 a shot, set to 52x and 39x for each, set avx to 0 to start..
i started with the 4096w power limit cooler setting.. instant 100c during small max power fft, and throttled badly.

I tried changing to 241w power limit.. it was 96 at first then 100c not long after, but still badly throttled, around 4.5ghz or less i think it was.

So i tried the same test without avx, avg ratio was still only 49x or less
The load was only averaging 1.29, sometimes 1.24 with avx off, i think maybe it was 1.20ish with avx on.

I think something else is amiss here, still working through it, but also that unify x board comes tomorrow.

 

[ir_cow]

I think something else is amiss here

yeah you! lol.

101 of trouble shooting, start with what is working before playing with voltages and overclocking.

 

[phanbuey]

Try load defaults and see what it's doing... what's the default voltage? Does it shoot to 100 at stock? What is the HWload wattage it's using?

 

[markm75]

im on the unify-x at this point.. did defaults like i did before, similar results, p95 with avx of course hits 100c, power max is like 232 watts. (with 4096 limits, its slightly better with the 241watt limit)


Im not seeing anything in hwinfo that says "yes" to throttling. per say though now.. i've changed some settings to manual and had better success, though cinebench is at 25900 or so, i guess due to throttling..

Currently im trying 1.295 adaptive and -0.04 or so for the offset. (5.2 p core, 3.9 e core, 4.1 ring)

note that when cinebench runs (avx 0), it cyles from 4.7 to 5.2 ghz randomly.
If i use intel avx stress test, its often 4.9ghz, sometimes 5.2, and notes "current throttled"
I tried setting the max power limit to 4096 and increased the volts some too, same thing keeps happening.

Overall this board feels much much more stable, im in windows 11 for reference, i've tried the high performance power option

Thermal velocity boost is set to disabled at this point

Edit changing power limits obviously is making the cores ratio stay up but I'm crashing pretty quickly even at a fixed voltage 1.33 and 5.2

Conclusion this chip is terrible
(setting to auto it has a max of 1.385 (sometimes 1.4ish), then p95 small ffts 100c near instant, max power, im at 1.20v under load and it doesnt crash if its on this auto setting and auto llc but the frequency goes to 4.7 (i have avx at -2 currently from 5.2))

on the auto volts cinebench will avg 1.30 and hit 100c at times but the frequency stays up higher, say 4.9ish on avg, avg temp 86c, score (1 pass) is lowly 26800 or so.

Seems i also have something odd with win 11, my scores R23 seem to avg 26k at 5.1

I booted to a clean w10 and it was 28700

I noticed Temps were higher in win 11 too maybe even voltage slightly

1.26 and llc8 cinench was bsod immediate

 

[ir_cow]

You might have a absolutely garbage binned CPU. Cinebench R23 for me on stock is 1.2v and scores 27850. Some people report 1.15v at stock.

 

[markm75]

its real strange, i've lowered the adaptive offset to 1.25 and llc4, with -0.025 (from 1.275 and 0.015 llc5) and im still pushing 1.28v in cinebench at 5.1ghz (228 watts max and 5000 to 5100mhz), decent scores though, but temps at 1.275 were 94 to 96c max and now at 1.25 more like 97c. (PL max power at 4096)

P95 max power fft is about 1.181 and 255 watt and 4700-4800mhz with avx at least at first. package 98-100c though

I would have thought this offset and base voltage id be closer to 1.2 but its not ending up that way so far.
(this on my test now win 11 clean partition, so the previous score issue and heat isnt from win11 per say but the OS itself, both clean win 11 and win 10 perform better)

EDIT:

tried 1.22v and 0.040 neg. offset adaptive, i got it to 1.23v under load with cine, but now the temps went up to 100c (27,903 score now), tried 1.22 and -0.015 about the same.. i would have thought bringing the volts from 1.27 to 1.23 it would have gone down, or things just heated up

went back to the 1.275 and -0.015, 100c in cinebench, i guess def. just hot now, the volts were down to 1.23 strangely, not 1.27v under load as before

 

[phanbuey]

temps at 1.275 94 to 96 seem ok actually. I think the chip maxes at 110C...

Can you set TVB ratios to downclock? Pick 5.2/5.3 @ 1.33v or whatever it needs, 5.0ghz @ 1.24 @ 90C from there and cap at 241W for powervirus prime small fft.

I think the ideal settings here is for the chip to run high clocks normally and then settle down when you want to crank some cinebench/prime since the auto voltage on that bin is pretty high.

You can also just cap it at 95C on max temp and call it a day, let it bounce off throttle @ 5.2 for cine and prime... always an option. Normal usage will be like 75-85 tops.

 

[markm75]

temps at 1.275 94 to 96 seem ok actually. I think the chip maxes at 110C...

Can you set TVB ratios to downclock? Pick 5.2/5.3 @ 1.33v or whatever it needs, 5.0ghz @ 1.24 @ 90C from there and cap at 241W for powervirus prime small fft.

I think the ideal settings here is for the chip to run high clocks normally and then settle down when you want to crank some cinebench/prime since the auto voltage on that bin is pretty high.

well i have all core at 5.1ghz as of now, i guess i could use tvb ratios.. i have all the tvb related bios settings to disabled as of now though.

 

[phanbuey]

TVB works well to stave off throttling ~ it can take the edge off the heat in the top end by just dropping to a 4.9ghz with 1.15-1.19v past 93C or something, which will effectively just keep the chip there under extreme load.

That also let's you pump more volts and frequency for normal usage, since you can normally run 5.2 just fine, it's only in like cinebench and prime that it will be a heatmonster, so just run a different tuning past 93-95C.

 

[markm75]

Ill try turning tvb back on (its default state anyway)

Just tried odyssey and the max package temp was 92C, before it was 80c on the old board with the same cpu but probably less volts for some reason then.

Well at least for the odyssey test, it was still at 91C and it locked up (the game, not windows). I think it was pulling around 1.31v load during the game (avx 0 setting, an avx game) edit: -2 avx was about the same, 94c temp.

edit: loading defaults into the bios, so auto cores, auto volts, odyssey maxed at 87c (this 87 to 94c is now with the case side on for more realistic load test)

Now I'm battling getting the ram to 6400 even on the new board

Before those tests were at 4800, 6400 and even 1.47v, the ramtest fails 30 to 80 min in.

I've tried upping voltages little by little (vcore)

Anyone know the max safe ram voltage? I wabt to say I could try 1.5 on both settings

 

[markm75]

my ram tests continue to fail, even at 4800mhz, using ramtest.exe

and i set the bios to 1.275v adaptive offset neg 0.030, yet the bios holds around 1.39 even during things such as memory tests, i find this odd, but maybe its not.

 

[ir_cow]

Ram shouldn't be giving you errors at 4800 for any normal reason. Have you been fiddling with the voltages and memory timings?

 

[markm75]

Ram shouldn't be giving you errors at 4800 for any normal reason. Have you been fiddling with the voltages and memory timings?

No it's xmp, I had ram at first at 1.45v and it also gave errors though.

I may put ram volts n vddq back to auto n see if the test fails or the ramtest tool isn't accurate

 

[ir_cow]

You don't want to be playing with DDR5 voltages unless you can identify what the memory ICs are used. you have micron, Samsung and SK HYNIX as the big three. Some memory doesn't respond well to voltage, others could be damage if its to high. I would be careful.

 

[coradaelu]

No it's xmp, I had ram at first at 1.45v and it also gave errors though.

I may put ram volts n vddq back to auto n see if the test fails or the ramtest tool isn't accurate

Were you able to solve the heating problem? I have the same problem with the 12900kf and msi a pro wifi DDR4 version