AMD RED BIOS EDITOR und MorePowerTool - BIOS-Einträge anpassen, optimieren und noch stabiler übertakten | Navi unlimited

[RX480]

but UL calculates the average clocks across all three tests in TS, so their reported average GFX or VRAM freq. are of little use.

No,
in my understanding the avg. clock is GT1+2 and loadtime between GT1+2 and may be loadtime GT1 but not during CPU-Test
therefore a high minClock can have a higher avg.Clock

btw. today
"impossible" spike reading at the end of GT2
(and no, the driver didn´t crash)

 

[PJVol]

No,
in my understanding the avg. clock is GT1+2 and loadtime between GT1+2 and may be loadtime GT1 but not during CPU-Test
therefore a high minClock can have a higher avg.Clock

I was wrong about CPU test, just like you were wrong about the test loading times, btw )
Here's what I found while digging through Systeminfo telemetry data. For testing purposes I ran TS with Fmax 2400mhz:
(you can find all data in monitoring.csv or rawmonitoringdata.json in your *.3dmark-result file, which is a regular zip)

1. FPS polling rate is 1/25 s
2. All other metrics, such as clocks, temps, etc, are sampled at 1s rate
3. For monitoring they use a) GPU-z for gpu, b) CPUID for other hardware, c) their own tool (BM) for FPS
4. GPU core and memory clocks are averaged only over data sampled during actual workloads, gt1 and gt2, i.e. the test loading intervals are discarded

 

[PJVol]

btw. today
"impossible" spike reading at the end of GT2
(and no, the driver didn´t crash)

What exact gpu settings, including mpt/mct/driver or whatever, do you use to get the core clock to reach the set Fmax value?

 

[RedF]

I was wrong about CPU test, just like you were wrong about the test loading times, btw )
Here's what I found while digging through Systeminfo telemetry data. For testing purposes I ran TS with Fmax 2400mhz:
(you can find all data in monitoring.csv or rawmonitoringdata.json in your *.3dmark-result file, which is a regular zip)

1. FPS polling rate is 1/25 s
2. All other metrics, such as clocks, temps, etc, are sampled at 1s rate
3. For monitoring they use a) GPU-z for gpu, b) CPUID for other hardware, c) their own tool (BM) for FPS
4. GPU core and memory clocks are averaged only over data sampled during actual workloads, i.e. the test loading intervals are discarded

Oh, the info might help me to find out why for me and many others the sysifo fails, and runs are not validated. I was told that Nvidia cards did not have such problems.

 

[RX480]

What exact gpu settings, including mpt/mct/driver or whatever, do you use to get the core clock to reach the set Fmax value?

the same run: ... mostly in the Powerlimit, ... only some Boosts
the normal max clock with my volts should be at the end of GT1 = lower then maxClock in the driverpaneel
(AFMF2-driver)

edit:
as check the Kombustor in attachment 2 (not power limited)

 

[PJVol]

the same run: ... mostly in the Powerlimit, ... only some Boosts
the normal max clock with my volts should be at the end of GT1 = lower then maxClock in the driverpaneel
(AFMF2-driver)

I don't get, are you saying that the only thing you change in MPT is Voltage Vmin/max for core and soc and all other settings were left intact?
What is PL=-1? Isn't msi 6800 trio default PL 215W?

 

[RX480]

What is PL=-1?

PowerLimit in this Example
218W - 1% = ca. 215,82W tgp

I like to use the powerslider in the driverpaneel if I want to jump between some settings.
218*1,15 would be 250, if required, ... may be useful in the TSE.

 

[PJVol]

the normal max clock with my volts should be at the end of GT1 = lower then maxClock in the driverpaneel

Yes, I see. That's basically the same behavior as in my case. Here is an example of "Superposition 1080p Extreme" run. The GFX clock hit the Fmax-50 wall no matter what settings.



I think I saw someone manage to get the GFX clock to Fmax without hardware PLL or SVI2 trickery, but I'm not sure now if that's really the case.
If this is true (i.e. GFX clock can sustain in 2790-2800mhz range), it'd be nice if someone could confirm this providing some evidence, like screenshots or something, no matter the type of workload used.
(rather than just saying "I know a person who heard that this is true" )