Ah, ok! Well, I’m gonna start doing some experimentation for a while and see if I can find some good settings. [sigh] Gonna have to learn how to do spreadsheets again…
Thank you, tk11 and everyone else in this thread. I’m sure I’ll have more questions soon enough =)
The motherboard is usually what sets the “peak” values available, though the EDC and PPT getting set quite that high sort of surprises me. I’d start at like 250/200/230 and trim it out from there; Prime95 Small FFTs remains a quite good way to get a thermal headroom check; if you can keep that below about 85C at whatever PPT you land at, you’re pretty much all set.
Theoretically, yes, if you let it do anything dumb enough. In practice it doesn’t seem to be terribly likely, as you’ll generally run out of thermal headroom before you hit damaging current.
Ah, thank you. I’m using Cinebench 23 for stress test atm. You think Prime is a better choice?
The rig is on a test bench with a dinky little air cooler, so it’s a bit hard to test the thermals properly atm. I gotta take my main rig apart, or … maybe I have a spare liquid cooler around here somewhere…
Either way, thank you for the advice! I’m testin’ some stuff out atm.
Cinebench is “good” in that it gives you a numeric/comparable result, whereas Prime95 is more stress test than benchmark; using it to set your “worst case” thermals works well. I’ve also had it “pick up” instabilities or single-core errors that Cinebench generally wouldn’t.
The tiny air cooler will probably want a pretty severe reduction in PPT just to keep temperatures reasonable, but it’ll let you “feel out” what each setting does while having a limitation that prevents it from trying to send parts to the shadow realm.
With which processor, thread configuration, ambient temperature, airflow, and cooler? There’s some variation with processor, instruction set, and core performance order but dual tower airflow configs in the 20-25 dB(A) range usually 4-6 cores pull 142 or 162 W PPT at 65-70+ °C ΔT. So it’s rarely difficult to hit a 90 or 95 °C thermal throttle. Temperatures start to drop around the eight core EDC intercept and with the ~120 W EDC allows spread across all dual CCD cores it’s more like 45-50 °C ΔT.
EDC limited is common for all core operation. I haven’t specifically tried PBOing EDC to see if there’s benefit for pushing SIMDy all core closer to PPT. FWIW, so far my experience going off processor default or eco modes is it’s an awful lot of fussing around trying to find something stable for next to zero return in workload behavior.
And Prime95’s more app than stress test. 
So is y-cruncher, though I find its bench side a useful complement to timing machine learning workloads and our in house SIMD code.
Yeah, I had already tuned PPT and curve optimizer and then lowered EDC but ran into stability issues with no apparent benefit so I didn’t feel compelled to bother any further with TDC or EDC because It didn’t look like there was really anything to be gained over using the defaults.
That’s a lot of current for not very many cores. Can’t say it surprises me to see that workloads are possible where that few cores can pull that on Zen 4/5, but the temperatures at that point are almost false. I say “almost false” as if they aren’t happening (they are); but the limiting factor isn’t really the cooler’s ability to ditch heat at that point, and it instead moves to getting the heat to the cooler. I’ve found that somewhere around 15W/core is where things start to get questionable in that regard (even back to Zen2), to where a twin-CCD chip could quite happily eat twice the power of a single-CCD equivalent under the same cooler.
Yeah, I’d agree with that, give or take some variation between processors and workloads. The 9900X I’ve got boosts to 16.2-16.4 W/core at stock and doesn’t necessarily go all the way to ~75 C ΔT, which IMO is pretty decent.
Railing up against 140 A EDC is consistent across the couple AM4 motherboards I’ve looked at, though I haven’t done detailed profiling on most of the 5900Xes and 5950Xes I’ve built.
Just looked at the 9900X with short AVX-512 and it hits 120 A TDC at 10 threads. Stays TDC limited from there, though PPT is 150 W+ from 6-24 threads. Short FMA stays at 162 W with 8-24 threads with TDC rising from 113 to 120 A. PPT-TDC are more closely balanced here than PPT-EDC on the dual CCD Vermeers I’ve looked at.
Highest 9900X temperatures are at six threads with FMA being more stressful. 95 C instead of 90. Thermalright Phantom Spirit with TF8 and the TL-C12s topping out ~890 RPM.
Hey guys, started testing the chip but, unfortunately, California’s wildfires are trying to melt my neighborhood down.
So, until the danger is clear, all the parts are packed up and ready to hightail it out if we’re forced to evacuate.
Hopefully it won’t be too much longer, but, apparently, next few days are going to be the most likely for the fire to spread. Here’s to hoping it won’t.
Thanks again for all the help. I appreciate all the info on esoteric tech knowledge, and I’ll update soon as I’m not in the marshmellow roasting zone.
This topic was automatically closed after 270 days. New replies are no longer allowed.