My motherboard is X570 Aorus Master on latest BIOS (F11)
Great! I have the same board. I'm on an older UEFI revision since I have a 3900x and I don't need 3950X support. You do, hence the need for a more-recent UEFI.
The Ryzen Master isn't exactly crashing, but sometimes it opens, and most of the time it won't open at all. The error message is always the same though "Ryzen Master unable to initialize. Kindly reinstall the program Hence Aborting!". Usually when I restart the machine it will open, but after I close it, it won't open again. Sometime it won't open up at all stating the same message. I tried reinstalling several times, deleting some registry items regarding Ryzen Master but still doesn't help.
I have had that happen before. Had to uninstall and reinstall to get it to work. Mighta done something else too . . . I think I ran DDU and flashed to a newer UEFI. You could always try using DDU, and if that doesn't help, you can reflash your UEFI after using DDU and see if that helps. Ryzen Master isn't required, but it is kinda handy.
I apologize, but I didn't get your explanation. This is my first time having an unlocked CPU and unlocked motherboard and I never overclock before so I don't really get what you are trying to explain here.
Ah, okay. Static OC refers to a manual overclock - you set a fixed voltage and clockspeed, and cross your fingers that it'll past stability testing. Doing so with Ryzen Master is really easy since you can quickly try different voltage/clockspeed settings without having to reboot constantly. But if it doesn't work then you'll have to do it the old-fashioned way.
So it's better to turn off PBO and try to do manual overclock ?
Yes. PBO should be helpful in that it unlocks power limits for your CPU, but it's notorious for feeding too much voltage for a given clockspeed. So you hit your power and/or thermal limits at too low a clock, and it's generally a mess. Some people still rely on it for whatever reason . . . on my system, I get higher performance with default settings (and a few odd voltage tweaks) than I do with PBO.
Manual overclocking lets you dial in the lowest possible voltage for a given clockspeed to help keep thermals in check, which is a must for Matisse since it has so many problems with hotspot temperatures. But it's hard to dial in a manual "static" OC that can apply to all workloads. For example, I've gotten 4.4 GHz all-core on a 3900x stable running Cinebench R20, but Prime95 SmallFFTs is difficult to stabilize past 4.2 GHz. Since all you're doing is transcoding, you can dial in a 4.2 GHz OC on your 3950X and tune for the lowest possible voltage that still allows the program to operate normally and call it done. You will wind up generating a good bit less heat (and lower temps) than if you let PBO try to hit that clockspeed on its own. Just don't expect that all-core OC to work on every workload.
I got suggestion to try offsetting the VCORE by -0.05V. To do this, I'm kind of confused. Is it to Dynamic VCore (DVID) or Dynamic VCORE SOC (DVID) ?
The first one. Change CPU Vcore from Auto to Normal, and then dial in an offset. I've gone as far as -.1v but you can try -.05V or whatever else you like. I don't recommend going further than -.1v though. I would also suggest switching LLC to Normal, which basically disables it (Auto won't give you that behavior). Then see how well your transcode works. Make sure everything else for your CPU is stock. The only things you want to change are your voltage offset and LLC.
Aside from that do you know a good tutorial for overclocking the CPU manually?
I haven't found one for Matisse yet. You can apply most of the same principles from Summit Ridge (1xxx) and Pinnacle Ridge (2xxx) CPUs with the caveats I listed above. Most people don't like static OC on Matisse since there isn't much headroom available. They don't understand the need to dial in tight voltages when PBO (and similar) does such a poor job.
I'm really a newbie and beginner here so it is kind of confusing on where to start. From videos I watched its like increasing CPU Clock Ratio and increasing CPU Vcore bit by bit and then testing for stability using blender / AIDA64.
That is normally what you do with an AMD system. With Matisse you need a slightly different approach.
The first thing I do is find a group of workloads that represent that most-stressful thing I'm actually going to do with my PC. Prime95 SmallFFTs doesn't count since it's so so hard to get that stable and in good temperature range on Matisse. It gets HOT. AIDA64 is pretty relaxed in comparison, so it's decent, but I found myself tuning against CBR20 and Blender Benchmark instead.
Then I test for thermal/power limits. I pick a safe speed - something like 3.8 GHz or, for a 3950X, I'd pick 3.5 GHz - and then I start raising voltage slowly until I figure out exactly how much voltage I can give the system without hitting unacceptable temperatures. Since you used Kryonaut I would set 80C as my peak temp. Anything above that, and you may cause the TIM to dry out - maybe. Though in reality the temp the TIM will reach will be far lower than the hotspots. Still, Matisse is much more efficient below 70C. Personally I prefer to keep my temps below 70, but it can be really hard to do. Especially when you have 16 cores.
Once I have my peak voltage dialed in, I start raising clockspeeds and I look for instability. Since PBO was able to get your transcode workload running @ 4.2 GHz on eight cores, you should probably try to hit 4.2 GHz stable in at least that benchmark. You should probably get lower volts/power draw/temps once you do things manually. Remember what I said about static OC being dependent on a workload-by-workload basis? Something that will be stable in your transcode workload might not work in Blender Benchmark.
And that's about it. Ryzen Master makes it really easy. If you can get it working.
And then try to overclock the memory to as close as 3733 Mhz
I would save that for later.
Are you actually saying that even with 100% utilization the temp will still be the same, because as with more cores it will automatically lower the voltage and the Ghz so that it will stay within the stated TDP range?
In default mode, pretty much yes. A workload that can hit 4.2 GHz on eight cores might only hit 3.5 GHz on sixteen cores. PBO doesn't diverge much. It supposedly raises TDP limits, but it starts hitting temp limits (exacerbated by stupid voltage choices from the PBO voltage/clockspeed/temp algorithm). The net result is that you rarely get to push beyond the default power limits or clockspeeds by much.
By 145.96 W it correlate with 93 C on temperature?
I hope not. That means your cooling solution is not doing a great job. But it certainly seems so . . .
Facebook
Twitter