Some cooling and noise questions

Valantar

SFF Guru
Jan 20, 2018
1,175
961
I am not using the LNA adapter. I was really trying my best to find an approach that could keep the processor cool in the tiny case, that was a higher priority for me than noise. I wanted to make sure that I could get the higher cpu fan speeds to keep things cooler.



I had read that same advice for Ryzen 3rd gen as well, but I haven't found that to really be the case. I think it might be a function of the fact that I built in a very small case with no real way to cool other than the cpu cooler or perhaps I have a processor that undervolts well. But I have actually found MUCH better performance (both single threaded and multi-threaded) with undervolting and running a fixed frequency than with stock conditions. I briefly tried to use voltage offsets, but I didn't find much success with them. Perhaps I need to investigate that further.

Stock cpu with memory OC:


4.2GHz all-core @ 1.15V
4.2 GHz at 1.15V is very good, I'd say borderline golden sample level. I've seen people needing 1.3V to hit 4.1 all-core on various 3000-series chips (and for the record, exceeding >1.3V is when silicon degradation starts becoming a problem, with 1.35 being pretty definitively unsafe over any extended time period. Stock operation exceeds this, but only for very short times, and monitors things closely to avoid degradation or damage. Software voltage readouts in stock operation are generally inaccurate due to how the chip's control and protection systems work.).

It might be that with a chip as good as yours in a low-spec SKU like the 3600 you're better off running fixed clocks and voltages like that simply because the specs of the chip won't allow it to boost that high for all-core loads. The 3600 doesn't AFAIK have the features to boost beyond its rated 4.2GHz peak speed, so given that you're hitting that all-core at a reasonable voltage you're doing as good as can be expected. But that's a very rare case.

Btw, is the stock operation pic with PBO on or off? It seems the current recommended mode of operations is leaving PBO off. IIRC this is due to it causing voltages to spike higher than necessary, triggering thermal protections too early. Looking at your stock image you're close to the EDC (peak socket current) limit at 96.7%, so increasing that slightly might be a good idea. There's still nearly 25% headroom before hitting the sustained current limit. My understanding is that adjusting EDC/TDC/PPT needs to be done carefully and generally only adjusting one parameter at a time, as just increasing everything tends not to work as one might think.

If you want to experiment my recommendation (mind you, this is all second-hand knowledge based on reading forums, not personal experience) would be to keep PBO off, UV on an offset, and possibly increase EDC incrementally, all the while running scored benchmarks to monitor if performance is actually changing. Monitoring voltages and frequencies generally won't do much, as they'll look good in software even if performance is degrading (say, due to too low voltages, current limits being exceeded, or conflicting parameters). Of course, as noted above, with your chip you're unlikely to ever see better performance than you're currently seeing. But your chip does sound very, very good.
 
  • Like
Reactions: paulesko

paulesko

Airflow Optimizer
Jul 31, 2019
269
201
That´s indeed a very good chip! I have a 3950x which is supposed to be very good binned from factory (I think it is) and I see (at 1.25-1.26v under load) 4.275 mhz on the bad cores and 4.475 on the good ones, so there seems to be a really high variation amongst diferent chiplets.

In fact, on my "bad" chiplet I have one single reaaally bad core that puts down the rest to that 4.275 mhz, but I´ve been searching the limit of the other 3 cores not asigning load on prime 95 to the bad one and they go up to 4.425 mhz before being unstable. I guess your chip is one of these chips that had a really bad one single core, but the rest are really good ones, so AMD on those cases would dissable the bad ones and sell them as 6 core chip. Lucky you!
 

txporter

Caliper Novice
Feb 1, 2020
29
5
4.2 GHz at 1.15V is very good, I'd say borderline golden sample level. I've seen people needing 1.3V to hit 4.1 all-core on various 3000-series chips (and for the record, exceeding >1.3V is when silicon degradation starts becoming a problem, with 1.35 being pretty definitively unsafe over any extended time period. Stock operation exceeds this, but only for very short times, and monitors things closely to avoid degradation or damage. Software voltage readouts in stock operation are generally inaccurate due to how the chip's control and protection systems work.).

It might be that with a chip as good as yours in a low-spec SKU like the 3600 you're better off running fixed clocks and voltages like that simply because the specs of the chip won't allow it to boost that high for all-core loads. The 3600 doesn't AFAIK have the features to boost beyond its rated 4.2GHz peak speed, so given that you're hitting that all-core at a reasonable voltage you're doing as good as can be expected. But that's a very rare case.

Btw, is the stock operation pic with PBO on or off? It seems the current recommended mode of operations is leaving PBO off. IIRC this is due to it causing voltages to spike higher than necessary, triggering thermal protections too early. Looking at your stock image you're close to the EDC (peak socket current) limit at 96.7%, so increasing that slightly might be a good idea. There's still nearly 25% headroom before hitting the sustained current limit. My understanding is that adjusting EDC/TDC/PPT needs to be done carefully and generally only adjusting one parameter at a time, as just increasing everything tends not to work as one might think.

If you want to experiment my recommendation (mind you, this is all second-hand knowledge based on reading forums, not personal experience) would be to keep PBO off, UV on an offset, and possibly increase EDC incrementally, all the while running scored benchmarks to monitor if performance is actually changing. Monitoring voltages and frequencies generally won't do much, as they'll look good in software even if performance is degrading (say, due to too low voltages, current limits being exceeded, or conflicting parameters). Of course, as noted above, with your chip you're unlikely to ever see better performance than you're currently seeing. But your chip does sound very, very good.
For the stock settings, I just reset to defaults in the BIOS and manually overclocked the RAM again. I believe that sets the PBO to Auto. I am not sure if that is actually off or on. I will take a look at what you detailed and see what I can come up with letting Ryzen boost itself.

Well, if it is near golden level chip than I got lucky. I love the small size of the ML06-E but I really did need to dial things back to lower the temperatures just to be able to use the case.
 

txporter

Caliper Novice
Feb 1, 2020
29
5
Ok, I did some more testing to try to understand what running with an offset can buy me vs fixed frequency/voltage. I can see a drop in power and temperatures with the offset voltage, but I seem to lose performance, if anything.

Here are my Cinebench numbers:
[email protected] = 3810/490 (MT/ST)
Stock (no PBO) = 3463/472 (MT/ST)

I only measured multithreaded for the two offset voltage conditions:
-0.05V = 3445
-0.10V = 3416

I think the problem is that with PBO disabled, I am seeing max 3.8GHz.



EDC-PPT graphs (#1 #2) TDC-Effective VID (#1 #2)
 

nulio

Trash Compacter
Original poster
Jan 30, 2020
45
34
Hi!

I got tired of waiting for any PC case (Ghost S1 was damaged and the FormD T1 had an incomplete address) and started trying to optimize my system.

I did a bunch of benchmarks at stock - Cinebench, all six CPU Blender ones, all six GPU Blender ones and Unigine Heaven - which I can post later if there's curiosity.
I'm ignoring temps and noise now, to focus on learning how to tweak stuff.
I'll never have things too hot and loud. It's basically an open benchtable for now and I still didn't remove the LNA in the cooler fan.

Using Thaiphoon Burner I found out my memory - CMK32GX4M2B3200C16 - is Micron Technology and not Samsung: https://nulio.htmlsave.net/

I watched a guide on how to use the DRAM Calculator and:


I read a Corsair guide for a Aorus board, but x570, and I watched a couple of videos.
Then I went to BIOS and play:

1- Peripherals > AMD Overclocking > DDR and Infinity Fabric Frequency/Timings > Infinity Fabric Frequency and Dividers > 1600

2- M.I.T. > Advanced Voltage Settings > DRAM Voltage (CH A/B) > 1.35V

3- M.I.T. > Advanced Memory Settings > Changed everything that matched what the calculator gave

But not everything had the same order as the calculator, despite one video guide saying it would.
I matched everything I could.

I couldn't find matches to:
- SOC Voltage, VDDG CCD Voltage, VDDG IOD Voltage and cLDO VDDP Voltage;
- Power Down mode, BGS and BGS alt. I looked in all menus, what was there had different names.

It didn't post and the system cleared the CMOS. I redid it, same thing. Meh

In the end I left the Infinity Fabric Frequency and Dividers at auto and used the Extreme Memory Profile for the memory.

It barely did anything =(

SYSTEM
- AMD Rizen 5 3600
- Corsair Vengeance LPX DDR4 3200 PC4-25600 2x16GB CL16
- Gigabyte B450 I Aorus Pro WIFI - BIOS Version F50
- Windows 10 Education 18363

STOCK
- CPU - Auto
- FCLK - Auto: 1200
- RAM - Stock: 1067 15-15-15-36 @1.245V

Cinebench Multi: 3490
Cinebench Single: 473

XMP
- CPU - Auto
- FCLK - Auto: 1600
- RAM - XMP: 1600 16-18-18-36 @1.39V

Cinebench Multi: 3512
Cinebench Single: 474


All this work for basically nothing haha. I was expecting more. Especially after reading this: https://www.techjunkies.nl/2019/10/15/__trashed/

I have some questions.
- Did I make a mistake using the calculator? If someone with more experience is able to double check it would be great.
- Did I make a mistake in the BIOS? Does someone have a nice guide for this MB?
- Were this results to be expected?

Cheers