Is there an EPYC Trubo Boost Clock Breakdown/Safe Max Temps Anywhere?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

IamSpartacus

Well-Known Member
Mar 14, 2016
2,515
650
113
Also don't forget if you set cTPD, set package power limit too.

For some reason changing these on 7282 didn't do anything. Works great on 7452 though, changing from default 155w to 180w pulls almost exactly 50w more from the wall on 100% load.

I set to max on my 7452 with determinism on performance. All cores are running 2823 MHz on full load. Running different number of threads predictably lets those cores boost higher.

But it's definitely a power limit, and not temp related because both CPUs are running ice cold on SM 4U heatsinks.

*Edit: One question I have is, does changing other settings that contribute to total package power, change boost behavior by allowing more cores to run at higher frequency? AMD's tuning guide has a handful of references like this: "Lowering the memory clock speed also results in power savings in the memory controller, thus allowing the rest of the SoC to consume more power potentially resulting in a performance boost elsewhere."
See 2.8GHz all cores for that CPU seems more like what I'd expect. If one can run all cores at boost clock speed what exactly is the point of having a base clock?
 

sno.cn

Active Member
Sep 23, 2016
211
75
28
Baseline for marketing? Maybe that's worst case with power settings so people don't buy a CPU they expect to run on boost clocks with all cores, and then complain or try to sue the manufacturer for false advertising.
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,515
650
113
Baseline for marketing? Maybe that's worst case with power settings so people don't buy a CPU they expect to run on boost clocks with all cores, and then complain or try to sue the manufacturer for false advertising.
Well given my current settings (only thing I changed is cTDP) it seems base clock batters as that's what I'm getting (3.0GHz) with all cores loaded. But yet you're seeing clocks almost 500MHz higher than your base clock on full load. Hard to parse this all out.
 

sno.cn

Active Member
Sep 23, 2016
211
75
28
Did you set both cTDP and package power limit? And did you set determinism to manual/performance?

Probably I have more headroom over base clock since mine's only 2.35. But 7452 max clock is 3350 MHz, and it does every bit of it when I drop load down to 16-24 threads (on 2P setup).

But if you look at 7402 with higher TDP, maybe it runs closer to max fully loaded?

I actually never checked loaded frequency before today since I've been getting excellent performance. But I think it almost negates the low base clock on the 7452 I posted in the deals section.
 

TXAG26

Active Member
Aug 2, 2016
397
120
43
I am running my ram at 2933 MHz, even though it is DDR4 3200. I wanted the slight latency benefits and didn’t need the slight bandwidth boost. According to the AMD documents, this reduces total package power and allows a bit more headroom for boost speeds.
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,515
650
113
Did you set both cTDP and package power limit? And did you set determinism to manual/performance?

Probably I have more headroom over base clock since mine's only 2.35. But 7452 max clock is 3350 MHz, and it does every bit of it when I drop load down to 16-24 threads (on 2P setup).

But if you look at 7402 with higher TDP, maybe it runs closer to max fully loaded?

I actually never checked loaded frequency before today since I've been getting excellent performance. But I think it almost negates the low base clock on the 7452 I posted in the deals section.
I did not set package power limit, I will look at that. When you say you hit 3.35GHz with 16-24 threads, that's out of 128 threads right? I'm curious what you could hit with that many threads on a single CPU. The 7452 is attractive if it can hit those high of clocks because for me I will hardly ever max my entire CPU at once. I mainly only need the higher clocks for a few select VMs.


I am running my ram at 2933 MHz, even though it is DDR4 3200. I wanted the slight latency benefits and didn’t need the slight bandwidth boost. According to the AMD documents, this reduces total package power and allows a bit more headroom for boost speeds.
Yea I've been meaning to test that as well. Need to get back into my BIOS and find the RAM speed config as I couldn't find it last time I was in there.
 

TXAG26

Active Member
Aug 2, 2016
397
120
43
Yea I've been meaning to test that as well. Need to get back into my BIOS and find the RAM speed config as I couldn't find it last time I was in there.
It is under Advanced->NB Configuration->Memory Configuration->Memory Clock->2933Mhz
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,515
650
113
It is under Advanced->NB Configuration->Memory Configuration->Memory Clock->2933Mhz
I'm not running the H11SSL anymore, I have the ASRock EPYCD8 board. I needed the additional PCIe x16 slot and M.2 slot.
 

TXAG26

Active Member
Aug 2, 2016
397
120
43
I'm not running the H11SSL anymore, I have the ASRock EPYCD8 board. I needed the additional PCIe x16 slot and M.2 slot.
That's right. I would think ASRock would likely use a similar basic AMI bios setup, but maybe not. I've never accessed an ASRock bios.
 

sno.cn

Active Member
Sep 23, 2016
211
75
28
I did not set package power limit, I will look at that. When you say you hit 3.35GHz with 16-24 threads, that's out of 128 threads right? I'm curious what you could hit with that many threads on a single CPU. The 7452 is attractive if it can hit those high of clocks because for me I will hardly ever max my entire CPU at once. I mainly only need the higher clocks for a few select VMs.
Running 16 threads pinned to one CPU, 3300 MHz. Running 32 threads pinned to one CPU, 3100 MHz.

Screen Shot 2020-01-28 at 9.14.01 AM.png
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,515
650
113
Ok found the RAM speed settings. In ASRock you have to enable RAM overclocking to be able to change the speed.




Also set the power package limit and determinism so I'll do some more testing now.