5900x PPT, TDC, EDC

[Shoopman]

3 hours ago, YoungBlade said:

If you want to test stability, I'd recommend using OCCT. Run the large, variable AVX2 stress test for 15 minutes on each core individually (advanced options, uncheck all but one core, also set it to check both physical and virtual, disable the other options). If errors show up, your curve offset is not truly stable.

Good call, I'll do that for sure. I haven't had any blue screens with -30 on all cores, but I definitely want to make sure it's rock solid before I go any further.

[YoungBlade]

6 minutes ago, Shoopman said:

Good call, I'll do that for sure. I haven't had any blue screens with -30 on all cores, but I definitely want to make sure it's rock solid before I go any further.

Mine appeared stable with a -25 offset after some basic testing, but it wasn't really. I'd get a random shutoff every week or two. It's highly unlikely to be 100% stable for all cores unless you got a golden chip. Of course, you can run it in a mildly unstable state if it never causes problems, but I don't know how valuable that is.

 

With the test method I used, my negative offsets for each core resulted in:

20, 15, 10, 25, 20, 25, 25, 25, 25, 30, 23, 25

 

Basically, just over half the cores can do 25 or better. All but two can do at least 20. Then the preferred cores are stuck at just 15 and 10. From what I've read online, that's pretty typical. Sometimes, the most preferred core can't do any offset at all, even.

[Shoopman]

23 hours ago, YoungBlade said:

Mine appeared stable with a -25 offset after some basic testing, but it wasn't really. I'd get a random shutoff every week or two. It's highly unlikely to be 100% stable for all cores unless you got a golden chip. Of course, you can run it in a mildly unstable state if it never causes problems, but I don't know how valuable that is.

 

With the test method I used, my negative offsets for each core resulted in:

20, 15, 10, 25, 20, 25, 25, 25, 25, 30, 23, 25

 

Basically, just over half the cores can do 25 or better. All but two can do at least 20. Then the preferred cores are stuck at just 15 and 10. From what I've read online, that's pretty typical. Sometimes, the most preferred core can't do any offset at all, even.

So far for me most of them are ok at -30 except for the preferred cores, but I've only been running a test for 15 minutes on each core. Not sure how long it would take to be 100% sure of stability.

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

UPDATE: After running OCCT for 15 (sometimes 30) minutes on every individual core I wound up with:

 

25, 30, 25, 30, 30, 30, 20, 15, 30, 30, 30, 20

 

Bolded cores are preferred (#0, 2, 6, and 7) hence the lower offsets. Core 11 may be the 5th best core but I don't know how to determine that.

 

I also ran OCCT on all cores, same settings, for 1 hour and had no errors so I think I'm good for now. Oddly Core 11 would throw errors on this all-core test consistently, but only sometimes on the single core test. I lowered the offset to 20 and the errors stopped in both tests.

 

I'm only boosting up to ~4400 in R23 all core though (Score of 22168)... not sure why since I hit 4700 in the OCCT all core test.

 

Clocks in gaming have not changed, was hitting 4900-4950 in RE8 and 4775-4825 in Halo Infinite

 

If I really want to waste my own time I might run each core again for 1 hour just to be 100% sure, but that's a huge ordeal that would take several days.

 

[Shoopman]

Update to the update, I have verified @maka0's optimal settings. 135 TTP, 100 EDC, and 90 TDC seem to give the best results. Single core performance, as far as I can tell, is virtually indistinguishable from stock settings (both with curve optimizer enabled). The difference in R23 was <10 points and the different in-game (Halo Infinite) was 1-2 FPS, basically margin of error. Stock settings in Halo Infinite gave 4825-4875 MHz, while the optimized settings did slightly better at 4850-4900, though it was at 4900 the majority of the time.

 

Multi-core took a very small hit, about 2%. Stock settings in R23 gave a score of about 22,300. This dropped to about 21,900 with the above settings.

 

I will re-verify that my curve optimizer settings are stable now that I've changed the TTP/EDC/TDC but I doubt anything has changed. I will report back if I have to make any changes.

 

EDIT: Here are my final settings. I verified them all with a 1 hour test in OCCT, plus a one hour multi-core test. Not willing to do any more because that would require paying for OCCT which is subscription only, unreasonable IMO. Still getting random restarts occasionally so not sure what that's about. Usually while I'm just watching Youtube and doing nothing else so it's not like anything is being stressed. Anyway, here they are:

 

0- 25*
1- 25
2- 20*
3- 30
4- 30
5- 30
6- 15*
7- 15*
8- 30
9- 30
10- 30
11- 20

 

Cores with an asterisk are preferred cores. Based on this I have to conclude that cores 1 and 11 are the 5th/6th best cores. Happy that everything else was able to stay at 30.

 

[maka0]

On 11/24/2021 at 2:49 PM, Shoopman said:

Update to the update, I have verified @maka0's optimal settings. 135 TTP, 100 EDC, and 90 TDC seem to give the best results. Single core performance, as far as I can tell, is virtually indistinguishable from stock settings (both with curve optimizer enabled). The difference in R23 was <10 points and the different in-game (Halo Infinite) was 1-2 FPS, basically margin of error. Stock settings in Halo Infinite gave 4825-4875 MHz, while the optimized settings did slightly better at 4850-4900, though it was at 4900 the majority of the time.

 

Multi-core took a very small hit, about 2%. Stock settings in R23 gave a score of about 22,300. This dropped to about 21,900 with the above settings.

 

I will re-verify that my curve optimizer settings are stable now that I've changed the TTP/EDC/TDC but I doubt anything has changed. I will report back if I have to make any changes.

 

EDIT: Here are my final settings. I verified them all with a 1 hour test in OCCT, plus a one hour multi-core test. Not willing to do any more because that would require paying for OCCT which is subscription only, unreasonable IMO. Still getting random restarts occasionally so not sure what that's about. Usually while I'm just watching Youtube and doing nothing else so it's not like anything is being stressed. Anyway, here they are:

 

0- 25*
1- 25
2- 20*
3- 30
4- 30
5- 30
6- 15*
7- 15*
8- 30
9- 30
10- 30
11- 20

 

Cores with an asterisk are preferred cores. Based on this I have to conclude that cores 1 and 11 are the 5th/6th best cores. Happy that everything else was able to stay at 30.

 

Sorry to bring this back from the dead! 

 

Just wanted an update on how you went? 

My friend is getting new 7000 series and I suggested looking at messing with the power values to give him the same performance at a lower temperatures

[freeagent]

Whew.. old thread. Due to AGESA and updates and all that I run my 5900X at 245/165/190. Things were different when this thread was new

 

She does about 518-520 GFlops with PBO.