Curve Optimizing My 7900 non X, how long ????

[Jamin280672]

Ive got bored of messing around with ram now on the new bios, it just doesnt seem to be giving me anything extra out of my kit, so ive taken to playing with CO instead to try and get my temps down a little.

Im running a 7900 non X, with 64gb (2 x 32gb) G.Skill Trident Z5 CL30 6000mhz @ Buildzoids timings and 6200mhz, on an ASUS X670e Gene, ive added a +125mhz boost to the CPU, I cant go any higher as it causes all sorts of problems, ive just ran quick tests like Cinebench to verify this for now.

So now im playing with CO, at the moment all my cores are at -15, I started at -10 a few hours ago, the way im testing is with OCCT who make it easy to do per core tests (you can see it in action in the pict below), so far ive managed to shave about 10oC off the chip, if I can get all cores stable at -20 then im going to probably stop there as im not going to fight to get maybe -1 or -2 more out of some cores and -5 out of others.

I just want to clarify that im doing this right, so ive set the timer for 2 hours, this allows for 2 passes of each core, 5min testing at a time on each core, a total of 10mins on each core across the 2 hours, im using the small iterations set and im starting with SSE instruction set as this gives the highest boost, and then testing 2 hours with AXV, then again with AVX2, a total of 6 hours running, total of 30mins on each core, if theres no errors, I then goto the bios, adjust by 5 points and run all the tests again, would this be the correct way to do it ? i'll eventually finish off with an overnight test for each instruction set to make sure, ive never messed with CO before so sorry if this sounds like a stupid question, the only chip I did in the past was my 5800X3D and that was easy, unless you got a bad one, they all did -30.

 

[Wing-Man]

i tested with CB R23 for 30min pre step with CO

 

[Mick-1965]

I'm new to AMD always used intel in the past.
But... My 7900 I set to -30 on all cores, boost overide to 125, and manually adjusted PPT, TDC, and EDC. 128w,100A and 160A to let the rope out a little from stock but not so much as it was too inefficient.
All seems stable, so that'll do me.
Tested with core cycler for several hours, cinebench R23. and AIDA SHA-3 benchmark many many conseutive runs (Apparently picks up undervolt/overclock instabilities very quickly (Look out for WHEA errors in the event log or HWInfo))

 

[Jamin280672]

I'm new to AMD always used intel in the past.
But... My 7900 I set to -30 on all cores, boost overide to 125, and manually adjusted PPT, TDC, and EDC. 128w,100A and 160A to let the rope out a little from stock but not so much as it was too inefficient.
All seems stable, so that'll do me.
Tested with core cycler for several hours, cinebench R23. and AIDA SHA-3 benchmark many many conseutive runs (Apparently picks up undervolt/overclock instabilities very quickly (Look out for WHEA errors in the event log or HWInfo))

Thanks, seem like my CCD2 is a weak one compared to CCD1, at the moment I have all the cores on CCD1 at -25, on CCD2 I have 1 at -25 (currently testing), 3 are -18 and 2 at -20
I'll see after this overnight run which ones can do -30, not holding up much hope for CCD2, in fact theres only 1 core that will possibly do -30 on CCD2 which is obviously the one at -25 now testing.
Ive decided to use OCCT for a quick 10 min test on the core ive adjusted, if it passes, im running an overnight test using Core Cycler, seems to pick up the errors quicker/better
Theres a few too many options in Core cycler though, an I supposed to run all the tests eventually ? like SSE all tests, AVX All Tests, AVX2 All Tests, Y-Cruncher All tests etc, at the moment im running Prime 95 / SSE / Large.

 

[Jamin280672]

i tested with CB R23 for 30min pre step with CO

This is not all good, CB23 uses AVX2, thats a high stress load on all the cores, giving the lowest boost clocks, thats why people test using SSE instructions with a large or huge data set, it gives a lighter load on the cores making the CPU boost to its highest clocks.

 

[alec]

On Zen 3 I preferred OCCT SSE test, because instability came at lighter loads
On Zen 4 after I found that AIDA benchmark suite was crashing, prefer OCCT AVX2 mode. Seems the main error mode is in heavy load not-quite-top-clocks.

For per core curve optimizer I prefer to go from bottom up.
Set every core to say -40, OCCT will happily find errors quickly and tell you which cores, bump those by 5 and repeat.
And only when everything seems to be stable for 15 minuts do a diverse multihour battery of tests, different games etc.
Usually if you just use "OCCT 15 min stable point" and bump every core by 5, it will likely be stable for everything

 

[Mick-1965]

at the moment im running Prime 95 / SSE / Large

I just ran it at at the default settings. which is Prime95 SSE "Huge" for several hours.

Try AIDA SHA-3 bench, Mess about with the parameters, HT on -off 1 core 2 core 12 core etc etc as it will pick up instabilites very quickly and give you a starting point. Just a shame you have to keep clicking start