Question Ryzen 9700X - - - large power draw disparity between cores during single core benchmark ?

[Neocortex_]

I recently built a new PC with a Ryzen 9700x. I enabled PBO but reduced the TPP limit to 110W to keep temps <65C during multicore loads, and am currently in the process of undervolting.

I noticed that, while benchmarking and stability testing with a cycling single core load (using OCCT), some cores draw quite a bit more power than others (17-18W vs 13-14W for the extremes). Which cores draw how much power is consistent after several hours of testing, and Tdie temps are more or less proportionate to the amount of power drawn (~75C vs ~60C) but otherwise stable.

The cores drawing the most power also seemed to have a bit of trouble keeping a constant 5550MHz clock speed, sometimes dipping to 5500MHz, but this might also be malcebo given the small difference. From what I can tell, the range (at least when it comes to Tdie temps) is mainly a result of the undervolt, causing the power draw of the currently better performing cores to go down, rather than the power draw of the concerning cores to be higher out of the box.

Edit: Running without undervolting, the cores of concern run about 1-2% slower than the other cores, and none hit the target frequency of 5550MHz. My guess is, during undervolting, the other cores hit the target frequency sooner and so will reduce the power draw. The cores that have me concerned barely reach that frequency with undervolting, and so keep that 17W power draw. Is this 1-2% performance difference between cores anything concerning or just natural variance that happens sometimes that has some scary but harmless appearances when pushed with undervolting?

Should I be concerned/looking into something?

 

[Misgar]

I recently built a new PC with a Ryzen 9700x. I enabled PBO but reduced the TPP limit to 110W to keep temps <65C during multicore loads, and am currently in the process of undervolting.

Why are you restricting the performance of the CPU? It's a bit like towing a trailer behind a sports car and keeping the handbrake on.

My 7950X peaks at +93°C and averages +85°C during long video transcodes with no signs of distress. I don't have PBO enabled and I'm not undervolting, because I need a stable system. If I restricted CPU power, render sessions would probably take 18 hours instead of 12.

 

[Neocortex_]

Why are you restricting the performance of the CPU? It's a bit like towing a trailer behind a sports car and keeping the handbrake on.

My 7950X peaks at +93°C and averages +85°C during long video transcodes with no signs of distress. I don't have PBO enabled and I'm not undervolting, because I need a stable system. If I restricted CPU power, render sessions would probably take 18 hours instead of 12.

The out of the box TPP of the 9700x is 88W, so compared to non-PBO configuration, it is not restricted. Setting it to 110W is, as far as I am concerned, the sweet spot for increased performance with PBO without my PC turning into a jet engine and my room turning into a sauna.

 

[TerryLaze]

That's completely normal with high core count CPUs, if they had to have all cores be exactly the same they would have incredibly bad yields or would have to clock their CPUs way slower to make them all run the same.

AMD has both "preferred" and "best" cores.
https://www.anandtech.com/show/15137/amd-clarifies-best-cores-vs-preferred-cores

 

[Neocortex_]

That's completely normal with high core count CPUs, if they had to have all cores be exactly the same they would have incredibly bad yields or would have to clock their CPUs way slower to make them all run the same.

AMD has both "preferred" and "best" cores.
https://www.anandtech.com/show/15137/amd-clarifies-best-cores-vs-preferred-cores

Alright, thank you ^^ I figured some variation between cores is to be expected, but I don't really have a frame of reference for how much variance could be considered normal. The first thing I noticed was also the big Tdie temp discrepancy. Since those were stable per core and there weren't any runaway thermals, I wasn't too concerned, but figured it wouldn't hurt to dig into a bit more and ask around

 

[Misgar]

110W is, as far as I am concerned, the sweet spot for increased performance with PBO without my PC turning into a jet engine and my room turning into a sauna.

That makes sense. I might not run too many 4K transcodes this summer if we have any heatwaves.

Fan noise isn't too bad in my ancient Lian Li 'Super Silent Case'. Two 120mm intake fans, one 120mm and one 80mm exhaust fan, plus Noctua NH-D15 air cooler and GPU fans. It's all fairly well baffled against noise with re-entrant air intakes.
https://www.pugetsystems.com/labs/articles/Case-Review-Lian-Li-PC-S80-25/

My 7950X/RTX 3060 consume 400W during 11 to 36 hour 4K transcodes. With the new RTX 4070, it'll be 430W of heat in the room.

My HP Xeon servers with four screaming Delta fans are much noisier. I can tolerate them for an hour or two when they slow down after the initial POST. Working in large server halls gives you a certain tolerance for fan noise (going deaf).