new 7950x, help understanding behaviour.

[robinlawrie]

Hi all, hoping for some education. my beastly old work machine (3990x, 128gb ram, huge external watercooling system) died in the middle of a job, vrm popped. ive got a new motherboard coming, and hoping with all my heart i dont have to throw a 4000 quid cpu in the bin. that would make me cry.)

in the meantime i was forced to buy a replacement machine very quick(there goes the fee for the job)

i couldnt get anything like what i had before, i had more cash back then. however i still need to be able to render my big 3d scenes in vray at a decent speed. so i got a 7950x, msi tomakawk x650 motherboard, 128gb ram (getting the ram up to speed will be another story, for later) for cooling i chose an endorfy navis 280mm AIO

im attempting to get the most out of it i can, as a daily driver. on the 3990x it was easy, enable PBO, set limits within reason, instant big clock boosts. my big cooler could keep the thing under 60 degrees even pushing 4.3ghz on all 64 cores.

the 7950x seems rather limited..pbo as i used it on the 3990x makes no difference. from my reading, ive tried setting the curve offset thingy to -15 and adding 100mz to the max clock.. this seems to improve the clocks a bit, on HWInfo, i go from 5ghz all core to 5.2 on ccd1 and 5.05 on ccd2.. the temps sit around 93.8 degrees, so not hitting the 95 degree limit.

however, my vraybenchmark scores have gone down? from 39180 (limited by ram stuck at 3600, vray isnt very ram speed sensitive, but it does make a difference, typical scores with faster ram are around 41-42000) to 38700.. if i leave the benchmark running for 10 minutes instead of the default 1, the score goes even lower, to 37,500

can anyone explain what might be happening? i dont seem to be thermally throttling, and the clocks remain higher than before the curve offset.. however, lower scores... i also find that the claimed 5.85gz top speed is only hypothetical.. seems to achieve it only for a millisecond every few minutes.. the other 99.999 percent of the time its capped at 5.5, even when doing stuff in a single threaded app.

thanks for any advice...

 

[Doco]

If you've just done a quick PBO and +mhz without too much stress testing you could have gotten into a scenario's where you are experiencing clock stretching, where the cpu is showing to boost to higher speeds (and with pbo at lower voltage) but in reality its not entirely stable, the system doesn't crash but you will see a deterioration in scores in benchmarks etc

If you are going to be playing about with PBO and a +mhz Overclock I always recommend doing the slow and tedious way of increasing one and then the other, running benchmarks after each change to see if it makes an impact.

 

[Jamin280672]

Hi all, hoping for some education. my beastly old work machine (3990x, 128gb ram, huge external watercooling system) died in the middle of a job, vrm popped. ive got a new motherboard coming, and hoping with all my heart i dont have to throw a 4000 quid cpu in the bin. that would make me cry.)

in the meantime i was forced to buy a replacement machine very quick(there goes the fee for the job)

i couldnt get anything like what i had before, i had more cash back then. however i still need to be able to render my big 3d scenes in vray at a decent speed. so i got a 7950x, msi tomakawk x650 motherboard, 128gb ram (getting the ram up to speed will be another story, for later) for cooling i chose an endorfy navis 280mm AIO

im attempting to get the most out of it i can, as a daily driver. on the 3990x it was easy, enable PBO, set limits within reason, instant big clock boosts. my big cooler could keep the thing under 60 degrees even pushing 4.3ghz on all 64 cores.

the 7950x seems rather limited..pbo as i used it on the 3990x makes no difference. from my reading, ive tried setting the curve offset thingy to -15 and adding 100mz to the max clock.. this seems to improve the clocks a bit, on HWInfo, i go from 5ghz all core to 5.2 on ccd1 and 5.05 on ccd2.. the temps sit around 93.8 degrees, so not hitting the 95 degree limit.

however, my vraybenchmark scores have gone down? from 39180 (limited by ram stuck at 3600, vray isnt very ram speed sensitive, but it does make a difference, typical scores with faster ram are around 41-42000) to 38700.. if i leave the benchmark running for 10 minutes instead of the default 1, the score goes even lower, to 37,500

can anyone explain what might be happening? i dont seem to be thermally throttling, and the clocks remain higher than before the curve offset.. however, lower scores... i also find that the claimed 5.85gz top speed is only hypothetical.. seems to achieve it only for a millisecond every few minutes.. the other 99.999 percent of the time its capped at 5.5, even when doing stuff in a single threaded app.

thanks for any advice...

Not all cores will do -15 on the curve, in fact your best cores will probably only do -5 if you're lucky, and some of them -0 as these are already heavily tuned from the factory to put it in laymens terms
The easiest way to find out which are you best cores is to download Ryzen master, they will be flagged as your best cores.
The only way to find out exactly what they will do is to test test test extensively with something like Core cycler, otherwise just set them to -5, if your scores are still rubbish, set them to -0 if you cant be bothered to test days upon days.
Everything on your PC will more than likely be using these cores first unless you've changed that in the bios (CPPC preferred cores)
So now as @Doco stated, you are probably clock stretching those cores and they are under performing because of it.
You have basically said I want an extra 100mhz out of all of the cores and use a lower voltage curve to do it.

For me, I tested with Core Cycler, I tested SSE, AVX, AVX2 and AVX512, Y-Cruncher, PI, and OCCT as well as Memtest Pro which I find picks up core errors quite quickly, it took well over a week, I set it to run before I went to bed and left it running all night and day whilst at work, if any core fails along the way, testing starts all over again from square 1, if they all pass, then adjust the negative curve, and start testing all over again.

The easiest way is just to enable PBO and be done with it, the performance increase between enabling PBO and fine tuning things is only about 5%, watch some Skatterbench video's based on the 7950X to find out whats truley involved and the performance difference you get with different configs.

Ryzen 9 7950X Undervolt & Overclock to 5950 MHz With X670E Aorus Master | SkatterBencher #45

In today's video, I overclock the AMD Ryzen 9 7950X Raphael CPU up to 5.95 GHz with the GIGABYTE X670E Aorus Master motherboard and EK custom loop water cool...

youtu.be