tresnugget

Have you done any testing with port royal and the 12900k with e cores enabled vs disabled + higher ring?

haven't done a lot of 3D on 12th gen tbh. i ran SLI 3090s with LN2 12900K. All cores enabled, HT off should produce best scores. Last night i'm running 12900KS at 5.4-5.5 with e cores at 4.1-4.3 and ring 4.5. It can go harder and I didn't hook up the chiller to cpu loop bc i wanted to see what kind of power the ti would draw before adding another 700-800w on the circuit lol
 
the circuit can prob take it, but it was bizarre to see the 8 pins drawing 145w each (and i can't see if the other 3 not on pmd are drawing this, but i assume they are) - when i do my SLI runs, if the chiller is in the mix for cpu i run an extension cord from another room for the second 1600w psu to power the second card and this avoids shutdowns for bench runs, but it really isn't the best setup haha
 
did you have more or less success tinkering with the P+E core configs? i believe generally the meta is:
 
TimeSpy = HT on, all cores on
Port Royal = HT off, all cores on
 
PR does benefit from higher CPU OC somewhat, but not as much as TS. and TS is boosted a lot by tighter timings/better ram effi whereas PR more or less doesn't care so long as ram oc is stable 

zippytek

tresnugget

Have you done any testing with port royal and the 12900k with e cores enabled vs disabled + higher ring?

haven't done a lot of 3D on 12th gen tbh. i ran SLI 3090s with LN2 12900K. All cores enabled, HT off should produce best scores. Last night i'm running 12900KS at 5.4-5.5 with e cores at 4.1-4.3 and ring 4.5. It can go harder and I didn't hook up the chiller to cpu loop bc i wanted to see what kind of power the ti would draw before adding another 700-800w on the circuit lol
 
the circuit can prob take it, but it was bizarre to see the 8 pins drawing 145w each (and i can't see if the other 3 not on pmd are drawing this, but i assume they are) - when i do my SLI runs, if the chiller is in the mix for cpu i run an extension cord from another room for the second 1600w psu to power the second card and this avoids shutdowns for bench runs, but it really isn't the best setup haha
 
did you have more or less success tinkering with the P+E core configs? i believe generally the meta is:
 
TimeSpy = HT on, all cores on
Port Royal = HT off, all cores on
 
PR does benefit from higher CPU OC somewhat, but not as much as TS. and TS is boosted a lot by tighter timings/better ram effi whereas PR more or less doesn't care so long as ram oc is stable 

Lol same. I've done very little on 12th gen with 3dmark. Pretty much only since I've gotten the 3090 Ti KPE. So far I've only ran my CPU on its daily settings at 5.3 p core, 4.2 e cores, and ring at 4.2. I can push to 5.4 and probably have more headroom on the e cores and ring as well but just running into a wall temp wise. But yeah, tonight I'll do more experimenting with the e cores and ring and turn off ht.

zippy is that one of the new praxis flat benches? or is it custom made?  either way - it looks really clean and well designed with the MORA.  I've been dying for an open, flat test bench with some good storage that isn't also humongous.  using a plain jane KP atlas test bench right now..
 
e: keep us posted on card 2 :P

That is not my bench, it’s a friends with 3090 Ti Hall of Fame OC Lab Limited edition

Port royal is 100% a gpu test only. CPU can help a tiny amount but only if it's preventing the gpu from being fully utilized. I've seen 4790Ks perform perfectly adequate. But yes slight improvements can happen with a faster CPU. As GPUs get stronger and fps increases I expect CPU will become more important just as it does in high refresh games.

mylittlepwny2
Port royal is 100% a gpu test only. CPU can help a tiny amount but only if it's preventing the gpu from being fully utilized. I've seen 4790Ks perform perfectly adequate. But yes slight improvements can happen with a faster CPU. As GPUs get stronger and fps increases I expect CPU will become more important just as it does in high refresh games.

Luumi says a faster cpu helps in PR. I’ve noticed this myself as well.

Sajin

mylittlepwny2
Port royal is 100% a gpu test only. CPU can help a tiny amount but only if it's preventing the gpu from being fully utilized. I've seen 4790Ks perform perfectly adequate. But yes slight improvements can happen with a faster CPU. As GPUs get stronger and fps increases I expect CPU will become more important just as it does in high refresh games.

Luumi says a faster cpu helps in PR. I’ve noticed this myself as well.

Yup. Noticed the same using the same GPU in 2 slightly different builds. Not sure if it was the CPU (5.2GHz vs 5.0GHz clock) or the RAM (4000MHz vs 3600MHz XMP) that made the difference, but there certainly was a difference.
 

I said cpu does slightly help. But it's marginal. Ram itself makes no real difference. Whole benchmark is 99.9% gpu bottlenecked. That last .1% comes from CPU. Don't believe me? Go look at how many 4790K scores are on the hall of fame.

The fact that XOC guys don't use LN2 on their CPUs while running the bench tells.you everything you need to know. There's simply no need. CPU just needs to be fast enough to get out of the GPUs way. It plays a slightly bigger role when running SLI and you're pushing 140+ fps average. But even then it's a minor difference.

AMD is different. Pre Zen 3 AMDs cpus have been much worse for efficiency. Running a zen 2 cpu for instance at same clocks as a 10900K you'll be 5%-7% lower scores at the same clocks. And it gets worse as you go older. The CPUs and memory latency really get in the way. Tested it all personally

 
I'll take your word for it since I don't do anywhere near enough benching to know, and there's always some variation from run to run even within the same rig on the same day. It was just that I happened to do a quick run only a couple of hours earlier on my 3090 on a Z590 FTW @ 5GHz + 3600MHz (4x8GB) ram, vs a run a while ago on the same GPU but Z590 Dark @ 5.2GHz, and 4000MHz (2x16GB) ram, and there was a significant difference on both Port Royal and Time Spy, BUT it was only 1 run of each today (I was testing an issue with BF2042 suddenly dropping down to 25-35 FPS), so my experience was certainly not conclusive.
 
FYI, it seems the problem was a corrupt DirectX installation. All good now, and back to ~140+FPS in BF2042, so I'm happy. Also finally got off the 472.12 driver - 516.59 appears to be good. Winning!
 
 

The CPU can definitely effect the PR score.  Weirdly, my 5800x3D always scores about 75-100pts lower in PR than my 5800x.  I'm not sure why.  Maybe the difference in clock speed, although I tried testing some all core OC's on my 5800x (4400, 4600, 4800mhz) and it made no difference in my PR score.  
 
Anyway I was surprised by that result.  Here is a comparison of runs.  Average temp was slightly warmer in the 5800x3D runs below, but I've repeated this result with equal temps as well various times with both my 3080ti and 3090.  Definitely not just the temp difference,
 
3080ti
https://www.3dmark.com/compare/pr/1586426/pr/1562416
 
3090
https://www.3dmark.com/compare/pr/1544792/pr/1546133