Read somewhere, that one of the bosses of AMD said, that they expect similar performance gains with Zen 4... I guess my 1600X won't last as long as I hoped to
Everything stock: CPU: AMD Ryzen 7 5800X CPU OC: Stock Memory: 64GB DDR4-3200 MHz Dual Memory timings: CL16-18-18-38 GPU: Nvidia Geforce RTX 3090
@Thomas Jansen re your replay test, that won't actually be stressing the CPU will it? Since AI isn't being calculated on the fly?
It absolutely is stressing the CPU, in fact, the fps is fully limited by the cpu, my 1080ti never went above 60% usage in the benchmark. It's just a test of which CPU handles the rendering part of the game the best, representative of a MP race, which is what I'm personally most interested in anyway. Sure with a bunch of AI running the fps will be limited even more and the gaps between CPUs might increase or decrease, but testing with live AI would be extremely hard to do consistently anyway.
@thomas, in your ACC benchmarks is the tuned memory profile using with Ryzen Memory Calculator settings - any chance you could post those timings? I've been in two minds about upgrading my CPU or GPU, but based on the benchmarks it seems that racing games have more to gain from the CPU. It'd be nice if there were more triple screen res benchmarks showing CPU/GPU scaling.
I am indeed using the 1usmus ryzen memory calculator: running this with one change being the dram voltage at 1.46, I recently bought another identical ram kit which couldn't even run at the 3333mhz preset though, so the silicon lottery definitely applies . The thing with CPU performance is that it's often not a direct fps increase, especially with triple screens or the high resolution VR headsets you will usually still be GPU limited. But a good CPU can give you the headroom to let the GPU really stretch its legs, and maybe most importantly, a good CPU will make for much more consistent performance, which is crucial in VR.
Thanks - I have 16Gb CL16 3200 matched, struggled to run the safe timing with the calc - had to stick to stock XMP profiles. I'm running 3x1080p triples - it's difficult without enough data points comparing different CPU performance with the same GPU or same CPU with different GPU to determine what scales better. I suppose I'll have to wait for the B450 beta bios for Ryzen 5000 to drop early next year and bite the bullet. I've added an extra benchmark for my laptop. I'll give the r3e replay bench a whirl on my main sys when work settles down. CapFrameX looks interesting.
an important factor is refresh rate on your monitors, if you just have 60hz monitors you most likely don't need a powerful CPU at all to feed the GPU. If you are running at 144hz or even higher the CPU starts to become very important.
All 3 are running at 144Hz, although I am running them with Freesync + LFC turned on. I could easily pick up a used Ryzen 3600 or 1080 non-TI for around the same price ~300 AUD.
With the latest bios with PBO2 and curve optimiser I gained another small bit, especially very good min. fps Almost at 5GHz single thread speeds now
@Thomas Jansen What VR headset do you use? I'm using the reverb g2 now. I have a 6 core cpu (4.9ghz AVX), and the r3e executable load is spread over all cores for some reason. I can handle a maximum of 74% super sampling (per eye 2840x2772) in steamvr now, any higher and it results in frame drops. According to fpsvr, almost all my latency is coming from the CPU. The 1080ti has significant load as well though. I think a faster CPU for me is needed, I'm very curious how much I can crank up the SS with these new AMD's, and if it's worth the upgrade. I was considering to buy the 6core version, but that might not cut it.
At that kind of supersampling I don't think you will gain anything from a faster CPU. fpsvr is only showing cpu frametimes because the VR in RR is basically 'hacked' in, so it doesn't report the frametimes as normal. A faster CPU will maybe get you a tiny bit more headroom for more SS, but a better GPU would probably make much more of a difference. SS is pretty much all GPU load.
This curve optimiser undervolting feature for ryzen 5000 is amazing I managed to get much lower temps at higher clocks in heavy stuff like cinebench (4.55ghz at 90c to 4.65ghz at 82c). and much more consistent boost behaviour in applications with lower load, it's very consistent at 4.95ghz on a couple of threads such as in the benchmark here. Finally able to take the crown from the mentally tuned 9900K here
A decent 5% gain from stock in the 720p RR benchmark: I saw a similar increase in ACC as well, and some Cinebench R20 scores for anyone curious:
very nice! would be interesting if you could give this one a try: https://forum.sector3studios.com/in...pu-benchmark-thread.13473/page-18#post-212197 Considering the huge gap between zen 2 and zen 3 I wonder if zen 2 was just really bad in RR or if zen 3 is just really good in RR, so I'm curious where intel would stack up
Set arrived 2 days ago and i am still testing on my testbench (with a crapy GPU) so raceroom is no option for the moment except for the slightly higher clock speeds and the optimized power / temp values I would not expect any big differences, Intel ipc is just the same and i expect your CPU to score better in single core performance
In the latest BIOS version, ASUS enabled this exclusive PBO Fmax Enhancer feature which should slightly increase single thread performance on 3000-series CPU's, in theory at least. And it appears to be working, because I'm getting a ~100 higher score than usual: BUT, it looks like I'm getting random BSOD's because of it. Strange because I'm basically on defaults except for the DOCP profile.