Test: CPU Core count and RAM scaling in ACC, AC1 and R3E

Today I remembered this post of the Lord himself about the testing method of @Daniel Costa.

https://www.assettocorsa.net/forum/...caling-in-acc-ac1-and-r3e.59867/#post-1084562

I've done all my test ingame live without using the replay. I think this is the main reason for results are being that different.

 

Ah yes, good point. And as the post right after Stefano's said: "Lord Kunos is right, a benchmark feature in-game running full physics would help to obtain more accurate results"

 

So, do you have the test results?

 

Sorry about the old thread revival thing, but I wanted to share this:

So after one of my RAM sticks stopped working properly, I decided to buy the 2x8 GiB 2400 Mhz CL11 sticks, since I can't do a bigger upgrade in the time being, might as well improve what I have for now. Result: game seems to run clearly smoother. Since I effectively lost that RAM and I don't want to mess with timings to make the new ones run as the older 2x8 GiB ones at 1600 Mhz (older ones were CL10), I didn't do a proper comparison vs my older sticks real perfomance, but in Userbenchmark RAM tests I went from a 62.9% (21.4GB/s - 20GB/s - 59.8ns) score with the older sticks to a 85.4% (30.5 GB/s - 23.1 GB/s - 53.7 ns) score, and the improvement that I'm seeing I can't believe it's unrelated to . Testing the new sticks in the game at just 1600 Mhz with default timings means considerably worse timings than my older sticks, so the perfomance was obviously terrible, but I think I have good enough memory to remember that the older RAM sticks didn't allowed my PC to run AC:C as good as it does now when XMP is enabled.

Maybe game updates or other config in my PC is helping somewhat, but seems that my i5 4690k @4.3 Ghz is having a better time now running the game at a 60 fps target and I believe the RAM is the (only or main) reason why. So yeah, guys, don't cheap out on it. I have yet to try online though, I did races with 24 AI's, usually testing with 17-19 visible AI's, and mixed settings. There's some stutters but are a lot less frequent and not as big.

 

Actually, probably I could've get a better deal if I sold my mobo-ram-cpu as seb 6th gear told me and, on top of that, spent a bit more money or settled for buying used parts, but I guess the extra hassle and my curiosity to test the difference with the RAM weighted in heavily on my decision. If I took the wrong choice, I still have a better PC now and added new feedback on this matter so I guess life goes on.

 

Doing more tests, I have to say that performance can unfortunately still get quite stuttery at times, I did tests in Nurburgring at sunset as I've read it's one of the worst case scenarios, and it hits harder than earlier tests, but I don't even remember playing against 24 AI's (17 visible) in that setting with my former RAM sticks (correct me if I'm wrong, but I suppose 24/17v opponents is more CPU taxing being AI compared to multiplayer), so probably it would have been borderline unplayable if not unplayable with my older RAM with the 60 fps target.

 

I have upgraded my 4G 960 to a 12G rx 6700 xt, and while it's clear that I'm still CPU limited it seems obvious that the lower GPU driver overhead plus the higher video memory (It usually ranges around 7 G in the game) plus the better raw power does a significant difference. It's not suddenly a perfectly locked 60 fps experience (my target, I have a 60hz monitor an a 9 year old i5, aiming higher here would be nuts), but it's definitely better. I don't know when Nvidia will improve their drivers regarding this, but they definitely should.

 

To add more info about it, that seems to be correct from the benchmarks I've looked into and my recent experience with the 6700 xt, the delta performance difference I encounter CPU wise seems to be related to how Nvidia and AMD handle the draw calls with the CPU. For games running DirectX 11 or older, usually not very well threaded, Nvidia has the potential to get more draw calls at the expense of more CPU power needed (even for the same FPS target), and for newer and/or low level, very well threaded APIs such as DirectX 12, Nvidia approach adds that extra CPU requirement vs AMD method for little to no improvement at all.

AC: Competizione seems to be a decent threaded game for being a DirectX 11 title, although probably not the best as it seems it doesn't scale that well over 4 or 6 threads, so this game probably enters the category of: AMD driver is good enough to run it well but Nvidia driver is better if you have enough CPU power to carry that extra power requirement, and in my case, obviously I don't have CPU power to spare, so the AMD driver plus the GPU upgrade makes up the difference.

Ideally, GPU manufacturers would let you choose which approach you want to use to get the best performance for each scenario, but seems unlikely. I think the closest thing you can do from the Nvidia side to get a closer experience of the AMD approach is to disable threaded optimization on the Nvidia Control Panel, but I briefly tried that with the GTX 960 before changing the card and, while it seemed maybe performance was a bit better, it could've been placebo effect. To be fair I didn't test it much, probably because I've already make up my mind to upgrade the GPU, but any Nvidia user in a CPU limited scenario could try changing that setting to see if it improves performance.

 

Appearently there's also a hardware side in the different approaches, so surely that can't be changed with drivers in current and older GPUs.