The ACC GPU Benchmark Thread

I always thought they were about on par, if you take the 6GB 1060.

But it's correct - Radeon underperforms in ACC, which is mostly down to UE4. You're a bit late to the party here ;-)

 

Are you going to add my results to OP ?

 

Yes, of course. I'm doing this in ~2 week turns, though. It'll be in, soon.

 

Post 1 updated. I changed the bar graph. The bars now show the average of all test runs with the same chip. The individual test runs are in the table (including yours, @Stains).

 

Thanks Mate.
It's weird how lots of reviewers have said the 5700xt is on par with the 2070 super which is on par with the 1080ti, the 2070s and 1080ti seem to be pretty equal but the 5700xt certainly seems to be quite a bit slower.
I really hope next AMD GPU's punch up more but by then nVidia will come out swinging again. AMD will always be trailing it seems!

 

Keep in mind the preconditions of this test:

• UE4 "naturally" favours NVIDIA.
• In ACC, GPU tweaking (undervolt/overclock/settings in driver) make a a significant difference to the FPS performance. Proper tweaking on my Vega gained me ~15 %.
• We don't have many test runs per GPU chip so the bar graph isn't (yet) 100 % representative. My Vega made 80 FPS AVG, Bernie's did a very low 56 FPS, possibly due to a CPU bottleneck.

AFAIK, in other titles, 5700 XT is indeed on par with the 2070 super and it's cheaper. So depending on the titles you wanna play, 5700 XT can be a very good choice.

But I'm also hoping that AMD will offer a good top-end GPU, soon.

 

5700 xt smashes the 2070 in other tests so not sure why its behind the 2060

 

Ryzen 3600 Rtx 2080 ti 16 gigs ram
2019-09-15 15:56:04 - AC2-Win64-Shipping
Frames: 9985 - Time: 60000ms - Avg: 166.417 - Min: 132 - Max: 218

 

Hey guys, could you re-run the benchmark? I saw reports of FPS improvements in VR with the latest update. Thanks.

 

VR definitely improved. Will rerun

 

so the 2080ti vs 1080ti is 100% more expensive for +25% perf ?? I will wait for the next gen....I will have +50% perf for the 3080ti....

 

CPU 9900k 5.0GHz custom LC
RAM 32GB 4000/CL17
NV drivers 436.30
ACC Version 1.0.10

GPU EVGA FTW3 RTX 2080Ti 2130/15500 (on air)
GPU Asus FE GTX 1080Ti 2040/11800 custom LC

2080Ti
2019-09-19 21:27:04 - AC2-Win64-Shipping
Frames: 8398 - Time: 60000ms - Avg: 139.967 - Min: 109 - Max: 225

1080Ti
2019-09-19 20:59:41 - AC2-Win64-Shipping
Frames: 6456 - Time: 60000ms - Avg: 107.600 - Min: 78 - Max: 181

Looks like 30% difference in avg fps and 39.7% in min fps, which is well above average for other games and Turing vs Pascal comparision.
Recently tested 20 games, comparing both cards and came out that 2080Ti is faster in avg fps:

• 22,7% in 1080p
• 36,7% in 1440p
• 37,3% in 2160

Same bench in different resolutions.






Comparing to other games on UE4: Moto GP 18, GRip, Gravel, MXGP Pro - ACC does not look bad.
Let's look at the latest Coalition/Microsoft hit Gears 5 on DX12.
Considering the amount of physics in racing game I think Kunos did a good job.

 

nice work on comparisons!

 

That driver is bad for Pascal.
Your numbers still don't justify and upgrade for me. I hope for AMD to get to the high end and Intel to jump into the fight.

 

Where can I read more about this?
I'm on 431.60 and always wary of upgrading when I find a good driver... I usually stick with it for a 3-6months...

 

I always go here when new drivers drop and check whats peoples impressions

https://www.reddit.com/r/nvidia/comments/d27ogr/game_ready_driver_43630_faqdiscussion/

This is the summary from the dude who does benchmarks every driver when they release

"Driver performance testing

DX12 Performance on The Division 2 seems this time slightly better.

Unfortunately all Dx11 tests have seen a performance regression in this driver. Every metric in every game is worse on 436.30 than on 436.15. FC5 sees the worse drop of the tested games."

Im still on 430.86 as it seems to be the last driver that worked ok on Pascal .

 

Don't speak too soon. We have a recent 2080Ti result that looks much better:

I don't know why this one is so good. It was achieved with a Ryzen, the other 2 in the 130 AVG range were achieved with Intel i7s. Next time I run the benchmark, I'll check for GPU usage. We

In general, I suggest to check the table for details - as I explained earlier, the graph is not (yet) representative. We need a lot more runs.

 

I wouldn't say it's bad for Pascal, but rather that 436.02 brings significant improvements to the Turing architecture and in some titles newer cards clearly step forward.
If You refer to the statement that "NVIDIA reduce performance for older cards" it's a myth that appears on forums, urban legend. Since I monitor performance, there has never been a situation where the previous generation lost performance. Optimisation, engine development, API development, better usage of new features, which means that new generations always increase performance over time, not the other way around. I base my experience on testing 780Ti, 980Ti / Titan X (Maxwell), 1080Ti cards in the last 5 years.

True, Turing hasn't made clear progress in traditional rasterization probably at the expense of adding tensor cores on the PCB.
This can be seen by comparing the average performance increment over previous gen's.

1440p res.
Kepler 780Ti to Maxwell 980Ti/TX → 56%
Maxwell 980Ti/TX to Pascal 1080Ti → 68%
Pascal 1080Ti to Turing 2080Ti → 36%

 

I dont know but i just did this ........with a little help from NotePad ++
Just maybe...

 

What is the consensus to use for a target frame rate? I’ve got a GTX1070 with a 4K G-Sync monitor and the 50’s seem acceptable. Just curious what everyone else is shooting for.