Power supply check

Yeah running all cores on max turbo clocks will definitely help a bunch in certain games.

I actually made a mistake earlier. When I got 80fps on Ultra. I had TAA turned on - which led to the 60-70% gpu usage. When I turned off TAA, fps jumps to 120fps and gpu utilization goes to mid 80%. Now the sample video gameplay I found for 1080ti sli has TAA turned on and 90% CPUs utilization. That’s throwing me off. How does anti aliasing increase gpu usage to the point of bottlenecking? Is that possible?

120fps on 4K ultra and 85% scaling i would say that you do pretty well.

Hmm yeah actually. It looks like battlefield’s TAA is what causes sli to mess up. Like you said scaling might actually be fine for what it is. But TAA (and certain types of anti aliasing are CPU intensive) and introduces compatibility issues with sli. Oy! SLI is a mess.

1 Like

I want to share something that’s quite interesting.

I turned of HyperThreading (no impact on performance). I also utilized window’s core affinity feature for running tasks and forced battlefield 1 to only use 12 cores which I also configured bf1’s user.cfg to also use 12 cores and the screenshot shows that all cores average about 80 - 85% while GPUs also Avg 80-85%. The spike with cores above 12 are caused by taking a screenshot. The real noticeable difference though is that framerate is rock steady vs before where it would dip from time to time. Quite interesting to get the most out of a bottlenecking set of CPU’s.


(TAA Turned off)

2 Likes

Seems like a numa problem - the dips are probably when memory was accessed from across CPU’s.

1 Like

I also have numa turned off.

It looks like the GPU utilization issue might be due to Battlefield 1 and not the system.

I ran unigine heaven benchmark on ultra with extreme tessellation and gpu’s were at 99%.

and the score I got is a few points behind someone else on youtube with a non-bottlenecking cpu. https://www.youtube.com/watch?v=B8gcVT3n0Ps

Yeah well benchmarks tools like unigen are generally made for,
stressing the components.
Its nice to use as a compairisson to similar system configurations.
But its not really a benchmark for games in general.
Because all games kinda differ in the way they utilize your hardware so to speak.
So yeah there will be games that are really optimized well.
BUt that are also games that are poorly optimized.
So yeah your experience will vary from game to game really.

its probablly just a matter of the said game being cpu intensive.
And with 2 1080Ti´s you cpu´s might exally have some issues keeping up with feeding those cards, cause of the lower clockspeeds.
Thats most likelly the issue you are running into with BF1.
BUt like i mentioned, this will not be the case with every game out there.
And in my opinion if you still get 85% utilization on both cards with stable frame rates wenn AA disabled.
Then in my opinion your setup does really well if you take the lower clock speeds of the Xeon´s in mind.

1 Like

That’s a really great point that I missed. You’re definitely right, BF1 is a cpu intensive game and there were patches where certain aspects of the game were unoptimized leading to 100% cpu usage. These cpu’s have given all they got and 85% utilization is good enough! until I find a new pair of xeons for cheap :slight_smile:

Holy Crap! I figured out the trouble with my sli bottleneck. CPU might be pushing it, but I realized over the weekend that my GPU are on different PCIe Lanes. GPU0 is on CPU0 and GPU1 is on CPU1 so there is now latency introduced by the QPI Link.

The even more amazing thing is that before with TAA turned on in BF1, GPU usage would go down to 65% but now it’s scaling just fine.

Seriously, though I am really glad you guys replied to the thread, it really helped me with the thought process and come up with troubleshoot methods.

XEONS can game really well.

Glad we could be of service - hopefully this helps one of us in the future as well.

1 Like

I hope it does. At least this gives an idea of some placement for older xeon cpu’s still being able to keep up with newer videocards.

Just making notes for later if anyone glances upon this thread for the same issue.

From an earlier reply by @MisteryAngel regarding the unigine benchmark: absolutely correct on this one. Can’t tell if GPU’s are bottlenecked by cpu or anything in the system through these. They are only meant to stress the GPU’s. As such in the screenshot and score I provided, the two GPU’s were still setup incorrectly with the two split between different PCIe buses controlled by each of the two xeons. It’s only in game where the game becomes apparent since most of these benchmarks use little to no CPU during runtime.

I think a good way to hone CPU usage if you have a lot cores is to find out how many cores the game actually uses. BF1 in this case allows user to manually set. I was able to force it to 12 cores then later back to 6 cores and then use Windows’ feature to set core affinity for a specific running process and set which cpu cores to run. It gives a better read out as the process stays locked. In my case I did not notice any cpu cores going close to 100% utilization. A few reached 80% (at times spiking to mid 90%) which means that this CPU may be on the edge of bottlenecking (assumption).

1 Like