squark Posted September 14, 2018 Share Posted September 14, 2018 I've recently swapped from a 4790k to a 2800x. Even when dedicating half the cores/threads to a single vm, I'm unable to get the frame rate for fortnite above 120fps. In heavy parts of the map, it struggles with 60fps. On native, these parts run at around 140fps and the rest around 250. The gpu isn't hitting max utilization (around 30-50%) and I think it's being cpu bottlenecked, however cpu usage sits around 60% so I'm confused as to what is causing this lack of performance. Quote Link to comment
david279 Posted September 14, 2018 Share Posted September 14, 2018 I think you mean 1800x. First gen ryzen has a problem on unRAID where the CPU won't boost correctly. You might wanna oc it to get it where it's supposed to boost. Or try out 6.6 rc and see if it may be fixed there. 2nd gen ryzen does not have this issue. Quote Link to comment
squark Posted September 14, 2018 Author Share Posted September 14, 2018 Sorry, I meant 2700x, however I'm running it on a x370 taichi board. Quote Link to comment
david279 Posted September 14, 2018 Share Posted September 14, 2018 Try setting GPU to maximum performance in the Nvidia control panel/3D settings. I have a 2700x as well with a 1070 in a windows VM and it plays just fine, and at 1440p at that. Download msi afterburner and make sure the GPU core and memory is where it should be. Quote Link to comment
Symon Posted September 14, 2018 Share Posted September 14, 2018 Im experiencing similar problems with games that are CPU dependent (World of Warcraft in my case). The performance of the CPU seems to be the bottleneck (I have a Threadripper 1950 OC to 3875 Hz and a 1080 ). I've tried a lot of things with CPU pinning / isolating but nothing really helped to improve the fps. The game still runs like it's palyed on a PC from 5 years ago :-) . However, this game is also known to be very CPU dependent and not being very well optimized for newer CPUs... Quote Link to comment
testdasi Posted September 14, 2018 Share Posted September 14, 2018 8 hours ago, squark said: I've recently swapped from a 4790k to a 2800x. Even when dedicating half the cores/threads to a single vm, I'm unable to get the frame rate for fortnite above 120fps. In heavy parts of the map, it struggles with 60fps. On native, these parts run at around 140fps and the rest around 250. The gpu isn't hitting max utilization (around 30-50%) and I think it's being cpu bottlenecked, however cpu usage sits around 60% so I'm confused as to what is causing this lack of performance. What cores did you pin? 4-7 + 12-15? If so try Pinning 2-3,6-7,10-11,14-15 Keep 4-7 + 12-15 but put your GPU on a different slot If I understand the diagram correctly (https://en.wikichip.org/w/images/7/75/amd_zen_octa-core_die_shot_(annotated).png) each die has 2 CCX with each having 16 PCIe lane connected to it. So it's quite possible that the bottleneck isn't CPU or GPU but latency having to constantly jump over a CCX to get to the GPU. 3 hours ago, Symon said: Im experiencing similar problems with games that are CPU dependent (World of Warcraft in my case). The performance of the CPU seems to be the bottleneck (I have a Threadripper 1950 OC to 3875 Hz and a 1080 ). I've tried a lot of things with CPU pinning / isolating but nothing really helped to improve the fps. The game still runs like it's palyed on a PC from 5 years ago 🙂 . However, this game is also known to be very CPU dependent and not being very well optimized for newer CPUs... Threadripper is quite different from Ryzen in term of optimisation. Try spreading your pins out across all 2 dies and 4 CCX (something like 3,7,11,15,19,23,27,31). Alternatively, pin only the cores that are connected directly to your GPU. The latter will give you best theoretical performance but does involve a bit of trial and error. The former at least reduces some jumping and easy to do and test. Quote Link to comment
Symon Posted September 14, 2018 Share Posted September 14, 2018 32 minutes ago, testdasi said: Threadripper is quite different from Ryzen in term of optimisation. Try spreading your pins out across all 2 dies and 4 CCX (something like 3,7,11,15,19,23,27,31). Alternatively, pin only the cores that are connected directly to your GPU. The latter will give you best theoretical performance but does involve a bit of trial and error. The former at least reduces some jumping and easy to do and test. I already used manajro as recommended by 3flappp to identify which die my GPU is connected to. Unfortunately, there wasn't much improvement in this for the Game performance. However, I still dont know if I can optimize the RAM for this as well. I think they are talking in this thread about it. I think the biggest improvement can be achieved by rising the frequency of the CPU to a higher level which can only be achieved right now through overclocking the base CPU frequency. Unfortunately, I'm worried that my system will overheat / get unstable. It would be nice if this would get fixed so the internal frequency governor would work properly in UnRaid for Ryzen / Threadripper. Quote Link to comment
squark Posted September 14, 2018 Author Share Posted September 14, 2018 I tried a combination of different pinnings and it seemed to improve consistency, however the performance isn't quite there. I installed some stuff to figure out what package is best for me to run cpu stuff on for my gpu, but wasn't quite sure how to interpret it. It seems I should be doing all latency cpu stuff on the first package? thebeast-diagnostics-20180914-2309.zip Quote Link to comment
squark Posted September 14, 2018 Author Share Posted September 14, 2018 (edited) Running the vm on 0-7 with unraid pinned to 8-9 has almost resolved it. The performance isn't quite 100%, but runs at 80 fps at worst and normally around 140 fps which is far better than it was initially. GPU load is around 95% once the textures have finished loading in which would probably indicate this is as good as I'm gonna get performance wise i would imagine. Its still lacking 30% of native, but is playable. Edited September 14, 2018 by squark Quote Link to comment
david279 Posted September 14, 2018 Share Posted September 14, 2018 Running your system using the performance CPU governor help a bit as well. I notice my CPU cores park at 4.1 to 4.2 when i use that governor. Quote Link to comment
testdasi Posted September 14, 2018 Share Posted September 14, 2018 32 minutes ago, squark said: Running the vm on 0-7 with unraid pinned to 8-9 has almost resolved it. The performance isn't quite 100%, but runs at 80 fps at worst and normally around 140 fps which is far better than it was initially. GPU load is around 95% once the textures have finished loading in which would probably indicate this is as good as I'm gonna get performance wise i would imagine. Its still lacking 30% of native, but is playable. How do you pin unRAID to 8-9? Quote Link to comment
squark Posted September 14, 2018 Author Share Posted September 14, 2018 isolcpus=0-7,10-15 in the syslinux config Quote Link to comment
testdasi Posted September 14, 2018 Share Posted September 14, 2018 That's interesting because when I tried to pin core 0, unRAID still uses it anyway. Quote Link to comment
squark Posted September 14, 2018 Author Share Posted September 14, 2018 I looked into core specific monitoring cause it seemed like I was having <1% usage on some of the pinned cores with docker disabled but I don't remembers whether it was core 0. Quote Link to comment
squark Posted September 16, 2018 Author Share Posted September 16, 2018 After putting playtime in, the performance is less than 50% of native so I guess I'll be rebooting into Windows Quote Link to comment
zaptrem Posted November 12, 2020 Share Posted November 12, 2020 I'm also encountering this issue with an i7 10700k and 3080. I didn't notice it before since I had a 1070. I'm not using any CPU pinning, just the exact thread/socket/core counts as native. Quote Link to comment
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.