Question Poor performance on gaming PC

[MrSquid_]

Hello,

So I recently bought a second-hand gaming PC that was custom built in 2018 :
ASRock Fatal1ty X99M Killer Intel i7-5820k (3.3 GHz, 6 cores) Corsair liquid cooling H100i GTX Asus GTX 1080 Ti Turbo (Driver : 461.57 | 27.21.14.5167) 16 GB CORSAIR Dominator Platinum 2666 MHz (4x4 GB) M.2 SSD 960 Evo (500 GB) SSD 850 Pro (512 GB) PSU: Corsair AX760i (760W) Case : Corsair Carbide Air 240 Windows 10 Pro (10.0.1941)

I'm currently using a random Dell FHD monitor (using display port) from work as this is all I have for now.
I ran some benchmarks (3DMark Free, UserBenchmark Free, Heaven Benchmark 4.0 Free) and the results are quite disappointing.

3DMark

Following : My overall score on Time Spy compared to the most similar configurations, from January 1st, 2020 to the present day, as to reduce the impact of driver and OS versions.

Spoiler: 3DMark distribution of most similar setups

As seen in this test, I'm a statistical outlier, on the wrong side of the distribution. I repeated a few times with almost identical results. In fact, 3DMark says my score lies between an office PC and a gaming laptop, whereas I was of course hoping to score between a gaming PC and a high-end gaming PC.
Following are some screenshots of some plots from MSI Afterburner took shortly after the test, first for the GPU

Spoiler: MSI Afterburner GPU related graphs

As seen, GPU temperature (first from top) seems quite fine, utilization and clock speed (resp. second and fourth from top) too as we see both graphic tests using 100%.
Next : MSI Afterburner screenshot for CPU, more specifically the CPU temperature (second from top) and overall CPU usage (third from top)

Spoiler: MSI Afterburner CPU related graphs

The spike from the CPU test is clearly seen, both in the usage and temperature. However the temperature seems to be quite high given the h100i liquid cooler... At this point it should be noted that I used the "Extreme" performance profile for the cooler, using iCUE (Corsair's management software).

So based solely on this test I don't understand what's wrong. I checked the negotiated BUS interface in GPU-Z and I am indeed using PCIe x16 3.0. I used DDU to wipe and reinstall the nvidia driver with no change in this test results.

Now here's some important piece of information : By default (i.e with no particular overclocking software, using only the nvidia driver), the GPU's fan speed was not adjusting correctly. In fact it seemed to have a maximum speed of 29%, resulting in frequent 85°C temperatures during this test, probably throttling. After clicking the "User defined" button in MSI Afterburner it took off to higher values, resulting in the more moderate temperatures you can see in the first Afterburner screenshot. But it didn't have any impact on my result.
Side note: The previous owner was not using this PC as a gaming rig but only for some CAD software, as he was more of a PS4 person. He told me the computer was indeed very quiet. However once I unlocked the GPU fan speed it actually became quite noisy, which leads me to think the GPU fan might have always been set on this low maximum speed ... That's just a supposition though.

UserBenchmark

My results : https://www.userbenchmark.com/UserRun/30640863
I understand UBM results are very misleading as many use hardcore overclocking but still. They indicate a particularly poorly performing GPU. However I'm not sure that it's the problem as we have seen (at least I think) normal temperature, clock speed and utilization from MSI Afterburner from the Time Spy test. But I'm no expert, that's why I need your insight.

Heaven Benchmark 4.0

Following, my results from the heaven benchmark using the Ultra preset

Spoiler: Unigine Heaven Benchmark on Ultra preset

I don't know much about this test though as I haven't found an easy way to compare with similar setups. From what I've read that's actually not too bad, but you tell me.

"Real life" test

I don't have many games available right now, but I tested the PC on Minecraft with high settings : Almost maxed everywhere except for the render distance (set to about 15 chunks), and the biome blender set to 5x5 I believe (I don't know what it does) and performance wasn't great either. I was frequently dropping beneath 100 FPS, punctually dropping under 60 FPS.

FYI by default my RAM is running at 2066MHz but I'm using an XMP profile from the X99M motherboard to get it to 2666 MHz, as it is advertised to run at this frequency. The previous owner assured me he never overclocked anything. He didn't strike me as a computer enthusiast so I tend to believe him on this point.

So here I am. I think it's safe to say the computer is performing below expectations. I don't know if the GPU has a problem, or if the CPU gets too hot which bottlenecks everything else... That's my two hypotheses so far. Any help would be greatly appreciated.
Sorry for this lengthy post. I tried to make it short but still giving you all the information I have so far. I'm slightly out of depth here.

 

[N. M.]

I may be crazy but I don’t think you should be hitting that high temps on your cpu. I bet you’re thermal throttling.

 

[MrSquid_]

Hi @N. M. and thank you for you reply. Here is a better visualization of the GPU/CPU charts right after running the time spy benchmark.

Spoiler: Time Spy comparative charts

Rapid explanation just to be 100% clear on what's going on in this plot:
By looking at the second graph from top (GPU utilization) we see 3 regions of activity. The first is the cinematic you get when starting time spy (it's the largest too). The other two are both graphic tests. Then there's the final CPU test that we can only see on the CPU usage (5th from top).

Now the way I see it:
During the launch cinematic (which I guess is more than just a plain cinematic but you probably know what I mean by that), we see that the CPU temp. (3rd from top) reaches 100°C which of course would cause thermal throttling. However, during both graphic tests, the CPU temp. cools down to settle slightly above 74°C, which I admit might be a bit high, but not enough to trigger throttling I reckon. Finally, the CPU test brings the CPU temp. to 100°C again.

So I'm not ruling out thermal throttling and evidently the temperatures I'm getting are an issue, but it might not be the issue causing the overall very poor performance, especially during the graphic tests, in which I'm running below 30 FPS. But that's just my reading of these graphs I could be mistaking.
Also, if thermal throttling was indeed the issue, could it be possible that the UserBenchmark identify only my GPU as performing below expectations ? Evidently their test is not bullet proof seeing as it barely even takes a minute to run, but I'm wondering.

Side note: We see on the CPU clock graph (2nd from bottom) that the overall CPU frequency is the most consistent when the CPU temp. is at or near its highest. I don't quite know how to reconcile that with the thermal throttling I'm probably experiencing during these moments.

 

[DSzymborski]

Did you do a full fresh install of Windows? That's typically the first thing to do on a used PC that you acquired. Then go about updating all the various drivers, from the motherboard manufacturer and Nvidia. Don't use any type of magic driver software.

Also, at this point, you don't reconcile the frequency and thermal throttling. The benchmark issues are a case of nausea; the temperature issues are a knife in the belly and need to be resolved immediately before you even think about additional issues.

 

[MrSquid_]

Hi @DSzymborski ,

I did not do a full fresh install, Windows was factory reset for the sale. Then I used DDU to remove the nvidia driver before reinstalling it. I was planning on doing a fresh install but I wanted to check the performances before so. The drivers installed after the factory reset were from the manufacturers though.

But yeah either way I need to fix the CPU temperatures issue. I just hope that I won't end up needing to replace the GPU as well.

 

[MrSquid_]

Hi,

So I opened a new thread for the temperature issues here and long story short, the AIO was dead so I got a new one. CPU temps are good now. Room temperature of 28°C, OCCT CPU stress test (large dataset) brings it at 55°C max, otherwise core temps are within the 29-35°C range IDLE. No CPU overclock for these values. So thank you for bringing this to my attention.

Regarding the GPU, I've monitored core clock, memory clock, fan speed, GPU temp, negotiated BUS link. I compared with the constructor's values and everything is as it should be. For me it means the GPU doesn't have any hardware fault. I guess it is possible that the UBM and TimeSpy results are completely worthless since I'm not overclocking the GPU, but I don't know. I still feel like I'm a bit far from the average. The new AIO didn't improve my TimeSpy results btw.

What do you guys reckon? Is there any other parameter I could check of other means of verifying this? Or is being in the orange in UBM expected when not overclocking?
If not, is there any chance it could be because I'm using an FHD monitor? I've heard about CPU bottleneck which hits harder on "low" resolutions, but that seems a bit extreme to me.