Question Terrible games performance but excellent benchmark performance ?

[jdgotts]

Hey everyone, I need some help here.

Specs:
Zotac 3080 LHR
10700k
32GB 3200MHz DDR4
850W PSU

GPU and CPU are untouched, overclocking wise.

Went on a 2 week trip earlier this week. Prior to the trip, PC worked as expected. Back from trip, gaming performance is bad. Like, truly terrible. I get 35fps in counter strike, when I used to easily push 300 at 2560x1440. This is repeatable for all games, but some seem more affected than others. For example, Battlefield 1 gets ~80fps at ultra settings, which, while lower than before, is far better than 35 in counter strike of all things. Still, no games are performing truly how they used to.

Here's my troubleshooting steps so far:
I immediately pulled out furmark and ran it, only for my card to get completely average results when compared online. Cinebench R13 for the CPU was the same. I performed a furmark test simultaneously with prime95 running to prove it wasnt a total available power issue, and both ran essentially the same as when tested alone, so I've ruled out my PSU being insufficient, or the GPU not getting "good" power or something like that.
I've done a complete DDU display driver reset, and then put on NVIDIA drivers from about 2 months ago, which performed the same in both benchmarks and games as I'm currently experiencing. I've since put the most recent drivers back, as downgrading proved to do nothing helpful.
I've gone through my entire windows even logs for system and application looking for warnings and errors, but get nothing but DCOM warnings and other unrelated things that have been recurring for years.

This potentially seems similar to these issues:

https://www.nvidia.com/en-us/geforce/forums/geforce-graphics-cards/5/525455/rtx-3080ti-performance-loss-after-reboot/

https://www.nvidia.com/en-us/geforce/forums/geforce-graphics-cards/5/472087/low-fps-on-every-reboot-until-reinstalling-drivers/3153034/?topicPage=2

Need to reinstall NVIDIA drivers after every reboot otherwise extremely low FPS

I have the following issue for about 10 days now: (Specs (laptop): Intel Core i7-11800H, NVIDIA GeForce RTX 3060, 16 GB RAM, Win 10). Every-time I reboot my computer (turn off and on), none of the games I have are playable (10 FPS). I then go and uninstall my drivers from the device manager and i...

linustechtips.com

However the largest difference is that my benchmark tests work completely fine. I'm on the lower end of average on the furmark and unigine leaderboards, but the zotac 3080 wasn't known for performance anyways, and I haven't overclocked at all. With the games performance I'm getting, I'd have to have like 3rd percentile performance, and I don't. During the stress tests and benchmarks, the card will have a full ~98% power draw, 99% utilization according to GPU Shark. When in a game, however, I'm getting something like 30% power draw, 100% utilization, and GPU Shark informs me that it is currently being limited by the GPU Voltage limiter, whereas during the stress tests or benchmarks, it will say it is being power limited.

It's also really interesting that with a combination of restarts and stress tests and checking game performance, there have been moments when games perform as expected. Following a system restart, they will go back to poor performance, until I get lucky again. That has happened probably three times amidst 30+ restarts and benchmarks and settings changes and control panel openings and driver installations.

Any advice or knowledge is massively appreciated.

EDIT: adding in a link to my userbenchmark score, as that also shows high performance. After getting that score, I went and tried playing Counter Strike, Deep Rock Galactic, and Battlefield 1, all of which performed poorly.

Please click the green human to continue - UserBenchmark

 

[TerryLaze]

Just for testing I would disable the iGPU from bios just in case something happened and games try to start on that.
Also have a look at windows power settings and put it on max performance, do the same for the nvidia control panel.

 

[jdgotts]

Yeah I considered the iGPU route for a bit, but honestly the performance that the games are getting at 2k resolution, while way worse than what I expect from my 3080, is beyond what the iGPU would be able to do. I have it disabled in bios either way, so it doesn't even show up under display adapters in device manager.

Power settings, as far as I can see, are all set to high/max. Under graphics settings, I tried out disabling hardware accelerated scheduling, as well as adding a game exe and manually telling it to run that game at maximum performance. Neither made an impact.

Also updated bios to the latest version, just for completeness' sake.

This feels like such a strange issue, because as far as the GPU is concerned, the game should be identical to the graphical stress testing; they're both just things to render. I have no idea how a variety of gpu stress tests (furmark, userbenchmark, unigine) all get very solid results, a variety of cpu tests do the same thing, even simultaneously with the gpu tests, but games just cannot utilize the gpu at all.

Another interesting bit is that the performance of the games doesn't change at all with resolution. CS:GO at 720p runs at 40fps, just like at 1440p. There's absolutely no setting I have that is limiting the performance, but it sure appears to be getting bottlenecked somewhere?

 

[Bob.B]

Yeah I considered the iGPU route for a bit, but honestly the performance that the games are getting at 2k resolution, while way worse than what I expect from my 3080, is beyond what the iGPU would be able to do. I have it disabled in bios either way, so it doesn't even show up under display adapters in device manager.

Power settings, as far as I can see, are all set to high/max. Under graphics settings, I tried out disabling hardware accelerated scheduling, as well as adding a game exe and manually telling it to run that game at maximum performance. Neither made an impact.

Also updated bios to the latest version, just for completeness' sake.

This feels like such a strange issue, because as far as the GPU is concerned, the game should be identical to the graphical stress testing; they're both just things to render. I have no idea how a variety of gpu stress tests (furmark, userbenchmark, unigine) all get very solid results, a variety of cpu tests do the same thing, even simultaneously with the gpu tests, but games just cannot utilize the gpu at all.

Another interesting bit is that the performance of the games doesn't change at all with resolution. CS:GO at 720p runs at 40fps, just like at 1440p. There's absolutely no setting I have that is limiting the performance, but it sure appears to be getting bottlenecked somewhere?

Run userbench like this.
Reboot
Wait a few mins
Run ubm with the browser closed.
Post a LINK to the results page.

 

[scout_03]

check your sdd firmware might need to be update also full name and model of ram did you compare the spd and memory tab in cpu-z for settings .

 

[jdgotts]

Run userbench like this.
Reboot
Wait a few mins
Run ubm with the browser closed.
Post a LINK to the results page.

So here that is: https://www.userbenchmark.com/UserRun/63781418

Ignore the NVME drive performance, the poor 980 pro is in an 3.0x4 slot and the other one is in a 3.0x2 slot due to my mobo, so they're getting gimped by that.

HOWEVER, HUGE NEWS LADS, I FOUND A REASON:

memory tab in cpu-z for settings .

scout, you telling me to open cpu-z made me check the graphics tab. Which shows that my GPU is running at a bus interface of not PCIe x16 4.0, but instead.... x1 3.0. Which is....much lower.

How the hell did my card suddenly stop working with x16 in its bus lane? I'm going to open up my case and unseat/reseat the card, make sure the contacts are ok, etc, will update in a bit.

 

[Hotrod2go]

So here that is: https://www.userbenchmark.com/UserRun/63781418

Ignore the NVME drive performance, the poor 980 pro is in an 3.0x4 slot and the other one is in a 3.0x2 slot due to my mobo, so they're getting gimped by that.

HOWEVER, HUGE NEWS LADS, I FOUND A REASON:

scout, you telling me to open cpu-z made me check the graphics tab. Which shows that my GPU is running at a bus interface of not PCIe x16 4.0, but instead.... x1 3.0. Which is....much lower.

How the hell did my card suddenly stop working with x16 in its bus lane? I'm going to open up my case and unseat/reseat the card, make sure the contacts are ok, etc, will update in a bit.

Does your bios allow you to force it to 4.0 x16 speed? Sometimes auto settings on boards just don't work properly.

 

[jdgotts]

New userbenchmark: https://www.userbenchmark.com/UserRun/63782840

So I got the card running at x16 3.0. I don't believe that my mobo, the asus prime z490, supports 4.0 at all after checking its specs, so that seems like full speed for me.

I unseated the card and reseated it and got x8 3.0 performance. I then unseated the card, and put it into the 2nd PCIe slot on my mobo, and got x8 3.0 performance. I then unseated the card, cleaned the contacts with alcohol, and got x8 3.0. Then, I gave up, placed the card back into the original 1st slot, and got x16 3.0. Funny how that works, right?

I have low hopes of it lasting for a long time after the fix itself was so incredibly finicky, but at least I know where to look for poor performance on my system in the future. Based on the card getting x8 on two different mobo slots, I believe that points to the issue being the card and not the mobo. If this issue returns, I'll have to get back to troubleshooting.

Thanks everyone for the help, I might've never found out about the bus performance without checking GPU-z again. I'm going to mark this as solved.

 

[SkyNetRising]

I then unseated the card, and put it into the 2nd PCIe slot on my mobo, and got x8 3.0 performance. I then unseated the card, cleaned the contacts with alcohol, and got x8 3.0.

Second PCIE x16 has only x8 connectivity. This is normal.

Then, I gave up, placed the card back into the original 1st slot, and got x16 3.0. Funny how that works, right?

Conclusion:
1. Clean contacts on card.
2. Insert back into top X16 slot.

Based on the card getting x8 on two different mobo slots, I believe that points to the issue being the card and not the mobo.

It can be card,
it can be motherboard,
it can be bent pins in cpu socket.

Since you got it working again, then most likely is was dust in pcie slot.

 

[jdgotts]

it can be motherboard...

The reason I think it's my card is that my mobo manual denotes all the PCIe slots as x16 slots, instead of one x16 and 2 x8 slots. I guess the manual could be wrong, but that's why the x8 performance in the first slot, then the second slot, then finally x16 back in the first slot seemed completely random.

 

[Bob.B]

So here that is: https://www.userbenchmark.com/UserRun/63781418

Ignore the NVME drive performance, the poor 980 pro is in an 3.0x4 slot and the other one is in a 3.0x2 slot due to my mobo, so they're getting gimped by that.

HOWEVER, HUGE NEWS LADS, I FOUND A REASON:

scout, you telling me to open cpu-z made me check the graphics tab. Which shows that my GPU is running at a bus interface of not PCIe x16 4.0, but instead.... x1 3.0. Which is....much lower.

How the hell did my card suddenly stop working with x16 in its bus lane? I'm going to open up my case and unseat/reseat the card, make sure the contacts are ok, etc, will update in a bit.

When you run out of things to do you might to look at why your ram is running slow.

 

[jdgotts]

When you run out of things to do you might to look at why your ram is running slow.

LOL, I swear every component in my pc has an asterisk next to it, except maybe the CPU at this point. The ram is a 3200MHz kit, but can't run at that speed stably any more. About a year ago I started having huge stability issues while playing games, like constant blue screens, and as I went through the troubleshooting, it turns out my PC failed memtest86. I took 200MHz off the ram and I stopped having issues. It's probably something I could've spent more time on, but the 200MHz didn't make too much of a difference, and I haven't had issues since.

It'll get worked out when I upgrade next, but I'm hoping I can make it a year or two more before that happens.