Question Strangely terrible 4090 performance ?

[matt91]

Hey all,

I was hoping to get some advice - I've upgraded from a 2080 to a 4090 with the intention to upgrade to a 4k monitor, but am seeing fairly terrible FPS even at 1440p which is concerning. I've had a look at previous 4090 performance threads but with little success.

System:
CPU: AMD Ryzen 5 5600X
RAM: 2x16GB DDR4 3600MHz (XMP enabled)
Motherboard: MSI MAG X570 TOMAHAWK WIFI
GPU: MSI GeForce RTX 4090 GAMING X TRIO 24G (using Phantek vertical mount + PCIe 4.0 riser cable)
PSU: Corsair RM850e 850W
Double monitor setup - 1080p 60hz and 1440p 165hz

What I'm seeing
My expectation was seeing performance vaguely equivalent to these benchmarks for 1440p: https://www.techspot.com/review/2544-nvidia-geforce-rtx-4090/. The system described is fairly similar to mine - we have same generation Ryzen with little difference in performance until you hit >8 cores. I have slightly faster memory, but not expecting that to make a huge diff here.

However I'm not getting close - a couple of examples I'm seeing (at 1440p highest settings):

- Hunt Showdown: benched at avg 329fps, I'm barely scraping 150fps
- CPU was not being capped on any cores - avg usage at ~50%, max of any core at ~60%
- GPU sitting at about ~40% constant util
- Tried shifting this down further to 1080p - for basically the same FPS

- Hitman (more CPU bound): benched at avg 203fps, I'm getting 65fps!?
- CPU not capped on any core ~ 70% across all cores
- GPU is at 95%!?

Fullscreen vs fullscreen windowed has made no difference.

Things I've tried so far
- Confirming XMP is enabled for any RAM bottleneck
- Turning off all GSync & VSync, any other frame caps - both in NVIDIA control panel and in-game options
- Flashing the bios to the latest version
- Full DDU wipe in safe mode and driver reinstall
- Updating the chipset drivers
- Benchmark to confirm the card isn't actually dodgy - FurMark score of 34749 (at 1080p), looks comparable to other scores online
- Setting the NVIDIA Power Management Mode to maximum performance

Almost always a CPU bottleneck is blamed, but it doesn't look like the case here - 65fps at 1440p is pretty shocking. Not really sure where to go next beyond a full Windows reinstall and pray. Maybe trying it directly in the PCIe to rule out the vertical mount? Feels like a long shot.

Any insight & ideas much appreciated!

 

[Moonstick2]

Trying to eliminate the obvious, but your riser cable is plugged into PCI_E1 (the upper slot), not PCI_E3 (the lower slot), correct?

Although they're both physically PCIe x16 slots, the lower slot only supports x4 mode.

 

[DSzymborski]

Have you tried removing the riser cable and the second monitor completely? When diagnosing things like this, it's important to simplify as much as possible. And after that, I'd also try to benchmark compare to something in which the *exact* settings can be determined and reproduced. Techspot's article is all over the place in how detailed they describe the ray tracing settings they use for any particular game.

 

[Phaaze88]

Report back with gpu parameters in game. Between Gpu-Z and hwinfo, Gpu-Z is probably the easiest to look at. Screenshot the Sensors window, but set all the parameters to Max, noted in green font.

 

[matt91]

Thanks for the replies!

Trying to eliminate the obvious, but your riser cable is plugged into PCI_E1 (the upper slot), not PCI_E3 (the lower slot), correct?

Yep - it's using the top PCI_E1 slot. Looking at GPUZ it also looks like it's correctly in x16 mode.

Have you tried removing the riser cable and the second monitor completely?

Just tried this now in steps, sadly the same performance with both omitting the riser cable, and then taking the second monitor out of the picture.

I've found similar bench/gameplay on YouTube (e.g. this) which feels slightly more realistic - avgs of 220 with some lows of ~180, but this still feels like a good 50% improvement of what I'm currently seeing.

Report back with gpu parameters in game. Between Gpu-Z and hwinfo, Gpu-Z is probably the easiest to look at. Screenshot the Sensors window, but set all the parameters to Max, noted in green font.

Attached a few screens of this, plus the graphics card info and OHM taken during gameplay:

 

[matt91]

Reinstalled Windows to no avail - still the same performance and behavior where neither GPU or CPU are technically 100% bottlenecked, but it does look like it could be CPU related?

Ran 3DMark Firetest and saw interesting and similar results. The graphics test get great results - the physics less so, with the worst being the combined, which sees similar results to what I see in actual gameplay - ~50% GPU util, ~60% avg high core utils, with low FPS of ~60:

Firemark (pg1): https://ibb.co/9tMvFYC
Firemark(pg2): https://ibb.co/VJSRX8t
OpenHardwareMonitor CPU per core: https://ibb.co/9ZDW2FF

 

[Moonstick2]

Try running your benchmarks/games and monitoring the CPU to see what clock speed it reaches when boosting.