Switching from a 2080ti to a 4090 on a Ryzen 3900x setup yields no uplift. 2080ti also showing signs of not being fully leveraged.

[sumolananas]

Hello world! I'll do my best to keep it short and sweet
The Setup: Win11 + 120Ryzen 3900x + ROG Strix X570-E Gaming + 2x16GB Ram G. Skill Trident Z 3600 CL16 + (RTX 2080Ti) + Corsair RMX850x
3dmark 4y ago when I built the setup: ~14500 (sure, cant truly compare apple to apples with older versions of 3Dmark), but I could see GPU Load all maxed out back then.

Sold a kidney and one eye and bought: RTX 4090 + MSI A1000G PCIE5
Run 3Dmark on the old setup and detected that the performance of my 2080ti was being really capped:
My score ~8500 VS average of the community with same hardware: ~14000
Looking at the pretty graphs from 3dmark:
CPU performance as expected (clock wise and CPU score), and temps never higher than 62c (eyeballing the graph).
Clocks of the 2080ti: Consistent 1750mhz GPU Memory, and close to 2000mhz GPU Core, GPU Load: ~50% (eyeballing the graph), Temps always bellow 60c

Thought to myself something was wrong, but who cares, I have a fancy new toy... Off with the old PSU, off with the old GPU. New PSU, new GPU, sexy and problematic single cable PCI5 cable. Everything boots up, new drivers from nvidia detected. Installed. Run 3dMark and I get... the same score range.. ~8500... GPU boosted a bit to 2800mhz, but seems to be averaging at 1200mhz. GPU Core temp max 48c.
I think to myself, whatever... who cares about synthetic benchs? I boot up No Man's Sky, and I was having around the same FPS I had with the 2080ti! (at 4K, 70ish FPS).

Ah... poorly optimized game.. sure... Cyberpunk.. crank everything up... (high quality than I used to play) and I get around the same FPS I had before! (At least the raytracing eye candy didn't seem to have impact in the performance).

Thats when I went down the rabbit whole:
Was it g-sync? No.
Was it power profile of the computer? No.
Was it the old BIOS? update to latest: No.
Is it the CPU virtualization that I need to run docker containers? No.
Was it the overclocked RAM? No.
High activity of some random program on the background? No.
Swamp the RAM sticks - their position between themselves- No
Windows pending updates? No
Should have removed all the graphic drivers before installing the new one? No worries, nuke windows and install it again: No
Some power limit set? Install MSI afterburner, confirm power limit set to 100%: No
Use MSI afterburner to train the Overclock curve and maybe it will work? No
Was my cat seating on top of the computer? No
Am I leaving performance on the table by pairing a 4090 with an old system? Yes, but I need to wait for my Kidney and eye to regenerate to sell them again (that's how it work, right?... right?)

What makes things more interesting, is that in the miserable score I get, in the distribution graph that shows the scores of ppl with the same core hardware I have, it shows a little bump around my poor score, indicating that there is a fair number of people getting the same kind of scores that I get. Attached you can see the screenshots of the detailed results of 3DMark.
I've also attached the 3dmark result files (if you want to open them with 3dmark to check all the details).
My goal really isn't to use 3dmark as the source of all truth and reality (although I did put most focus on it, cuz it's the only instrumentation I got) but gaming is also showing that something is not ok.

My knowledge, limited experience in debugging these issues, and whatever ability I have in googling things, leave me at thinking that ... maybe the motherboard is having a problem? But it's I don't know what best way to debug from here. Any tool that is able to capture more detailed information on what's happening with the card? Perhaps power delivery? I have no idea where to pick this from.

I'm hopping that this community is able to help me, and all the other 3dmarkers that seem to be having the same problem, and I have a feeling that there might be people on a budget that might be extending themselves so they can enjoy their games, and be facing the same issue. The truth is that over the last 4 years, I went through a backlog of games, and I'm not able to tell when I started having this issue with the 2080ti. Jumping from old games, to AAA games, I believe that I just didn't notice performance drop, and accepted the underutilized reality of my 2080ti, since before this built I had not been playing for 10y.

Thanks for reading all the way here, thanks in advance if you chose to help, and now I'll go curl into a ball and cry myself a river.


 

2080ti - 3DMark-TimeSpy-8651-20230725214529.3dmark-result 4090 - 3DMark-TimeSpy-8675-20230726171831.3dmark-result 2080ti when first built in 2019 - 3dmark-autosave-20190929205015.3dmark-result

[Tan3l6]

DDU tried?

[hollyh88]

36 minutes ago, sumolananas said:

Hello world! I'll do my best to keep it short and sweet
The Setup: Win11 + 120Ryzen 3900x + ROG Strix X570-E Gaming + 2x16GB Ram G. Skill Trident Z 3600 CL16 + (RTX 2080Ti) + Corsair RMX850x
3dmark 4y ago when I built the setup: ~14500 (sure, cant truly compare apple to apples with older versions of 3Dmark), but I could see GPU Load all maxed out back then.

Sold a kidney and one eye and bought: RTX 4090 + MSI A1000G PCIE5
Run 3Dmark on the old setup and detected that the performance of my 2080ti was being really capped:
My score ~8500 VS average of the community with same hardware: ~14000
Looking at the pretty graphs from 3dmark:
CPU performance as expected (clock wise and CPU score), and temps never higher than 62c (eyeballing the graph).
Clocks of the 2080ti: Consistent 1750mhz GPU Memory, and close to 2000mhz GPU Core, GPU Load: ~50% (eyeballing the graph), Temps always bellow 60c

Thought to myself something was wrong, but who cares, I have a fancy new toy... Off with the old PSU, off with the old GPU. New PSU, new GPU, sexy and problematic single cable PCI5 cable. Everything boots up, new drivers from nvidia detected. Installed. Run 3dMark and I get... the same score range.. ~8500... GPU boosted a bit to 2800mhz, but seems to be averaging at 1200mhz. GPU Core temp max 48c.
I think to myself, whatever... who cares about synthetic benchs? I boot up No Man's Sky, and I was having around the same FPS I had with the 2080ti! (at 4K, 70ish FPS).

Ah... poorly optimized game.. sure... Cyberpunk.. crank everything up... (high quality than I used to play) and I get around the same FPS I had before! (At least the raytracing eye candy didn't seem to have impact in the performance).

Thats when I went down the rabbit whole:
Was it g-sync? No.
Was it power profile of the computer? No.
Was it the old BIOS? update to latest: No.
Is it the CPU virtualization that I need to run docker containers? No.
Was it the overclocked RAM? No.
High activity of some random program on the background? No.
Swamp the RAM sticks - their position between themselves- No
Windows pending updates? No
Should have removed all the graphic drivers before installing the new one? No worries, nuke windows and install it again: No
Some power limit set? Install MSI afterburner, confirm power limit set to 100%: No
Use MSI afterburner to train the Overclock curve and maybe it will work? No
Was my cat seating on top of the computer? No
Am I leaving performance on the table by pairing a 4090 with an old system? Yes, but I need to wait for my Kidney and eye to regenerate to sell them again (that's how it work, right?... right?)

What makes things more interesting, is that in the miserable score I get, in the distribution graph that shows the scores of ppl with the same core hardware I have, it shows a little bump around my poor score, indicating that there is a fair number of people getting the same kind of scores that I get. Attached you can see the screenshots of the detailed results of 3DMark.
I've also attached the 3dmark result files (if you want to open them with 3dmark to check all the details).
My goal really isn't to use 3dmark as the source of all truth and reality (although I did put most focus on it, cuz it's the only instrumentation I got) but gaming is also showing that something is not ok.

My knowledge, limited experience in debugging these issues, and whatever ability I have in googling things, leave me at thinking that ... maybe the motherboard is having a problem? But it's I don't know what best way to debug from here. Any tool that is able to capture more detailed information on what's happening with the card? Perhaps power delivery? I have no idea where to pick this from.

I'm hopping that this community is able to help me, and all the other 3dmarkers that seem to be having the same problem, and I have a feeling that there might be people on a budget that might be extending themselves so they can enjoy their games, and be facing the same issue. The truth is that over the last 4 years, I went through a backlog of games, and I'm not able to tell when I started having this issue with the 2080ti. Jumping from old games, to AAA games, I believe that I just didn't notice performance drop, and accepted the underutilized reality of my 2080ti, since before this built I had not been playing for 10y.

Thanks for reading all the way here, thanks in advance if you chose to help, and now I'll go curl into a ball and cry myself a river.


 

2080ti - 3DMark-TimeSpy-8651-20230725214529.3dmark-result 519.53 kB · 0 downloads 4090 - 3DMark-TimeSpy-8675-20230726171831.3dmark-result 713.64 kB · 0 downloads 2080ti when first built in 2019 - 3dmark-autosave-20190929205015.3dmark-result 501.2 kB · 0 downloads

if the gpu load is at 50% without upscalers in games its a clear sign that either 1. your gpu has to do fck all and the game is easy to run (think of minecraft stardew valley etc). or in difficult demanding games it is a clear indication of a cpu bottleneck. in this case yup that it be.

you cannot and i repeat cannot pair a 4090 with a 3000 series amd cpu and expect good results. you should get more fps sure but its being held back so much that it basically seems the same to you. swapping it for a 5800x3d should show immediat performance boosts. but in whole honesty you shouldnt even pair a 4090 with a 5800x3d. cards like that should be paired with the best of the best. 

but still if you do not want to sell more body parts i would highly suggest upgrading your bios and getting a 5800x3d.

and also synthetics are nice but if you want to show how your system is running it is better to do it in games. Its a better indication paired with an overlay.

 

[191x7]

1 hour ago, sumolananas said:

Hello world! I'll do my best to keep it short and sweet
The Setup: Win11 + 120Ryzen 3900x + ROG Strix X570-E Gaming + 2x16GB Ram G. Skill Trident Z 3600 CL16 + (RTX 2080Ti) + Corsair RMX850x
3dmark 4y ago when I built the setup: ~14500 (sure, cant truly compare apple to apples with older versions of 3Dmark), but I could see GPU Load all maxed out back then.

Sold a kidney and one eye and bought: RTX 4090 + MSI A1000G PCIE5
Run 3Dmark on the old setup and detected that the performance of my 2080ti was being really capped:
My score ~8500 VS average of the community with same hardware: ~14000
Looking at the pretty graphs from 3dmark:
CPU performance as expected (clock wise and CPU score), and temps never higher than 62c (eyeballing the graph).
Clocks of the 2080ti: Consistent 1750mhz GPU Memory, and close to 2000mhz GPU Core, GPU Load: ~50% (eyeballing the graph), Temps always bellow 60c

Thought to myself something was wrong, but who cares, I have a fancy new toy... Off with the old PSU, off with the old GPU. New PSU, new GPU, sexy and problematic single cable PCI5 cable. Everything boots up, new drivers from nvidia detected. Installed. Run 3dMark and I get... the same score range.. ~8500... GPU boosted a bit to 2800mhz, but seems to be averaging at 1200mhz. GPU Core temp max 48c.
I think to myself, whatever... who cares about synthetic benchs? I boot up No Man's Sky, and I was having around the same FPS I had with the 2080ti! (at 4K, 70ish FPS).

Ah... poorly optimized game.. sure... Cyberpunk.. crank everything up... (high quality than I used to play) and I get around the same FPS I had before! (At least the raytracing eye candy didn't seem to have impact in the performance).

Thats when I went down the rabbit whole:
Was it g-sync? No.
Was it power profile of the computer? No.
Was it the old BIOS? update to latest: No.
Is it the CPU virtualization that I need to run docker containers? No.
Was it the overclocked RAM? No.
High activity of some random program on the background? No.
Swamp the RAM sticks - their position between themselves- No
Windows pending updates? No
Should have removed all the graphic drivers before installing the new one? No worries, nuke windows and install it again: No
Some power limit set? Install MSI afterburner, confirm power limit set to 100%: No
Use MSI afterburner to train the Overclock curve and maybe it will work? No
Was my cat seating on top of the computer? No
Am I leaving performance on the table by pairing a 4090 with an old system? Yes, but I need to wait for my Kidney and eye to regenerate to sell them again (that's how it work, right?... right?)

What makes things more interesting, is that in the miserable score I get, in the distribution graph that shows the scores of ppl with the same core hardware I have, it shows a little bump around my poor score, indicating that there is a fair number of people getting the same kind of scores that I get. Attached you can see the screenshots of the detailed results of 3DMark.
I've also attached the 3dmark result files (if you want to open them with 3dmark to check all the details).
My goal really isn't to use 3dmark as the source of all truth and reality (although I did put most focus on it, cuz it's the only instrumentation I got) but gaming is also showing that something is not ok.

My knowledge, limited experience in debugging these issues, and whatever ability I have in googling things, leave me at thinking that ... maybe the motherboard is having a problem? But it's I don't know what best way to debug from here. Any tool that is able to capture more detailed information on what's happening with the card? Perhaps power delivery? I have no idea where to pick this from.

I'm hopping that this community is able to help me, and all the other 3dmarkers that seem to be having the same problem, and I have a feeling that there might be people on a budget that might be extending themselves so they can enjoy their games, and be facing the same issue. The truth is that over the last 4 years, I went through a backlog of games, and I'm not able to tell when I started having this issue with the 2080ti. Jumping from old games, to AAA games, I believe that I just didn't notice performance drop, and accepted the underutilized reality of my 2080ti, since before this built I had not been playing for 10y.

Thanks for reading all the way here, thanks in advance if you chose to help, and now I'll go curl into a ball and cry myself a river.


 

 

 

 

 

2080ti - 3DMark-TimeSpy-8651-20230725214529.3dmark-result 519.53 kB · 0 downloads 4090 - 3DMark-TimeSpy-8675-20230726171831.3dmark-result 713.64 kB · 0 downloads 2080ti when first built in 2019 - 3dmark-autosave-20190929205015.3dmark-result 501.2 kB · 0 downloads

 

Your CPU is slightly underperforming and the GPU is severely underperforming.

 

There's a CPU bottleneck, but it should not be that high and not with both cards, at least not even nearly the same.

 

Can you provide CPU_Z, GPU-Z, and Passmark RamMon screenshots?

 

Which CPU cooler are you using?

[sumolananas]

Hey all!

@Tan3l6Didn't try, instead did a fresh install of windows.

@hollyh88: Followed your advice, decide to spend more time installing some games, playing around (The Divison 2 & Hitman 3).
The division to play at 4K I would have to be tinkering with the settings around to be at 75fps, and now I can crank everything up and it plays at consistent 90fps. I don't remember the settings I used to play with to be honest, so can't compare the same settings. But while playing Division, afterburn shows the card at >80% usage.

Then tried Hitman Benchmark, ray tracing, all the bells and whistles, no DLSS Frame Generation and managed the score of 72fps, and the GPU would be >85% usage.
I used the same benchmark settings of the video bellow, of a 12900k ddr5 7200mhz. Spoilers, in the video, the average is 85fps, my old pc trying to be young, scored 72fps.
https://www.youtube.com/watch?v=TbkPECZNbAo

So yeh, definitely upgrading my platform will fully utilize the card, but, at least as far as 3DMark goes, it seem to be broken for me, and not a representation of what it means to actually play with it.

I'll keep an eye for someone trying to get rid of an 5800X3D. Or I'll sell the cat, to get the GPU a proper gameworkstation worthy of it.
Thanks for the inputs/suggestions. I'll update the post once/if I update my rig, just to satisfy the curiosity of anyone, to know the difference it makes adding a 5800X3D (although I know there are several videos on youtube, showing how it is pairing a 4090 with different generations of Ryzen, eg: https://www.youtube.com/watch?v=yqviii515kI