RTX 3080 FE performance issue

[freelancer23]

Hello everybody,

I'm experiencing a performance issues and I thinks I've almost tried everything to figure it out. I'll explain, I hope shortly.
I had an 2080Ti and I recently had the "chance" to get an RTX 3080 FE edition.
I'm having the same if not slightly lower performance as my 2080Ti from my previous game benchmark but for make it simple I'll speak only about the Unigine Heaven Benchmark.

My config :

Aorus x470 Wifi 7
AMD 3700x
16Go Corsair 3600MHz (XMP profile)
RTX 3080 FE

My previous bet run from my 2080Ti :

I OC my card and my RAM if I remember well.

And this is my best run so far with my new RTX 3080 FE

And a few games I ran the same benchmark and almost always got less than with my 2080Ti

My Window and driver are up to date, my Bios also. my RAM with the XMP profile at 3600MHz.

My GPU is used at around 96-99% temperature around 77-80° clock is 1950MHz VRAM 9500MHz (MSI afterburner data)
I put 104% power limit (to allow higher temperature) with a more silent curve (I tried differents setup but there is no benefit to OC and make more noise IMO)

I even run a 3D mark bench, which was hopefully normal so my Card has the really power somewhere hidden !

So far what I've tried :
In the Bios to make sure that it's using the Gen 3 of the PCIe
Went to Safe mode, clean uninstall the driver with DDU, go back to safe mode and install the latest Driver without Geforce experience.
Maybe I forgot some. My next step is to try with a fresh install of Windows and try once is updated to go back and back into the version...

Anyone has experienced the same ? Any suggestions ? Otherwise I have no reason to keep it...

 

[LFaWolf]

What is your power supply model? If I remember correctly your GPU temp is too high and throttling. It should be around 70c max. How is your case airflow?

 

[Falkentyne]

Try deleting the files in the NV_Cache folder or delete the shader cache or disable it, reboot then enable it (after deleting the files).
Also set both options in the Nvidia CP to default for both profile settings and reboot also.

 

[BarbaricSoul]

According to TPU's review of the RTX 3080 FE, the 3080 only gives an average of 13% more performance than a RTX 2080ti at 1080p. How much of an increase in benchmark scores were you expecting?

Your Time Spy score is actually above average for a 3700X/RTX3080 combo

 

[jboydgolfer]

i noticed very little performance increase in the games i play , when switching from my 2080ti to my new 3090. in artificial situations like 3dMark, etc, i see a increase, but in real life situations its far less noticeable. the 2080ti is a very capable GPU. 25% is a reasonable performance increase from 1 generation step, its best to ignore advertised performance claims, they lie all the time.

 

[freelancer23]

What is your power supply model? If I remember correctly your GPU temp is too high and throttling. It should be around 70c max. How is your case airflow?

Thanks for your reply, I forgot to talk about the PSU

Corsair HX 1000 which is quite a high end Power Supply I think.

But with my previous 2080Ti Card that I've been using for 1 year and half in the wrong connection way according to this schematic :

I now have connected the RTX 3080 FE like it's showing on this picture. But from before "DON'T DO THIS" to now there where now difference at all during my benchmarks.


About the temperature, I thinks it's normal, When I first launch the card I was worried about the noise because on my 2080Ti I coupled it with the Kraken G12 + Kraken x72 from NZXT so the quitest and best way to cool down any GPU. But I was really surprised about the noise of the stock fan curve which they where around 66-70% with a steady 77° after more than 15min on load.

My Airflow :
Front 3 Noctua Chromax intakes passing through the 280mm radiator from the CPU (I know it's not fully fresh air.. )
Top is closed with glass.
Back 1 Fractal Venturi HF 14 140mm for the exhaust which is among all my research the very best Fan 140mm for the airflow.

I don't really see any difference between open case and when it's closed. So I assume the airflow is quite good.
And as I said I even lower the fan curve to never exceed 60%

 

[mouacyk]

As long as your GPU usage is 98-99%, then it's normal. Your lower performance can be attributed to:
1) VRAM temp throttling, because GDDR6X runs hot and throttles at 110C. If you search in the Ampere owner's thread, almost everyone has to add cooling for VRAM.
2) Power Limit throttling, will give you a lower Effective clock than what you're seeing, sometimes -100MHz. Install latest version of HWiNO64 to see this metric and Memory Junction temp.

To overcome 1), install additional cooling for VRAM and 2) undervolt using MSI Afterburner profile, and maybe 3) water cooling. I have done all 3 and my TimeSpy score is 19,151. Some benches/games don't scale well at 1080p - FYI.

 

[jboydgolfer]

almost everyone has to add cooling for VRAM.

like what?

 

[mouacyk]

like what?

Add/replace thermal pads. Do additional backplate modding. I have a large server copper heatsink screwed into the backplate on mine.

 

[freelancer23]

Try deleting the files in the NV_Cache folder or delete the shader cache or disable it, reboot then enable it (after deleting the files).
Also set both options in the Nvidia CP to default for both profile settings and reboot also.

Thanks for the suggestion, I found NV_cache delete, then turn off the setting in CP reboot, and still the same again.

According to TPU's review of the RTX 3080 FE, the 3080 only gives an average of 13% more performance than a RTX 2080ti at 1080p. How much of an increase in benchmark scores were you expecting?

Your Time Spy score is actually above average for a 3700X/RTX3080 combo

I expect around 215-220 FPS at 1080p because I found an screenshot from an 3080 with an intel 10th gen (don't remember which one) having a result of 226 fps, knowing my CPU is a bit less powerful I expect a bit less.
I also have seen an better result than mine with an 3070 + AMD Ryzen 3600 which are both below my config and had marked 188 fps which is even better than me, so I definitely have a big issue.

I shou've tried to compare with 1440p at least I know, I have an 5140x1440 screen also, and again I should try to see some benchmark with this resolution also. but if in a few games and on unigine heaven I got the same result as with my 2080 Ti ( and worst than an 3070) this is very frustrating.

As long as your GPU usage is 98-99%, then it's normal. Your lower performance can be attributed to:
1) VRAM temp throttling, because GDDR6X runs hot and throttles at 110C. If you search in the Ampere owner's thread, almost everyone has to add cooling for VRAM.
2) Power Limit throttling, will give you a lower Effective clock than what you're seeing, sometimes -100MHz. Install latest version of HWiNO64 to see this metric and Memory Junction temp.

To overcome 1), install additional cooling for VRAM and 2) undervolt using MSI Afterburner profile, and maybe 3) water cooling. I have done all 3 and my TimeSpy score is 19,151. Some benches/games don't scale well at 1080p - FYI.

Okay then I installed HWiNO64 and my temperature juction were around 106-108° I'm a bit shocked, then I just put the fans on maximum and after 5min my GPU when from 82° to 62° and Junction temperature from 108° to 88° so basically -20°
and I ran the same test again, and I rose my score only to 187 fps !

Is that enough to exclude the temperature as the issue ?

I definitely need to try to reinstall the whole windows ...

And about undervolting by lower the core frequency and rising the frame speed, I will try that once I will be on my new windows, and hopefully with an 3080 fully working I start to desperate thanks for your help !

I also can imagine to adapt my G12 and add my kraken x72 to it, and take the heasink I manually crafted for my 2080Ti to put them on my 3080 also.. but I wanted to make things easy...

So I'm here a few hours later with a fresh new windows install and a lot of soft to reinstall.

But with the excact same result...

This is out of the box I simply Updated Windows and Geforce experience with the latest Game Ready driver install Uningine an run the bench.

I found a video from a COD player on Cold war how to fix a low FPS with RTX3000 series. the problem come from Gen 2 PCIe recognition from the Bios.
But as I said I new that before and already set to Gen 3. But just to be curious I donwgrad to Gen 2 and ran my Bench again, and the result is quit clear :

The problem certainely comes from the PCIe. it's like if it's running my RTX 3080 at Gen 2 speed and not Gen 3. I don't see anything else.

 

[freelancer23]

I have done a lot of test and nothing has changed, here what I did :

I tried 4 differents drivers from differents dates, also Studio Driver, every time the same.
I always made clean uninstall in safe mode with DDU.

After that I even tried to uninstall Windows Update, I could only uninstall the two last MAJ, again nothing has changed.

After that I decided to get further in my investigation by running some benchmarks, there aren't extremely precise but they will do the comparaison.

All my bench where made with my 3080, then I put back my 2080Ti and ran all the benchmark with both stock and OC.

It's Obvious that there is a big problem in 1440p, lower perf prooved compare to my 2080Ti.
But in 5120x1440p if we compare to the stock config of my 2080Ti we're not so far from the 30% gain, but it remains only +11% compare to my OC. and much quiter as well, I can't really imagine OC a lot this 3080 like I did with my Asus Strix O11G 2080Ti, it's also the very best Card (except the Matrix)

I will continue to look on internet there are so many people experiencing the same...

 

[Falkentyne]

I have done a lot of test and nothing has changed, here what I did :

I tried 4 differents drivers from differents dates, also Studio Driver, every time the same.
I always made clean uninstall in safe mode with DDU.

After that I even tried to uninstall Windows Update, I could only uninstall the two last MAJ, again nothing has changed.

After that I decided to get further in my investigation by running some benchmarks, there aren't extremely precise but they will do the comparaison.

All my bench where made with my 3080, then I put back my 2080Ti and ran all the benchmark with both stock and OC.

View attachment 191270

It's Obvious that there is a big problem in 1440p, lower perf prooved compare to my 2080Ti.
But in 5120x1440p if we compare to the stock config of my 2080Ti we're not so far from the 30% gain, but it remains only +11% compare to my OC. and much quiter as well, I can't really imagine OC a lot this 3080 like I did with my Asus Strix O11G 2080Ti, it's also the very best Card (except the Matrix)

I will continue to look on internet there are so many people experiencing the same...

Are you using ray tracing in COD:CW? Because if you are not, something is wrong with your CW config. I get 165-230 FPS on a mildly overclocked 3090 FE with RT disabled.
RT enabled however, your numbers look spot on.

 

[Mussels]

Whats your CPU Cooling?

You've added 100W+ into your system, you may have added enough heat to raise CPU temps and lower your boost clocks, therefore CPU limiting yourself.

Try an undervolt curve on the 3080 to keep the heat down, and see what happens (1800Mhz 0.825V worked for me under extreme loads on 3080 and 3090)

3700x is gunna limit the max FPS of a 3080 for sure, i literally did that upgrade to a 5800x and saw my FPS jump in just about every game by 30FPS or so. (165Hz monitor, went from ~120ish to flat 165 in a few titles like halo MCC)

 

[nguyen]

Well nothing too surprising since 2080 Ti has plenty of OC headroom (10-15% as OP has found out, even higher with TDP unlocked BIOS), meanwhile 3080 has very little OC headroom even with 450W BIOS.
Ampere uarch only show its prowess when you are playing at 4K and above
I would still keep the 3080 though, HDMI 2.1 is important when you want to use an OLED TV as a monitor later on.

 

[jboydgolfer]

i have seen little change when moving from my 2080ti to my 3090. aside from the 3090 doing everything at a lower temp.
im still considering upgrading my systems other components, as this build aside from GPU & PSU is a couple years old now.

 

[nguyen]

i have seen little change when moving from my 2080ti to my 3090. aside from the 3090 doing everything at a lower temp.
im still considering upgrading my systems other components, as this build aside from GPU & PSU is a couple years old now.

Just grab yourself a LG OLED screen (C9 or CX) and you don't have to upgrade anything else in order to fully utilize your 3090

 

[Falkentyne]

Well nothing too surprising since 2080 Ti has plenty of OC headroom (10-15% as OP has found out, even higher with TDP unlocked BIOS), meanwhile 3080 has very little OC headroom even with 450W BIOS.
Ampere uarch only show its prowess when you are playing at 4K and above
I would still keep the 3080 though, HDMI 2.1 is important when you want to use an OLED TV as a monitor later on.

An overclocked 2080 Ti does NOT come close to beating a 3080.
An overclocked 2080 Ti on chilled water or sub zero does, however.
His Cold War problems have nothing to do with him having a 3080. Call of Duty FPS issues are rampant, especially with people with AMD CPUs. This has been known and discussed for months now.

 

[nguyen]

An overclocked 2080 Ti does NOT come close to beating a 3080.
An overclocked 2080 Ti on chilled water or sub zero does, however.
His Cold War problems have nothing to do with him having a 3080. Call of Duty FPS issues are rampant, especially with people with AMD CPUs. This has been known and discussed for months now.

You will be surprised that many OCed 2080Ti on air can reach 17k+ in Timespy Graphic, especially with XOC BIOS, even my shit non-A 2080Ti can reach 16k Timespy graphic with the 310W BIOS under normal watercool loop (non-A chip can't be flashed with XOC bios).
So yeah, maybe OP has a bad 3080 sample that barely overclock and a very good 2080 Ti.
At 4K though no amount of overclock can help 2080Ti overtaking 3080, OP testing at 5120x1440 shows just that, 1440p is a different story.

 

[HenrySomeone]

According to TPU's review of the RTX 3080 FE, the 3080 only gives an average of 13% more performance than a RTX 2080ti at 1080p. How much of an increase in benchmark scores were you expecting?

Your Time Spy score is actually above average for a 3700X/RTX3080 combo

Precisely - especially at 1080p, a 3700x just won't be able to extract all of 3080's performance (but as others have mentioned, even with a top tier cpu, the real difference will only be apparent at 4K)

 

[freelancer23]

Are you using ray tracing in COD:CW? Because if you are not, something is wrong with your CW config. I get 165-230 FPS on a mildly overclocked 3090 FE with RT disabled.
RT enabled however, your numbers look spot on.

All my benchmark are with all the max settings in every game, Ray Tracing ultra included.

Whats your CPU Cooling?

You've added 100W+ into your system, you may have added enough heat to raise CPU temps and lower your boost clocks, therefore CPU limiting yourself.

Try an undervolt curve on the 3080 to keep the heat down, and see what happens (1800Mhz 0.825V worked for me under extreme loads on 3080 and 3090)

3700x is gunna limit the max FPS of a 3080 for sure, i literally did that upgrade to a 5800x and saw my FPS jump in just about every game by 30FPS or so. (165Hz monitor, went from ~120ish to flat 165 in a few titles like halo MCC)

My 3700x is cooled with NZXT kraken x62, which adds a little heat inside because they are in the intakes radiator, I could probably change that but so far I do my test with my case open to exclude this heating factor.

Changed for the 5800x! And you got that much difference!! That's amazing if it's only that ai could try to change but I'm really surprised.

 

[P4-630]

Thanks for your reply, I forgot to talk about the PSU

Corsair HX 1000 which is quite a high end Power Supply I think.

But with my previous 2080Ti Card that I've been using for 1 year and half in the wrong connection way according to this schematic :

View attachment 191085

I now have connected the RTX 3080 FE like it's showing on this picture. But from before "DON'T DO THIS" to now there where now difference at all during my benchmarks.


About the temperature, I thinks it's normal, When I first launch the card I was worried about the noise because on my 2080Ti I coupled it with the Kraken G12 + Kraken x72 from NZXT so the quitest and best way to cool down any GPU. But I was really surprised about the noise of the stock fan curve which they where around 66-70% with a steady 77° after more than 15min on load.

My Airflow :
Front 3 Noctua Chromax intakes passing through the 280mm radiator from the CPU (I know it's not fully fresh air.. )
Top is closed with glass.
Back 1 Fractal Venturi HF 14 140mm for the exhaust which is among all my research the very best Fan 140mm for the airflow.

I don't really see any difference between open case and when it's closed. So I assume the airflow is quite good.
And as I said I even lower the fan curve to never exceed 60%

Well I do the "don't do this" and have 0 issues....

 

[freelancer23]

Well nothing too surprising since 2080 Ti has plenty of OC headroom (10-15% as OP has found out, even higher with TDP unlocked BIOS), meanwhile 3080 has very little OC headroom even with 450W BIOS.
Ampere uarch only show its prowess when you are playing at 4K and above
I would still keep the 3080 though, HDMI 2.1 is important when you want to use an OLED TV as a monitor later on.

Yes there is obviously a big potential for OC with the 2080Ti. Probably less with the 3080. I didn't tried yet. And if I do that I'll probably need watercooling. I'm not sure if the G12 will fit.
Or downclock because of the very high temperature of vram would be also an option.

Regarding the screen I will probably never change I have the Samsung G9 which is overkill for any new GPU for the next 2 or 3 generation.

 

[P4-630]

Thanks for your reply, I forgot to talk about the PSU

Corsair HX 1000 which is quite a high end Power Supply I think.

But with my previous 2080Ti Card that I've been using for 1 year and half in the wrong connection way according to this schematic :

View attachment 191085

I now have connected the RTX 3080 FE like it's showing on this picture. But from before "DON'T DO THIS" to now there where now difference at all during my benchmarks.


About the temperature, I thinks it's normal, When I first launch the card I was worried about the noise because on my 2080Ti I coupled it with the Kraken G12 + Kraken x72 from NZXT so the quitest and best way to cool down any GPU. But I was really surprised about the noise of the stock fan curve which they where around 66-70% with a steady 77° after more than 15min on load.

My Airflow :
Front 3 Noctua Chromax intakes passing through the 280mm radiator from the CPU (I know it's not fully fresh air.. )
Top is closed with glass.
Back 1 Fractal Venturi HF 14 140mm for the exhaust which is among all my research the very best Fan 140mm for the airflow.

I don't really see any difference between open case and when it's closed. So I assume the airflow is quite good.
And as I said I even lower the fan curve to never exceed 60%

And btw "TWO PCIe SLOTS" and "THREE PCIe SLOTS" , I assume you mean GPU power connectors and not talking about SLI....

 

[freelancer23]

An overclocked 2080 Ti does NOT come close to beating a 3080.
An overclocked 2080 Ti on chilled water or sub zero does, however.
His Cold War problems have nothing to do with him having a 3080. Call of Duty FPS issues are rampant, especially with people with AMD CPUs. This has been known and discussed for months now.

Yes probably cod is the problem. But if we don't take this game into consideration, it's only a slight improvement, not matching. but less impressive that I thought

You will be surprised that many OCed 2080Ti on air can reach 17k+ in Timespy Graphic, especially with XOC BIOS, even my shit non-A 2080Ti can reach 16k Timespy graphic with the 310W BIOS under normal watercool loop (non-A chip can't be flashed with XOC bios).
So yeah, maybe OP has a bad 3080 sample that barely overclock and a very good 2080 Ti.
At 4K though no amount of overclock can help 2080Ti overtaking 3080, OP testing at 5120x1440 shows just that, 1440p is a different story.

Yes excactely, in 1440p which I will never play anymore, there is like an issue. But for dual 1440p I'm running at higher fps without any OC done yet.

 

[Mussels]

Well I do the "don't do this" and have 0 issues....

depends on the quality and design of your PSU, it varies between systems. It's good practise to run seperate cables, but not a fire hazard if you dont.