Nvidia 4090 GPU stuck at PCIE1.1x16

[WakeX]

Nope, only exact same as the last beta:
- AGESA ComboPI 1.0.0.7a updated with the required solution.

So far after punching in all my settings again, I've noticed the following two new entries in the BIOS:
Under OC there is a setting below Memory Context Restore called Memory Failure Retry Count which defaults to 1, nice to have for memory settings which only sometimes fail to train I guess.

Under ACPI Settings there is a new entry called CPU Fan Fail Warning Control which defaults to Disabled, which is also a good addition.
Not sure why it's set to disabled as default though, shouldn't the opposite the case and only need disabled if you're not running a fan/pump but some other external solution like LN2 or a chiller maybe?


The time spent on post code 15 (memory training) was also cut down massively on my settings (MCR auto, PD off) to around 32 seconds - Last BIOS time reported by task manager was way over 100 seconds easily on the prior beta BIOS, now it's consistently below 50 seconds. I wonder if I force MCR to enabled if it would affect anything and be even faster (or if I'd get BSODs like some reported with PD off), gonna try later.

Unfortunately the PCIe 1.1-bug still is present and showed up quite fast on the third and fourth reboot in a row, while on the fifth everything was fine again:

Once it runs at 4.0 I don't notice any instability, crashes or the like - so I'm not sure why that happens, if the signaling was bad wouldn't it have issues at 4.0?

I've got a Crucial T700 5.0 4TB on the way, maybe even this week already which will replace the slow WD Blue SN550 3.0 250GB (which will go into an external case) in the slot closest to the CPU - wonder if changing that will impact this pesky PCIe 1.1-issue, if at all.


Oh yeah, almost forgot:
Trying to get into M-Flash to update the screen froze and I was greeted with a 00 on the post code.
I pressed the power button on the case first to try getting a shutdown but that just restarted the system and then it went into M-Flash - after which I flashed the new BIOS without a hitch (should've probably rebooted cleanly and try again but yeah...).
Odd behavior to say the least.

 

[scuzzy]

Nope, only exact same as the last beta:
- AGESA ComboPI 1.0.0.7a updated with the required solution.

So far after punching in all my settings again, I've noticed the following two new entries in the BIOS:
Under OC there is a setting below Memory Context Restore called Memory Failure Retry Count which defaults to 1, nice to have for memory settings which only sometimes fail to train I guess.

Under ACPI Settings there is a new entry called CPU Fan Fail Warning Control which defaults to Disabled, which is also a good addition.
Not sure why it's set to disabled as default though, shouldn't the opposite the case and only need disabled if you're not running a fan/pump but some other external solution like LN2 or a chiller maybe?


The time spent on post code 15 (memory training) was also cut down massively on my settings (MCR auto, PD off) to around 32 seconds - Last BIOS time reported by task manager was way over 100 seconds easily on the prior beta BIOS, now it's consistently below 50 seconds. I wonder if I force MCR to enabled if it would affect anything and be even faster (or if I'd get BSODs like some reported with PD off), gonna try later.

Unfortunately the PCIe 1.1-bug still is present and showed up quite fast on the third and fourth reboot in a row, while on the fifth everything was fine again:

Once it runs at 4.0 I don't notice any instability, crashes or the like - so I'm not sure why that happens, if the signaling was bad wouldn't it have issues at 4.0?

I've got a Crucial T700 5.0 4TB on the way, maybe even this week already which will replace the slow WD Blue SN550 3.0 250GB (which will go into an external case) in the slot closest to the CPU - wonder if changing that will impact this pesky PCIe 1.1-issue, if at all.


Oh yeah, almost forgot:
Trying to get into M-Flash to update the screen froze and I was greeted with a 00 on the post code.
I pressed the power button on the case first to try getting a shutdown but that just restarted the system and then it went into M-Flash - after which I flashed the new BIOS without a hitch (should've probably rebooted cleanly and try again but yeah...).
Odd behavior to say the least.

That 1.1 bug is super strange, have you had a workig BIOS that doesnt show this?

 

[lostking20015bf02e6]

Nope, only exact same as the last beta:
- AGESA ComboPI 1.0.0.7a updated with the required solution.

So far after punching in all my settings again, I've noticed the following two new entries in the BIOS:
Under OC there is a setting below Memory Context Restore called Memory Failure Retry Count which defaults to 1, nice to have for memory settings which only sometimes fail to train I guess.

Under ACPI Settings there is a new entry called CPU Fan Fail Warning Control which defaults to Disabled, which is also a good addition.
Not sure why it's set to disabled as default though, shouldn't the opposite the case and only need disabled if you're not running a fan/pump but some other external solution like LN2 or a chiller maybe?


The time spent on post code 15 (memory training) was also cut down massively on my settings (MCR auto, PD off) to around 32 seconds - Last BIOS time reported by task manager was way over 100 seconds easily on the prior beta BIOS, now it's consistently below 50 seconds. I wonder if I force MCR to enabled if it would affect anything and be even faster (or if I'd get BSODs like some reported with PD off), gonna try later.

Unfortunately the PCIe 1.1-bug still is present and showed up quite fast on the third and fourth reboot in a row, while on the fifth everything was fine again:

Once it runs at 4.0 I don't notice any instability, crashes or the like - so I'm not sure why that happens, if the signaling was bad wouldn't it have issues at 4.0?

I've got a Crucial T700 5.0 4TB on the way, maybe even this week already which will replace the slow WD Blue SN550 3.0 250GB (which will go into an external case) in the slot closest to the CPU - wonder if changing that will impact this pesky PCIe 1.1-issue, if at all.


Oh yeah, almost forgot:
Trying to get into M-Flash to update the screen froze and I was greeted with a 00 on the post code.
I pressed the power button on the case first to try getting a shutdown but that just restarted the system and then it went into M-Flash - after which I flashed the new BIOS without a hitch (should've probably rebooted cleanly and try again but yeah...).
Odd behavior to say the least.

The built in render test does not put enough load. My 4070 Ti does this with GPU-Z also.
Using Unigine Heaven, the PCI-E runs at 4.0 without any issue.

 

[WakeX]

That 1.1 bug is super strange, have you had a workig BIOS that doesnt show this?

Not sure about that because I was running a different GPU while on older BIOS versions, the 3060 Ti FE never got into a bugged state like this.

The built in render test does not put enough load. My 4070 Ti does this with GPU-Z also.
Using Unigine Heaven, the PCI-E runs at 4.0 without any issue.

That's not true though, as you can see in this gif I recorded:

The 4090 goes to 4.0 for a moment before settling on 2.0 but you can still see the change at least and in heavier loads like games it will stay at 4.0 as usual, the 3060 Ti in contrast stays at 4.0 at all times in GPU-Z already.

Once it's in the broken state it does not change from 1.1 at all and games will run at 11-12 fps.

 

[lostking20015bf02e6]

Not sure about that because I was running a different GPU while on older BIOS versions, the 3060 Ti FE never got into a bugged state like this.


That's not true though, as you can see in this gif I recorded:

The 4090 goes to 4.0 for a moment before settling on 2.0 but you can still see the change at least and in heavier loads like games it will stay at 4.0 as usual, the 3060 Ti in contrast stays at 4.0 at all times in GPU-Z already.

Once it's in the broken state it does not change from 1.1 at all and games will run at 11-12 fps.

If it gets stuck at 1.1 in the games, yes, that is the issue.

 

[Alan J T]

Not sure about that because I was running a different GPU while on older BIOS versions, the 3060 Ti FE never got into a bugged state like this.


That's not true though, as you can see in this gif I recorded:

The 4090 goes to 4.0 for a moment before settling on 2.0 but you can still see the change at least and in heavier loads like games it will stay at 4.0 as usual, the 3060 Ti in contrast stays at 4.0 at all times in GPU-Z already.

Once it's in the broken state it does not change from 1.1 at all and games will run at 11-12 fps.

Can you set it to PCIE Mode 4 in the bios instead of AUTO

 

[WakeX]

Can you set it to PCIE Mode 4 in the bios instead of AUTO
View attachment 174335

Sure can do but tbh I often leave the PC running in the night for several days when it's downloading something anyway (not the fastest net here).
So once it works I don't reboot or shutdown that often.

I also tried to provoke earlier it by restarting but gave up after 5 times when it still was working fine each time, so it's really odd that it sometimes happens twice in a row too on the other side of the coin.

 

[Encryptme]

Not sure about that because I was running a different GPU while on older BIOS versions, the 3060 Ti FE never got into a bugged state like this.


That's not true though, as you can see in this gif I recorded:

The 4090 goes to 4.0 for a moment before settling on 2.0 but you can still see the change at least and in heavier loads like games it will stay at 4.0 as usual, the 3060 Ti in contrast stays at 4.0 at all times in GPU-Z already.

Once it's in the broken state it does not change from 1.1 at all and games will run at 11-12 fps.

Just tested my 4090 and it does the same thing, also using the same version of GPUZ which is the latest.

 

[lostking20015bf02e6]

Just tested my 4090 and it does the same thing, also using the same version of GPUZ which is the latest. View attachment 174381

View attachment 174382

Try to verify with some game work load, such as Unigine Heaven. The GPU-Z render test is kinda of light. Sometimes it doesn't need full bandwidth.

 

[Encryptme]

Try to verify with some game work load, such as Unigine Heaven. The GPU-Z render test is kinda of light. Sometimes it doesn't need full bandwidth.

Tried with Superposition and I do get 4.0 on the bus interface!! Thanks

 

[Alan J T]

Tried with Superposition and I do get 4.0 on the bus interface!! Thanks

YA just had a look and no way enough load on a 6900XT let alone a 4090 to trigger it out of low-power mode

 

[jaba1337]

Unfortunately the PCIe 1.1-bug still is present and showed up quite fast on the third and fourth reboot in a row, while on the fifth everything was fine again:

Once it runs at 4.0 I don't notice any instability, crashes or the like - so I'm not sure why that happens, if the signaling was bad wouldn't it have issues at 4.0?

I've got a Crucial T700 5.0 4TB on the way, maybe even this week already which will replace the slow WD Blue SN550 3.0 250GB (which will go into an external case) in the slot closest to the CPU - wonder if changing that will impact this pesky PCIe 1.1-issue, if at all.

Do you have PCIE Link State Power Management turned off in Windows power settings? That can cause some weird pcie speed issues when it is on. https://www.thewindowsclub.com/link-state-power-management-windows

 

[WakeX]

Just tested my 4090 and it does the same thing, also using the same version of GPUZ which is the latest. View attachment 174381

View attachment 174382

Do you mean it gets stuck at 1.1 completely in a broken state on your system too?
If so, that would make 3 cases (including that user from LTT-forums and me).

Try to verify with some game work load, such as Unigine Heaven. The GPU-Z render test is kinda of light. Sometimes it doesn't need full bandwidth.

YA just had a look and no way enough load on a 6900XT let alone a 4090 to trigger it out of low-power mode
View attachment 174385

Doesn't matter what load, if it doesn't change at all from 1.1 to 4.0 and settle on 2.0 during the GPU-Z render test it's in a broken state (which just happened again after a cold boot on my system):

3DMark PCI Express feature test (normal is 26.8x GB/s):

Games will run at ~10 fps if that, for example I fired up a Witcher 3 benchmark-savegame (where fps is generally lower=more load than in other areas of the game):

Do you have PCIE Link State Power Management turned off in Windows power settings? That can cause some weird pcie speed issues when it is on. https://www.thewindowsclub.com/link-state-power-management-windows

Link State Power Management was always turned off in the Windows-power settings.

Can you set it to PCIE Mode 4 in the bios instead of AUTO
View attachment 174335

Rebooted from the broken 1.1-state, went into BIOS and set the slot to 4.0 as you suggested but on next boot after saving it the broken 1.1-state continued to persist:

Frametimes look less jumpy but I wonder if that was affected by forcing 4.0 at all or something else entirely probably.

Also since setting it manually to 4.0 didn't help, I went back into BIOS and set it back to Auto - then when I hit Save+Reboot it tried to reboot, got stuck with only _ showing on screen and post code showing F0 which apparently means "Recovery condition triggered by firmware (Auto recovery)"...gg !

 

[abysal2023]

Do you mean it gets stuck at 1.1 completely in a broken state on your system too?
If so, that would make 3 cases (including that user from LTT-forums and me).



Doesn't matter what load, if it doesn't change at all from 1.1 to 4.0 and settle on 2.0 during the GPU-Z render test it's in a broken state (which just happened again after a cold boot on my system):

3DMark PCI Express feature test (normal is 26.8x GB/s):

Games will run at ~10 fps if that, for example I fired up a Witcher 3 benchmark-savegame (where fps is generally lower=more load than in other areas of the game):

Link State Power Management was always turned off in the Windows-power settings.


Rebooted from the broken 1.1-state, went into BIOS and set the slot to 4.0 as you suggested but on next boot after saving it the broken 1.1-state continued to persist:

Frametimes look less jumpy but I wonder if that was affected by forcing 4.0 at all or something else entirely probably.

Also since setting it manually to 4.0 didn't help, I went back into BIOS and set it back to Auto - then when I hit Save+Reboot it tried to reboot, got stuck with only _ showing on screen and post code showing F0 which apparently means "Recovery condition triggered by firmware (Auto recovery)"...gg !

Is your GPU connected via riser cable extension by any chance? Forgive me if this was already asked. PCIe 4.0 is finiky with those extensions, like those used for vertical GPU mounts.

 

[WakeX]

Is your GPU connected via riser cable extension by any chance? Forgive me if this was already asked. PCIe 4.0 is finiky with those extensions, like those used for vertical GPU mounts.

No riser cable, it's in the PCIe-slot directly.

4) Also shortly checked this GPU-Z PCIe issue discussed. My 3080 is mounted in the PCIe Slot directly (no riser cable) - Idle shows x16 1.1 - starting the test shortly spikes to 4.0 and then settles at 2.0. In Game (Star Citizen) performance is as expected and Bus goes to x16 4.0.

Yeah, that's the expected behavior but sometimes after a cold boot or reboot it just hangs on 1.1 with performance being in the dumps accordingly.
There's a high chance it stays stuck at 1.1 even after trying several reboots. In that case it's better to try a shutdown and then cold boot again I guess.

 

[KrisO]

No riser cable, it's in the PCIe-slot directly.


Yeah, that's the expected behavior but sometimes after a cold boot or reboot it just hangs on 1.1 with performance being in the dumps accordingly.
There's a high chance it stays stuck at 1.1 even after trying several reboots. In that case it's better to try a shutdown and then cold boot again I guess.

I have just checked and 1.1 information about bus interface speed is absolutely normal in idle. As soon as your graphic card is engaged the speed goes up to 4.0. I have not observed any negative impact in 3D Mark tests or in games. I would not worry about it.

 

[Alan J T]

No riser cable, it's in the PCIe-slot directly.


Yeah, that's the expected behavior but sometimes after a cold boot or reboot it just hangs on 1.1 with performance being in the dumps accordingly.
There's a high chance it stays stuck at 1.1 even after trying several reboots. In that case it's better to try a shutdown and then cold boot again I guess.

Is there a Power saving mode of High-performance mode in the Nvidia Driver setting
I know there something like that you can try with the low power option disabled and High Performance enabled

Best Nvidia Control Panel Settings [High FPS & Performance]

This guide will cover the best Nvidia Control Panel settings for maximum gaming performance in 2024. These work on Laptops and Desktops.

exputer.com

 

[Alan J T]

I have just checked and 1.1 information about bus interface speed is absolutely normal in idle. As soon as your graphic card is engaged the speed goes up to 4.0. I have not observed any negative impact in 3D Mark tests or in games. I would not worry about it.

Seems AMD just don't care as mine just stays @x4x16 now I pulled the USB card out of slot two since PCIE-3 is sorted out

 

[RugaCH]

Seems AMD just don't care as mine just stays @x4x16 now I pulled the USB card out of slot two since PCIE-3 is sorted out
View attachment 174445

I just ran into the PCIe 1.1 issue too (Diablo 4 just crashes outright, WoW at least runs):

The readings from the GPU are very odd.
Still on v17 BIOS atm - and it was fine before I went to bed last night, after booting today I got this issue.

RTX 4090 on PCIe 1.1

Hello! I built a new PC two weeks ago when Ryzen 7 7700X launched and now have a Gigabyte RTX 4090 gaming OC hiding my whole motherboard. 😄 Upgraded from R7 2700X + GTX 1070. After 2 days of gaming, I opened GPU-Z 2.50.0 this morning and discovered that I was actually running on a PCIe 1.1 x16 bu...

linustechtips.com

They're discussing it here too.

 

[RugaCH]

Also interesting:
in Afterburner, the Temp and Power Limit are set to 0:

When I change them to the max, they just fall back to 0.