BUG: 200-1040 MHZ GPU LOCK ON 3.5.5!

[Alecslee]

So, you have moved the 3.5.5 to stable version KNOWING it has a MAYOR bug related to the gpu getting stuck between 200 and 1040mhz, dropping the performance a LOT, wich has been reported SEVERAL times in the last month:

https://github.com/ValveSoftware/SteamOS/issues/1176

But its obvious your priority was to launch the oled model as soon as posible , wich needs some.of the 3.5 functions, instead of polishing the 3.5.5 after getting it to the stable, and now you have your 3.5.5 update post full of reports of that same bug...well played:

https://steamcommunity.com/app/1675200/eventcomments/7529517132614912523?snr=1_2108_9__2107&ctp=10

Come on valve.....i would fix it as soon as posible if i were you, cuz the bast mayority of the new customers that are going to have this bug, doens't know anything about the gpu, the tdp, or the frecuency, so they are going to asume the machine performs poorly , and its going to be your fault.

[The Asgorian Bugsong]

Probably might help to post specific games are still showing the performance bug on 3.5.7 to help narrow down games to test.

[ReflexShift]

I played a less demanding game today and it played well. GPU was 200mhz when things were light on processing and when I was in area I know taxes my pc it was 1040mhz. The game played at 60fps and was functioning properly. Played a more taxing game and it was a stuttering mess until I set the GPU clock manually. I am starting to think that the performance issues were having with some games might not be because of the clock speed bug that happens. It's just the update changed how the device interacts with some games.

edit: Maybe the update today Nov 22 helped a bit.

[Ihtiandr]

updated to 3.5.7 here, still present.
Checked in Forza Horizon 5. medium settings FSR highest quality, 45-50 fps, temp around 73-75
Set manual gpu clock to 1600 and 60 fps immediately, and the temps jump to 80-81.

[ffrasisti [ARG]]

Ihtiandr님이 먼저 게시:

updated to 3.5.7 here, still present.
Checked in Forza Horizon 5. medium settings FSR highest quality, 45-50 fps, temp around 73-75
Set manual gpu clock to 1600 and 60 fps immediately, and the temps jump to 80-81.

Can you show them in the github issue thread?

[Kennosuke]

Same here just noticed it running a vita emulator & now even upon a restart its constantly hitting these 2 figures. It's causing me to still turn off smt for games & manual applying 1600mhz when I shouldn't even need that for a vita game! 😮‍💨

[The Asgorian Bugsong]

Feels like everyone's having wildly different test results, some saying tearing is causing it, some saying it's not, some saying its steam versions of games causing it, some saying its not

[ingramli]

The Asgorian Bugsong님이 먼저 게시:

Feels like everyone's having wildly different test results, some saying tearing is causing it, some saying it's not, some saying its steam versions of games causing it, some saying its not

Yes, there may be multiple bugs causing similar behavior, in my case, linux based simulator shown obvious performance regression with the GPU frequency jumping between 200 and 1040mhz, shown by recorded video by some folks, at least one of the related bug is unrelated to proton, other(s) may, it seems a complicated matter and takes time to sort out, that's why some folks including me, hope Valve could provide the option (another Steam OS update channel) for folks to return to earlier stable builds (3.4.x) while the devs do the investigation....

[Rui_PT]

ingramli님이 먼저 게시:

The Asgorian Bugsong님이 먼저 게시:

Feels like everyone's having wildly different test results, some saying tearing is causing it, some saying it's not, some saying its steam versions of games causing it, some saying its not

Yes, there may be multiple bugs causing similar behavior, in my case, linux based simulator shown obvious performance regression with the GPU frequency jumping between 200 and 1040mhz, shown by recorded video by some folks, at least one of the related bug is unrelated to proton, other(s) may, it seems a complicated matter and takes time to sort out, that's why some folks including me, hope Valve could provide the option (another Steam OS update channel) for folks to return to earlier stable builds (3.4.x) while the devs do the investigation....

This would be a good option, maybe tell them in GitHub?

[ingramli]

Rui_PT님이 먼저 게시:

ingramli님이 먼저 게시:

Yes, there may be multiple bugs causing similar behavior, in my case, linux based simulator shown obvious performance regression with the GPU frequency jumping between 200 and 1040mhz, shown by recorded video by some folks, at least one of the related bug is unrelated to proton, other(s) may, it seems a complicated matter and takes time to sort out, that's why some folks including me, hope Valve could provide the option (another Steam OS update channel) for folks to return to earlier stable builds (3.4.x) while the devs do the investigation....

This would be a good option, maybe tell them in GitHub?

https://github.com/ValveSoftware/SteamOS/issues/1176#issuecomment-1822069279
I did, but received no response so far...

[Alecslee]

I think the key is in here:

++++++++++++++++++++

"> A reproducible issue: South Park The Stick of Truth, the game always eats too much CPU unless you manually limit the TDP, the behavior on pre 3.5 version is that you set the TDP to 4-5 watts and the game will run smoothly and not consume too much battery. On 3.5 if you limit the TDP to 4-5 watts the FPS drops to basically nothing because the GPU can't clock right, if you manually set the frequency to 700mhz it will go back to smooth 30fps with the TDP at 4-5 watts

------------------------------------

THATS EXACTLY the behivor i saw with emulation and what i've already reported here a couple of days ago.

Its clear that in some light games, if you cap the TDP the gpu isn't capable of adjusting itself properly, so it produces a bottleneck betwenn the tdp and the gpu, but, if you adjust MANUALLY the gpu to be 1-2 points ABOVE the cpu, you recover the performance without having to change the tdp.

To be precisely, my test was with donkey kong tropical freeze, and the behivor was:

- Doing nothing, it mantained 60 fps but the total wattage was high, so i capped the tdp as i usually do.

- Then, with an 8 tdp, wich is DOUBLE than the game usually needed, i noticied the performance dropped to half the normal one.

- Then i looked into the gpu an it was AUTOMATICALLY CAPPED to low values betwenn 200 and 1040mhz.

- So i decided to MANUALLY adjusting the gpu and surprise! i recovered the 60fps WITHOUT having to increase the tdp

- That alone proves the gpu was not adjusting the right way according to the tdp, but there is even more:

- While WITHOUT adjusting the gpu manually, i NEEDED AT LEAST A 9tdp to mantain 60fps, turns out that MANUALLY ADJUSTING the tdp, i could set a 5tdp!!! with a 700mhz GPU, and i mantained the exact same performance of 60 consistent fps but with 4-5watts less of tdp, so its not only a performance issues, but an efficency and battery one


-+++++++++++++++++++++++++

The south park is a gidhub report, the second one IS mine, and both of them proves the exact same problem of the gpu not being able to adjust irself properly according to the tdp, so its an easy scenario to reproduce: light Weight games, capping the tdp, without capping the gpu, and see.if you can achive the SAME performance in both cases, with manual and auto gpu.

AAAANNND another EXACT SAME CASE, this time with video included:

https://github.com/ValveSoftware/SteamOS/issues/1176#issuecomment-1824792986


Mario 3d world with manual tdp but without adjusting the gpu, 19 fps, and adjusting the gpu, 60....so the patern is clear and easy to reproduce, so now we'll see if the problem was the lack of "reproducible" cases, or they simlply don't know how to solve it ans were winning time.

[Broseph Joseph]

Alecslee님이 먼저 게시:

I think the key is in here:

++++++++++++++++++++

"> A reproducible issue: South Park The Stick of Truth, the game always eats too much CPU unless you manually limit the TDP, the behavior on pre 3.5 version is that you set the TDP to 4-5 watts and the game will run smoothly and not consume too much battery. On 3.5 if you limit the TDP to 4-5 watts the FPS drops to basically nothing because the GPU can't clock right, if you manually set the frequency to 700mhz it will go back to smooth 30fps with the TDP at 4-5 watts

------------------------------------

THATS EXACTLY the behivor i saw with emulation and what i've already reported here a couple of days ago.

Its clear that in some light games, if you cap the TDP the gpu isn't capable of adjusting itself properly, so it produces a bottleneck betwenn the tdp and the gpu, but, if you adjust MANUALLY the gpu to be 1-2 points ABOVE the cpu, you recover the performance without having to change the tdp.

To be precisely, my test was with donkey kong tropical freeze, and the behivor was:

- Doing nothing, it mantained 60 fps but the total wattage was high, so i capped the tdp as i usually do.

- Then, with an 8 tdp, wich is DOUBLE than the game usually needed, i noticied the performance dropped to half the normal one.

- Then i looked into the gpu an it was AUTOMATICALLY CAPPED to low values betwenn 200 and 1040mhz.

- So i decided to MANUALLY adjusting the gpu and surprise! i recovered the 60fps WITHOUT having to increase the tdp

- That alone proves the gpu was not adjusting the right way according to the tdp, but there is even more:

- While WITHOUT adjusting the gpu manually, i NEEDED AT LEAST A 9tdp to mantain 60fps, turns out that MANUALLY ADJUSTING the tdp, i could set a 5tdp!!! with a 700mhz GPU, and i mantained the exact same performance of 60 consistent fps but with 4-5watts less of tdp, so its not only a performance issues, but an efficency and battery one


-+++++++++++++++++++++++++

The south park is a gidhub report, the second one IS mine, and both of them proves the exact same problem of the gpu not being able to adjust irself properly according to the tdp, so its an easy scenario to reproduce: light Weight games, capping the tdp, without capping the gpu, and see.if you can achive the SAME performance in both cases, with manual and auto gpu.

AAAANNND another EXACT SAME CASE, this time with video included:

https://github.com/ValveSoftware/SteamOS/issues/1176#issuecomment-1824792986


Mario 3d world with manual tdp but without adjusting the gpu, 19 fps, and adjusting the gpu, 60....so the patern is clear and easy to reproduce, so now we'll see if the problem was the lack of "reproducible" cases, or they simlply don't know how to solve it ans were winning time.

I have observed this same behaviour whilst playing sonic 3 AIR (non steam game). I make pretty extensive use of TDP limiting for games, and used to be able to run AIR on 4 watts. Now i find the gpu clock sometimes inappropriately gets stuck at 200mhz if i TDP limit, which trashes the framerate well below 60, there is no reason a machine with the power of the deck cant play sonic 3 at 60fps at all times.

My laymans idea of whats going on is that there is something wrong with the gpu frequency scaling governer below 1040mhz, as if there are no intermediate frequencies between 200 and 1040, so the scaling drops all the way down to 200mhz as soon as anything below 1040mhz is called for, instead of to a more appropriate intermediary frequency like 600 or 800 or 1000mhz etc for example.

[jaaylow]

Not sure if this is common knowledge but you can downgrade OS to 3.4.X and it'll work just fine (in most cases). Then you just need to keep the device in standby because switching it off will auto upgrade the OS when you next boot up.

Not ideal but a workaround until they issue hot fix!

[Alecslee]

10 minutes with my new OLED 1TB, only 3 random games tested, and i've been able to reproduce the bug xD, and i have to believe that the enginiers of valve haven't able to do It...come on.

There you have another example, video included:

https://github.com/ValveSoftware/SteamOS/issues/1176#issuecomment-1824880754

Again, locking the tdp at 8, with gpu in auto, 23fps, but adjusting the gpu at 1000mhz, 45 with the same tdp.

[Alecslee]

jaaylow님이 먼저 게시:

Not sure if this is common knowledge but you can downgrade OS to 3.4.X and it'll work just fine (in most cases). Then you just need to keep the device in standby because switching it off will auto upgrade the OS when you next boot up.

Not ideal but a workaround until they issue hot fix!

Search in this same topic, i've commented about one command for the konsole that allows you to keep your actual boot version despite rebooting etc.

[ingramli]

jaaylow님이 먼저 게시:

Not sure if this is common knowledge but you can downgrade OS to 3.4.X and it'll work just fine (in most cases). Then you just need to keep the device in standby because switching it off will auto upgrade the OS when you next boot up.

Not ideal but a workaround until they issue hot fix!

Boot to 3.4.11, go to desktop mode, open konsole, paste the following,

sudo rauc status mark-active booted