MSI AB / RTSS development news thread

Question for @Unwinder
Here i tested Forza Horizon 4 with Single channel RAM vs. Dual channel and i found useful feature "Stutters counter"

Maybe possible to add that feature in to MSI AB that calculates "Stutters" and shows in real time?

 

Because the suggestion thread is locked, I will post my question/suggestion here. @Unwinder

May it be possible to implement the following:

- trigger a specific (next/previous) boost state/frequency (via hotkey) and lock the GPU in that state/frequency (for example +- and pressing + lifts the GPU a boost state up and vice versa)

I am asking because the boost logic of the Nvidia driver doesn't seem to be optimal or work properly always. It would be handy, to have a manual overwrite mode, where you can force the GPU into boost state 1/2/3 and then lock it there.

I have the issue in some games, that the automatic boost mode of the Nvidia driver doesnt work. And my GPU is stuck at 800MHz and MSI AB shows a usage of 99-100%. The driver doesnt seem to notice, that the usage is 99% and boost up correctly. I found out, that a workaround for this in some games is, to alt-tab out, open a Chrome window, load a page, then alt-tab quickly back into the game, and now the GPU boosts up correctly and FPS become stable at 60, instead of 10-30.

When this is triggered, the usage suddenly goes down from 99 to 70-80 and FPS become stable and good. Though even the clock goes back down to 900-1000 the usage stays around 80-90%. It seems the voltage maybe is risen resulting in proper FPS when this happens.

 

In Nvidia Control Panel change the setting Power management from "adaptive" or "optimal power" to "prefer max. Power", for these games or even in general.

 

That is not a satisfying solution. Also it doesn't work always for some games (another bug?). I dont want max performance where there is no reason for. I dont want the GPU clocking at max frequency boost, where being in the next upper boost from 800MHz would work fine for most games I play (like World of Warcraft on low settings for example). It seems there is a bug in the Nvidia logic or it is not clever enough to notice, that in some games the current boost isnt enough to support 60FPS, even AB states a usage of 99-100%.

It would be nice, if you could tell AB via hotkey: Toggle into next boost state/frequency and also lock it there.

Nvidia control => bugged and doesn't work always
Max Performance => not a satisfying solution to this problem

I have a lot of games, where it runs good at 60FPS with a boost around 900-1300MHz (also memory at 810MHz instead of 3500MHz works fine for most games for 60FPS). 1700+ would be overkill for it and produce not needed fan noise and energy drain.

I dont understand, why the Nvidia driver doesnt notice the GPU usage is 99% and trigger the next boost state by that. This seems to be a bug. It is not in all games though.

 

the nvidia driver uses a utilisation % before ramping up the power state, some games are always just going to need to be set to prefer max, starcraft2 and assassins creed come to mind.

 

Can someone explain me how do I choose the value for this scanline sync ? I just want to use it and be happy. I've tried 1080 1085 -50 -200 and so on but most of the times I have tearing on screen ?! frame rate is above refresh rate which is 75 and vsync and similar functions are off.

 

You decrease it until the tearline disappears at the top of the screen. If that's not possible (meaning the tearline just randomly appears at the bottom and top of the screen with no way to hide it), then you need to try SyncFlush=1 in the profile file. If that also doesn't help, then you need to try SyncFlush=2. If that results in stutters, then you need a faster GPU and/or need to lower the graphics settings of the game in order to lower GPU load.

 

Do I start from 0 and go negative until the line disappears ? Is this option using a significant amount of GPU computing power ? I do not need to understand what SyncFlush does but from the name of it quite explanatory...

Will test and see ! Is there a way I can test this value without actually firing up a game ? like in a browser something like UFOs (https://www.testufo.com) maybe I can fine a walking line to set this up.

Thanks for your time !

 

No. You start from -1. 0 disables scanline sync completely.

It does not use GPU by itself, but it needs a fast GPU to work well. It's about how fast the GPU can clear all pending operations to the screen. The faster the GPU is, the faster it can do that. If it's not fast enough, the tearline will not be stable on the screen.

SyncFlush=1 is usually fine in DX10, DX11 and GL, since it's an asynchronous flush. If the GPU is not fast enough, it will simply make the tearline unstable. In DX9, DX12 and Vulkan, it doesn't work and and it will use SyncFlush=2 instead, which is a forced flush, and if the GPU is not fast enough, you will get major stuttering. So keep in mind the API the game uses. When you use SyncFlush=1 but you're playing a DX9, DX12 or Vulkan game, then it means SyncFlush=2 and that one really needs a fast GPU (or a very non-demanding game; very old games work perfectly fine even on mid-range GPUs, for example.)

I don't think so. It need something RTSS can actually hook into. Note that you don't have to quit the game and restart it to change the scanline value. You can do that live while it's running.

 

In RTSS beta 5, there is no SyncFlush command in the game config files. Should we add it in?

 

Yes, in the [Framerate] section.

 

I hope this is the right spot to post this, I've noticed that when scanning for OC the resulting memory clock is lowered by 200. Im normally running a +900, so I can't actually compensate by changing it to +1100, I'm not sure how to clock it past what the bar let's me do (1000).

 

I've had that problem too on some cards that go over +1000. The 200MHz loss is from the driver switching into compute optimized mode.

I have a similar limit issue with going BELOW stock clocks on memory as well.

Anyway, see no reason to pick 1000 as the highest number. It could be 951 or 1342, why 1000?. - Same with the fan slider lower limit of 25%. I guess that's a bit more justifiable but I personally dislike it since you have to fiddle around with the curve to test anything below 25%. Testing fans to see where they stop spinning and what the minimum startup value is . A lot of sliders are useful with wider ranges, anyone who could cause serious problems for themselves simply from that would screw themselves up anyway, I don't' see why it's worth limiting fan from 0-100 or memory from a large range (I would say +/-1500 but I'm just pulling a number out of my hat, Unwinder would definitely know best on this subject...). I really do hope the memory slider is increased on both sides.

 

Have you tried disabling this with nvidia profile inspector? Scroll down to CUDA - Force P2 State and set it to OFF.
Of course, this may be more error prone (which is why it's on by default) and could cause the OC scanner to find problems that will only exist in the P0 state, but it seems to me that should be the correct scanning method anyway, since one would expect the resulting OC to work at P0.

 

WOW thanks! That makes things a heck of a lot simpler for compute loads and profiles!