I have trouble stabilizing the framerate of the Metroid Prime games using Ubershaders.
They generally run very well, but occasionally have a bit of trouble.
I am currently busy with measuring how much resources dolphin requires for these games of the system. Although it does not seem to be all that much.
What I am currently wondering, are there certain requirements to make Ubershaders work more smoothly in general?
Since the GPU core and memory are not close to being full, I suspect more complex things like the bus bandwidth are more important to Ubershaders.
I already tried to look up on the internet whether this question has been answered before, but the closest I came was 'get a beefy computer'.
Your system should be able to handle hybrid ubershaders, or "Ubershaders (Asynchronous)" on latest builds. "Ubershaders (Synchronous)" will be far too much for your GPU.
Also, check that your internal resolution settings are reasonable. I would guess your GPU can't push more than 3x in Dolphin without lagging in some games. and Metroid is a great GPU stresser.
One of the reasons why you may not see 100% GPU utilisation even when your GPU is the bottleneck is that the GPU has to spend a considerable amount of time waiting for the CPU to tell it what to do, so it may only be busy 60% of the time, but still not be done rendering a frame before the screen needs it.
Since when a GTX 950 too little for Synchronous Ubershaders? Sure, at 3x IR it'll probably be terrible, but at 1x or 2x IR it should be plausible.
(05-09-2018, 08:47 AM)JMC47 Wrote: [ -> ]Since when a GTX 950 too little for Synchronous Ubershaders? Sure, at 3x IR it'll probably be terrible, but at 1x or 2x IR it should be plausible.
I ran 3x IR on a radeon 460 (which should be a bit slower) and didn't notice any issues - but I don't have Metroid if that really causes so much of a difference....
But I guess it depends on which 950 you have - the 1024 shader version may be faster than the 768 shader one, and the laptop version will likely be quite a bit slower than that.
Thank you all for the quick replies.
(05-09-2018, 04:54 AM)Helios Wrote: [ -> ]Your system should be able to handle hybrid ubershaders, or "Ubershaders (Asynchronous)" on latest builds. "Ubershaders (Synchronous)" will be far too much for your GPU.
Also, check that your internal resolution settings are reasonable. I would guess your GPU can't push more than 3x in Dolphin without lagging in some games. and Metroid is a great GPU stresser.
Thanks for the information. I will play around with the resolution a bit more.
I always play on lowest on my laptop (not the listed specs) and never saw that much of an improvement, but it is worth a try.
(05-09-2018, 06:12 AM)AnyOldName3 Wrote: [ -> ]One of the reasons why you may not see 100% GPU utilisation even when your GPU is the bottleneck is that the GPU has to spend a considerable amount of time waiting for the CPU to tell it what to do, so it may only be busy 60% of the time, but still not be done rendering a frame before the screen needs it.
So fast communication between CPU and GXF is crucial? I suppose only the motherboard or faster RAM can improve this?
(05-09-2018, 08:47 AM)JMC47 Wrote: [ -> ]Since when a GTX 950 too little for Synchronous Ubershaders? Sure, at 3x IR it'll probably be terrible, but at 1x or 2x IR it should be plausible.
Yeah I will look into the resolution.
(05-09-2018, 09:00 AM)JonnyH Wrote: [ -> ]I ran 3x IR on a radeon 460 (which should be a bit slower) and didn't notice any issues - but I don't have Metroid if that really causes so much of a difference....
But I guess it depends on which 950 you have - the 1024 shader version may be faster than the 768 shader one, and the laptop version will likely be quite a bit slower than that.
Metroid is among the hardest games to get right, from what I have read an experienced.
the 950 is a desktop version, altough it wasnt the most expensive rebrand around. There is a good chance it only has 768 shader units.
(05-09-2018, 06:16 PM)RabbitBones22 Wrote: [ -> ]So fast communication between CPU and GXF is crucial? I suppose only the motherboard or faster RAM can improve this?
No, I think he means that the GPU has to wait for the CPU to finish processing stuff before it can render it so that lowers your utilization average.