(11-06-2014, 12:04 PM)Shonumi Wrote: [ -> ]kirbypuff Wrote:On Linux (Ubuntu), the the OpenGL backend performance is even worse (OpenGL on Windows is still faster)
Careful how you say that. As a general statement, that most certainly is not true for all GPUs and drivers (I have verified this on my system, Linux + OGL runs as fast or faster than Windows + OGL). It's really not something you can have a blanket statement about, especially since there are a lot more factors that can influence Linux performance than on the Windows side of things (distro, kernel, drivers, distance from the sun). It's really something that has to be decided on a case-by-case basis, since user reports have been all across the board in this regard. The only thing I will say that's pretty much established currently is that the OSS drivers for either AMD or Nvidia are inferior to the proprietary drivers, performance wise. Intel performance is actually pretty good since Intel actually does work on MESA and seems to care about Linux.
I'm pretty sure he was talking about AMD. Which is completely true. AMD on Windows OpenGL performance has always been better then Linux.
The thing is we've had Dolphin users for both Nvidia and AMD use both Windows and Linux, some of them report one being faster than the other, but these accounts are not consistent in any way.
Furthermore, benchmarks that are not Dolphin-based aren't entirely applicable when talking about Dolphin performance. If we are going to throw around benchmarks,
Phoronix shows that OpenGL on Ubuntu matches Windows pretty consistently when talking about Nvidia, beating it in fact more than losing. It's only in GPUTest that it significantly falls behind, but as with most any synthetic benchmark, it doesn't represent real-world performance as accurately as other indicators.
AMD is a different story, true. However, benchmarks and tech demos don't always give accurate accounts of performance in real-world applications (as evidenced by Xonotic section, it's basically a toss-up between the OSes).
It's important to keep in mind that Dolphin isn't a PC game, and it doesn't use the GPU quite the same way as a PC game does. It's also not benchmark software. The only sure-fire way to determine how Dolphin will perform with certain GPUs on certain OSes is to test them directly. So again, with Linux vs Windows on OGL, it's case-by-case when you're talking about how it will run Dolphin.
kirbypuff Wrote:The dkcr_moving_melters_paused benchmark (the most demanding GPU benchmark ever for EFB to Texture)
And what about efb to ram?
Did you test both GPUs all the way up to 7x?
NaturalViolence Wrote:And what about efb to ram?
EFB to RAM also requires a fast CPU with a decent memory controller (high system memory bandwidth). If the requirements are met, 6x or 7xIR is no problem for the GPU.
With Tino's Ishiiruka build, 6x/7xIR with EFB->RAM is possible even on AMD CPUs.
Quote:Did you test both GPUs all the way up to 7x?
Yes.
kirbypuff Wrote:EFB to RAM also requires a CPU with a decent memory controller and high system memory bandwidth.
This is a bit misleading. While memory bandwidth requirements do go up a bit that's not the main source of framerate drops with efb copy to ram. Even with 50GB/s memory bandwidth many games still encounter a big framerate drop when using efb copy to ram. Even though only a fraction of the memory bandwidth is being used.
Have you compared your results with other games?
(11-07-2014, 10:54 AM)NaturalViolence Wrote: [ -> ]This is a bit misleading. While memory bandwidth requirements do go up a bit that's not the main source of framerate drops with efb copy to ram. Even with 50GB/s memory bandwidth many games still encounter a big framerate drop when using efb copy to ram. Even though only a fraction of the memory bandwidth is being used.
It's not just the system memory bandwidth. It also needs a fast CPU to properly "feed" the GPU. AMD GPUs perform much better with a fast CPU. The faster the CPU - the better the performance.
EDIT: Updated the Dolphin GPU stress test with a better one. The previous one was already pretty extreme, but new one is brutal (about 10% more demanding than the previous one) - it represents the worst-case scenario.
The results didn't change much - the 7850 still managed to score over 60fps at 7xIR.
Any GPU that passes this test at a specified IR is guaranteed to run pretty much everything at that IR.