Will PCSX2 ever use 4 cores and vulkan?
#11
(04-20-2016, 05:20 PM)gregory Wrote: Yeah I know the impact of BA. But it will be better to improve the driver. Or to add the MT layer in GSdx. It ought to be done one day for vulkan/dx12 anyway.

Edit: that being said. BA is slow because it requires lots of draw call. Future hardware (at least latest Maxwell dunno for AMD) could do it in a single draw.

Single Draw call. Near zero performance loss in Maxwell. Hope same apply for pascal GPU
Reply

Sponsored links

#12
Yes it apply to Pascal too. It is a GL extension GL_ARB_fragment_shader_interlock. I know Nvidia requires recent hardware. But it might also work on recent AMD/Intel GPU.

However the perf overhead will be 0 on the CPU. But GPU will require extra work (and likely extra memory bandwidth). BA was only a proof of concept Wink
Reply
#13
(04-20-2016, 02:20 PM)gregory Wrote: So good move for AMD, bad move for everyone else (nvidia, dev, all users as code will be optimized for a couple of card)

I don't see how this could be bad for AMD GPU users. Smile
Reply
#14
Easy, every time you update the application you must ensure that performance will stay the same on all GPU architecture. So there is a high probability that game will be optimized for a couple of generation card. Potentially it will become slower if you upgrade your GPU. Or an update of the game could make it slower for an older gen.

It is the same for the driver but multiplied per applications. It is much easier to test performance of common code rather that performance of thousand of games.

So far, feedback said, it is very easy to optimize the CPU but it is very hard on the GPU. It is annoying because most app are GPU limited anyway.
Reply




Users browsing this thread: 1 Guest(s)