QCOM problem is that they're at crossroads. Their iGP is extremely efficient and performant for the Mobile market. It's been custom tailored through the years for this market demands.
But now they have to scale up Adreno to Desktop Class graphics computation. And there, the direction GFX is going is totally different from Mobile.
Mobile is basically TBDR + small batch FP16 workloads, with careful shaders that don't go overload. Desktop is completely different, with focus on plenty of bandwidth, Compute Shaders and next-gen geometry.
QCOM will need to design a GPU arch that can still be extremely area and power efficient for mobile while also being able to serve Desktop current and future demands.
Mobile is basically TBDR + small batch FP16 workloads, with careful shaders that don't go overload. Desktop is completely different, with focus on plenty of bandwidth, Compute Shaders and next-gen geometry.
So they have the same problem Apple does with its TBDR based GPU that doesn't perform/benchmark well in stuff (esp. games) designed for a traditional PC GPU pipeline.
So they have the same problem Apple does with its TBDR based GPU that doesn't perform/benchmark well in stuff (esp. games) designed for a traditional PC GPU pipeline.
Well Apple simply doesn't care, because they aren't concerned with running Windows stuff that requires high GPU performance. People will run native for productivity type tools, and they mostly don't game on Macs either.
I think all mobile gpus are like that, like the Mali G610MP4 was considered high end just 3 years ago. But in Linux running desktop apps it performs worse than a 2019 Vega 3 IGP.
Not sure what the problem is exactly, but i belive mobile just has other requierements, and these gpus are specially bad in FP ops. I belive it is because mobile gpus are still tailored for OpenGL ES.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.