I’m told the front end of the compiler is all Apple and compiles to a shared IR, but the backends are blobs provided by the GPU vendors
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Is there a switching cost that doesn't amortize well for short benchmarks?
-
Happens even if I force discrete.
End of conversation
New conversation -
-
-
There's a lot of things that could be going on. Unified memory architecture helps some tasks. Mac OpenGL drivers are awful, etc.
-
IIRC Intel GPUs also blowing away both nVidia + AMD for Geometry Shader performance. Doubt you're using those tho, not sure Mac GL has them.
- 2 more replies
New conversation -
-
-
could it be memory is less far away on intel gpus? for datastructures and dependent reads.. possibly
-
I don't get how you guys can be so much into GPUs and into Macs at the same time. Ever hear of Razor, Alienware, XPS, Gygabites or MSI ?
- 1 more reply
New conversation -
-
-
surprising, could the Radeon be bandwidth limited on these tasks, while integrated is using shared memory? or is the task fill rate limited? also, MBP is reported to throttle discrete GPUs in some cases too
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Tried using half floats for vertex data in alacritty. Huge win on Iris and absolutely bombed on Radeon
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
curios: what ML library do you benchmark/use?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.