But you may be in a situation in which your CPU is loaded with stuff that isn’t vector rendering—e.g. JS—and the GPU would otherwise be idle. Or the CPU might be slow, as in mobile. But then again, maybe not: viewing static PDFs on desktop is something people do a lot!
-
-
Show this threadThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
If you have the luxury of being able to dispatch redundant requests and then avoiding redundancy for the loser (while still dispatching a little over time to shift if resource until changes) it can auto-tune the decision
-
You can also do things like shift bias between 5-95% and use a small fixed sliding window of your cost metric to pick the bias. Cost is another point of design freedom. Pretty simple control theory application can go really far for stuff like this
End of conversation
New conversation -
-
-
Don't decide now: collect statistics at runtime and at runtime balance between CPU and GPU.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
How much is this affected by a specific choice of CPU and GPU?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
You may also have contention issues on a heavily-loaded a box and far more CPU than GPU cores available. This is an interesting class of problems, at some point I suspect we will wind up with HLLs that compile to both and choose just in time.
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.