If you understand what preemption is then you know it makes no sense to say it's used to run concurrent tasks.
Let me tell you how concurrent tasks run on Pascal. It's really easy. You see, some SMs are idle and the HW scheduler decides to schedule some independent compute (or graphics) work on them.
Done. It wasn't that hard If you cast away all the unfounded statements we heard from AMD less technically savvy supporters you see there is really no mystery.
Now we just need to repeat this for three or four months and with a bit of luck the message will percolate. No wait, who am I kidding
1. I didn't said anything about used for concurrent workload.
2. Jee, what was I posting again? Right, a question about How Pascal does Async Compute . Does that doesn't qualified as a question? If so, how could you accused me of saying the
bold.
3. Pascal used a software-based scheduler. Where did it get the HW scheduler from? Fermi?
4. A more technical answer would be more helpful and educating would you agree?
5. Don't be a douche. I asked nicely for an explanation and did not try to offend your Pascal Overlord k.
From what I gather it's down to granularity. AMD can do it per shader.
Nvidia on the other hand used to be able to split different kinds of calculations between different gpc's, effectively lopping off a large portion of rendering power in order to allocate it to compute calculations.
With Pascal different types of calculations can be allocated to each SMM cluster within each GPC group, and the amount of shaders within each cluster has been halved compared to Maxwell allowing for finer granularity.
Thanks @Deders for a genuine explanation . But, isn't Pascal can do it per shader now too (Maxwell was GPC only, Pascal per SM inside GPC) What's confused me is some said it's a clever hack (Pre-emtion, NV driver) so... yeah. Confused.