airfathaaaaa
Senior member
- Feb 12, 2016
- 692
- 12
- 81
people at ocn and beyond3d have done so many extensive tests almost a year now that they have covered pretty much all the possible ways to explain itI think you have to be careful in saying that some GPUs can't handle Async compute. Ive done a fair amount of research on this subject, and from what I gather Async compute is simply running shaders in the compute queue that is independent of the main queue. Think this is why the term "asynchronous" was coined for this. (Although it also seems like you could synchronize it with the main queue but will result in heavy performance loss).
Now all modern day GPUs can handle this. However performance on the other hand is a different story as TheELF has pointed out. This is the part where its heavily debated on. Its quite hard to exactly understand how nVIDIA deals with graphics+compute tasks (parallel) because there is hardly any information on what happens inside their GPUs.
However at the end of the day, this is all to increase GPU utilization which naturally results in better performance (side effects could be perhaps higher power consumption also since more units are active). So Im beginning to think that there is no one approach to the debate. Plus shared resources also plays a big factor and this is but one tiny parameter in the design that affects overall performance (What normally gets discussed here is very high level).
even the small arm chip that they found on the cards(that wasnt on any paper) is to be blamed too