Odpověď na názor

Architektura Nvidie je zamerena na buffering. A buffering, pokud bys vedel neco o programovani, je zaklad pro dosazeni dobreho throughput (propustnosti). Nevyhoda bufferingu muze byt delsi odezva dilcich operaci, takze je potreba odvest dobrou praci v driverech a scheduleru, aby to spravne fungovalo.

AMD se nevybralo lepsi cestu, pouze jednodussi. Osobne si myslim, ze z duvodu nedostatku zdroju sli po jednodussi implementaci. Ale ve vysledku, diky dobre zvladnutym ovladacum a scheduleru Nvidie, je na tom Nvidia lip. A na vykonu je to videt.

"Here is an old presentation about CUDA from 2008 that discusses asynch compute in depth - slide 52 goes more into parallelism: http://www.slideshare.net/angelamm2012/nvidia-cuda-tutorialnondaapr08 And that was ancient Fermi architecture."

"Async compute has been a feature of CUDA/nVidia GPUs since Fermi. https://www.pgroup.com/lit/articles/insider/v2n1a5.htm

NVIDIA GPUs are programmed as a sequence of kernels. Typically, each kernel completes execution before the next kernel begins, with an implicit barrier synchronization between kernels. Kepler has support for multiple, independent kernels to execute simultaneously, but many kernels are large enough to fill the entire machine. As mentioned, the multiprocessors execute in parallel, asynchronously.
That's the very definition of async compute."

Takze Nvidia dela Async Compute od roku 2008 a architektury Fermi a to pro DX11 hry. (DX12 bylo predstaveno 2014 a uvedeno 2015).

Zdroj: https://www.reddit.com/r/nvidia/comments/3j5e9b/analysis_async_compute_is_it_true_nvidia_cant_do/


Jedna vec je citovat whitepapery, druha vec je jim rozumet.