Well, here is some good news for NVIDIA users. It appears that NVIDIA will fully implement Async Compute via an upcoming driver.
Via driver huh? Well, I guess it's better than nothing
Emulated perhaps, via CPU? I dont know, I dont understand this stuff!! lol
It was only a matter of time either way, and it's time Nvidia can easily afford.
Pretty curious now about Pascal and whether NVIDIA has fixed all these DX12 issues
I hadn't taken the time to read a lot of technical info on why Nvidia cards had trouble with this. But I had assumed it should be able to be added via driver. The hardware itself is pretty straight forward, I wasn't aware of anything that should have limited it from doing async.
At this point, it's a software/hardware combo solution for NVIDIA's cards (unlike pure hardware solution for AMD's GCN cards)
Nvidias 900 series cards have async computer
I don't fully agree but again I'm no expert in the matter. How I see it though is just like how it works with other components I.e. Memory, cpu etc... It has to be hardware designed so the power runs through the circuitry in parallel. That's how the components are designed in parallel pipelines then how can they run in parallel? This adding it by a driver update doesn't make sense to me at all. Unless it was already built that way but was purposefully disabled until the point were they could sell it as an extra feature & just enable it in future & they've been caught short. What they're talking about would surly consist of a complete driver rebuild from the ground up from scratch. Either way, the fact that it's software based means Nvidia controls exactly how much power they allow to their customer depending on which card they buy. Then lower it again in time for a release of a new card with benchmarks that show huge improvements rather than what it actually should of been.
All cards need software i.e. drivers to say how the hardware works and what features are present and available. Not sure why everyone is saying NVidia is just emulating async, can someone explain this fully?
They explain at the end of the article, Maxwell 2 will use a mix of hardware and software whereas GCN uses hardware: "Maxwell 2: Queues in Software, work distributor in software (context switching), Asynchronous Warps in hardware, DMA Engines in hardware, CUDA cores in hardware. GCN: Queues/Work distributor/Asynchronous Compute engines (ACEs/Graphic Command Processor) in hardware, Copy (DMA Engines) in hardware, CUs in hardware.”
Thanks, so that is emulating Async then and not proper Async?
Yea it's definitely not proper Async, I have a feeling it's not going to work too well in comparison with AMDs GCN native hardware support, can't see emulation matching the real thing.
No one seems to understand that Nvidia just does Async Compute differently from AMD. Hardware support is there, software support isn't up to scratch yet, that's the only async compute "problem" Nvidia users are having right now.
Yea i was just typing about that.
To summarise : Nvidia's async compute works just fine, only with a certain known limit which devs failed to notice or rather nvidia failed to notify, but it works just fine. So the ahes dev programmed nvidias async as if its similar to amd's async hardware, which just caused problems for nvidias hardware. If you read all the articles it comes down to this. Nvidia's cards have had async built into them for years, in fact generations, but their interpretation is slightly different than amd as async was not a unified thought back then and has a long way to go before all devs use it fully, its in its basic form now, in fact this was stated by multiple devs over many articles. Only with DX12 did async become more usable, and so some devs are trying this out, which of course being the FIRST dx 12 game will have issues, Nvidia even notified them their driver isnt ready for the Ashes version where it was clearly programmed assuming both Nvidia Async and Amd are the same, which they arent. Ashes dev said something which was then blown out of proportion and now nvidia is gonna patch in the automatic limitation im assuming so devs can clearly see the limit on not program the two different brands the same way.
N4G is a community of gamers posting and discussing the latest game news. It’s part of NewsBoiler, a network of social news sites covering today’s pop culture.