This is Nvidia, but it's still machine learning and for sure the consoles will support it.
They added the hardware on the shader cores. Not only that they have said since last year directx ML makes use of any ML hardware available including the tensor cores.
my assumption: devs submit super-high resolution images so it's learning information that can't be resolved a 1080p (like that Wolfenstein 16K resolution ground truth image Nvidia shown)
The Xavier line of Tegra chips have built-in tensor cores, so it isn't that far fetched considering Nvidia has been at it with mobile tech for automated cars.
The tensor cores are just hardware support to deal with int matrix math. Both consoles have hardware for that even if not dedicated cores.
Of course it's applying as having dedicated hardware and just running it on your standard pipeline are two completely different things.
yes, but the code needs to be continuously tied to a back end that keeps it updated. that was the problem with DLSS 1.0, it was on the devs to do the updating and they had no general pool to learn from
aye, as far as I can find, the tensor cores add 1.25mm2 per SM (this is comparing a TU106 with a TU116)
Reddit - Dive into anything
www.reddit.com
You are looking this the wrong way. Tensor cores allows the workloads to be independent from the shaders but they take significant die space which takes away from overall processing of the gpu.
Ok, they still won't get dlss2.0 since consoles are using amd tech.
Yeah, I could see DLSS happening if they go all in on it and dedicate a much larger relative area of the die to tensor cores compared to what we have on Turing. Apparently the tensor cores on the 2060 sum up to ~22mm2, which is 18% of the die area for the X1, so it's feasible.
:o
www.eurogamer.net
Even if it does make its way, people always saying "it can't possibly run on (insert next Nintendo console)" will definitely try to find other excuses... :p
But that's probably what's going to happen in most cases unfortunately.
We still do not know if PS5 supports increased rate int8.
The difference between having separate hardware (NVIDIA's tensor cores) and having hardware support inside the shader cores (XSX approach) is that in the former scenario, you don't have to compromise on GPU compute power for native rendering in order to support a DLSS-like feature, while in XSX case, you need to share resources between native rendering and the upscaling AI algorithm. For example: the RTX2060 has 100 TOPS of tensor core performance. If we assume that we need 25 TOPS to do upscaling from 1080p to 4K, then it can do that on the tensor cores, and use its full 6 TFLOPS of ALU power to render the native 1080p image. The XSX, on the other hand, needs to apply 25 TOPS of shader compute power out of its 49 TOPS total, or roughly half of its GPU, to the DLSS-like algorithm, leaving only half of its GPU for native 1080p rendering. If it takes half the GPU to perform the DLSS algorithm, you lose quite a bit of the potential benefit.
Oh, very interesting! Looks like RT and Tensor combined are only about 20% of the GPU die size, so that in itself shouldn't be the major issue with including that hardware I think.aye, as far as I can find, the tensor cores add 1.25mm2 per SM (this is comparing a TU106 with a TU116)
Reddit - Dive into anything
You have some things incorrect here. RTX 2060 is around 100 TOPs - the XSX is 49 for int8. int4 TOPs is double for both - so around 200 for RTX 2060 and 97 for XSX.
Oh I see, I accidentally found the founders edition, and used the half-precision tensor cores number by accident. Will update my post to reflect it!
It blows my mind how much they can do with a 540p image. You can barely make out the lampposts in the non-DLSS one
This is even more impressive
This is nuts..
I like how TAA is is still one of the worst temporals AAs out there for majority of games. lol. Even nvdia had their own TXAA , that was short lived
