AMD Could Do DLSS Alternative with Radeon VII through DirectML API

i could really care less about DLSS, its image quality improvements are of questionable value. just because you can do it, doesn't mean you should...

 

Plus, how many Turings are out there? there are much more Pascal GPU's on the market (Bought new by gamers or 2nd hand ex-mining cards) so Pascal gamers wil left in the dark while New turing cards will enjoy the DirectML era, vs AMD's Vega GPU's that will get Free performance boost over 1080Ti.

 

The quality of the image is based on a bunch of factors that are constantly improving. Look at neural net imaging from 5 years ago to today, it's night and day.

Given Nvidia's market saturation and the time it will take AMD to pull together a GPU Open variant of DLSS on DirectML (Then train a few games and integrate it) Turing will most likely have shipped more cards then AMD with Vega.

Capable yes but the performance is kind of an unknown. Microsoft showed a 275x improvement over CPU on a Titan V on Tensor cores. Vega won't nearly get that much performance and Pascal doesn't packed FP16 so the performance will be even worse - if it's even worth doing at that point. Microsoft basically said under a certain point and you can't do it in real time reliably.

 

It's what was stated by MS during DirectML announcement. There's also no reason why Vega or any other cards would get any performance boost from DirectML.

 

Maybe I'm an idiot, but DirectML doesn't seem like some Tensor AMD response. Tensor cores are at their heart interesting for DLSS because the processing is done separately, and the idea is there is little/no performance impact as the Cuda cores continue to focus on the game.

 

Honestly I think this is where Nvidia went wrong with DLSS, the marketing.

The technology is quite interesting and I were more keen on DLSS than the whole RT paradigm when the RTX cards were announced, simply because RT isn't going to see wide adoption for quite a while.

Unfortunately Nvidia chose to market DLSS as a performance uplift, in conjunction with upscaling, and compared to TAA no less - when postfx AA has been a curse inflicted on PC gaming rather than anything that actually benefits image quality.
I believe Nvidia should have sold us on DLSS not as a performance enhancer at 4K but as a way to reintroduce SSAA for modern titles with a lower performance hit.

Looks almost as good as TAA at 4K? Yeah, not impressed.
Computationally cheap SSAA at 1080/1440p? Sold!

Give me a way to help relegate postfx AA to the garbage bin of history and I'll start throwing money at the screen.

 

I think there would be some tradeoff. You'd basically use some CU's for packed FP16 & trade that power for an algorithm to save performance overall.

But yeah you're right - Tensors are separate while any implementation regarding this would basically be 'stealing' from overall shader/compute performance.

 

Would it have to be though?

I mean surely Nvidia could implement driver middleware which would leverage the tensor cores for DirectML on Turing cards?

 

Waits with Popcorn in one hand and soda in another for said fireworks to happen.

I do like the Nvidia DLSS it pretty much Free AA, if it works with everything it great other wise those cores are waste of space, cause that it there only only gaming card, if look better or worse then TAA is irrelevant to me cause it would be FREE AA with zero hit and I sure as time goes by, would pay the price for these said RTX card hells no

I hope this work out for AMD

why would they do that ? I mean they could by why, when they have the RTX card that will do there version of it? , just gsync thing happening. Dont get me wrong I not fan of priority crap the gpu companies pull, if there is break threw it should be shared with all gpu makes for the benefit of the gamer, but it all about the money.

 

It's pretty obvious that DirectML will run on tensor cores on Volta and Turing. There's little reason to even create a new API otherwise.

 

Yeah I meant on AMD - Rereading it, I wasn't clear, sorry.

RPM works by shoving two similar FP16 operations down an FP32 register. So if DirectML is utilizing RPM to accelerate on AMD it's essentially stealing from it's overall shading performance. On Nvidia, with Volta/Turing, DirectML would leverage the tensor cores and run the operations in parallel. Which is what currently happens with DLSS, Nvidia just implemented it's own "DirectML" with NGX.

 

DLSS is not free, you still need tensor cores lol

 

DirectML is not free, you need a GPU.
Freesync is not free, you need a computer.

 

Not really good analogy to diss him down

Anyone has a dx12, while only a few tensor cores.

 

AMD GPU's are so weak as it is how are they going to cope with the new load?

 

I am just curious if AMDs DLSS through this API will carry as much benefit as Nvidia's method of using separate tensor cores to handle the job. You would think Nvidia's dedicated solution would perform better in that scenario but I am just guessing logically without knowing how well Radeon 7 could do with that. Maybe AMD can leverage the additional compute of Radeon 7 to handle DLSS nicely.

Additionally, I hope to see Nvidia add additional uses for the tensor cores on their GPUs that could especially help those choosing not to use DLSS. Probably the one down side on Nvidia is idle tensor cores especially right now with no games supporting just yet.

We might not have got the GPUs we thought we wanted at this point but I think I like where things are going. It is also confirmed that AMD will be working on Ray Tracing GPUs in the near future. Ray Tracing is only going to progress forward at this point and we should be seeing new things in the GPU world. Which is good. Lets just keep the prices in check

 

Aside from the fact that an open standard would probably win in the long run, regarding ease of implementation and so on, presumably DirectML would be used for more things than DLSS. Since Nvidia's newer architectures has dedicated hardware to handle such work, why not leverage it?

 

Precisely what I've said before. DLSS can be thought of a set of instructions that has been deep learned, it can be executed on any hardware so long as the work is put in.

 

I just got info that Nvidia CEO is losing the leather jacket and buying large amounts of moisture wicking t-shirts because AMD is putting some pressure on him lately.