I don't think this fits. What does it matter? The technology is utilizing a hardware GTX cards just don't have. Is Nvidia not supposed to innovate? or not utilize hardware until some arbitrary amount of time passes so owners of old cards can feel better? Like how come you're not complaining about AV1 decode not coming to GTX or RDNA1? Raytracing/Sampler Feedback/Mesh Shaders with older AMD cards, Smart Memory Access - should all those features be cut on your card because older ones don't have access?
You do realize that Pascal has support for DP4a instructions. So it could run DLSS in shaders. It would not be as efficient as tensor cores and performance would not increase as much, but it would be a way to improve performance for many gamers. nVidia did use shaders to run DLSS 2.0 in a version of Control. And Intel's XeSS has support for it's own tensor units, and for DP4a. So it will run on Pascal, RDNA2 and RTX cards. At a time when it's so difficult to upgrade your GPU, giving support of DLSS on Pascal would be a welcome addition. Don't you think?
No one is saying NVidia nor AMD should not innovate. They can innovate to their heart's content. The issue is that we are no longer in a normal upgrade cycle. A statistically sizeable portion of PC gaming enthusiasts has been priced out of their hobby for the time being. The least either manufacturer could do is to lend them a helping hand through driver software optimizations - even if it means looking back instead of forwards for once. Even corporate greed ought to have its limits.
I would rather the finite resources at Nvidia are spent at improving DLSS with tensor cores further like what they are doing now. Pascal owners can sell their GPU and buy newer ones, they had plenty of opportunities to do so. Heck even selling my old 1080Ti right now could get me a brand new 3060 (not much of an upgrade I know, but 3060 is still a better GPU than 1080Ti feature-wise)
Is it clear exactly how this tech works? Is it like DLSS but forceable from the driver? Is it DLSS but with dynamic input resolution? I'm just a bit unclear how this replaces DSR. Usually DLSS was used to improve performance while DSR was used to just brute force higher input resolutions, no? Thanks,
Pascal has 1/4th performance of INT8 of Turing and cannot concurrently issue INT/FP - How do you know DLSS/DLDSR is even utilizing INT8? DLSS '1.9' is not an example. How do we know it wasn't running in FP16? Either way it was worse quality than 2.0 and significantly worse than what we have now (2.3) - also the performance advantage of DLSS isn't entirely from Tensor cores it's mostly from rendering the image at a lower resolution. Run RSR or NIS on Control and compare it to DLSS 1.9, $10 says the spatial upscalars will do a better job with the image with the same performance benefit.
For older games in particular I'm really excited to try out those features. I do hate that these have to be done via GeForce overlay and not the Nvidia Control Panel -- I much prefer enabling things there if I can.
DP4A is not the same as supporting packed 8-bit integer math of some kind, it's a specific instruction doing 8x8->16bit multiplication and saturating accumulation into a 32-bit integer supporting various combinations of signed and unsigned operands. Presumably DP4A and DP2A counted as a single instruction, not a quad/double rate operation run at the same rate as either 32-bit integer or float instructions. From nVidia itself: https://docs.nvidia.com/cuda/pascal-tuning-guide/index.html#int8 GP104 provides specialized instructions for two-way and four-way integer dot products. These are well suited for accelerating Deep Learning inference workloads. The __dp4a intrinsic computes a dot product of four 8-bit integers with accumulation into a 32-bit integer. Similarly, __dp2a performs a two-element dot product between two 16-bit integers in one vector, and two 8-bit integers in another with accumulation into a 32-bit integer. Both instructions offer a throughput equal to that of FP32 arithmetic. DLSS 1.9 still had a decent performance boost, despite running on shaders. In terms of quality, it was lacking but I guess it wasn't using motion vectors and temporal accumulation for the final result. I think that only came with DLSS 2.0 And of course, with training, nVidia has made huge steps with more recent version of DLSS 2.x On the other hand, we can look at the Intel presentation where they compare render times of XeSS on DP4A and tensor units. As you can see, DP4A might be 2.5X slower than XMX, but it's still faster than native rendering.
I did not see that must missed it, i just previous post saying it was done via GFE overly but that might been referencing something else i will have to double check.. If it done via CP that i most intrested to see how it down scales 1440p/4k to 1080p and if look clear and more readable special on desktop. would save me the trouble of buy new monitor for time being, IF it works how i think it would anyway and i coudl be wrong about that too
@BlindBison was presumably talking about the Freestyle filters also mentioned in the OP - which are enabled through GFE.
probably, I know why I missed screenshot of DLDSR listing in CP though... I didnt have coffee yet when I was looking threw the thread. I still only install what I need for drivers to work which drivers physx and HDaudio, hell i still running 466.11 drivers, I see point chaning them when everything works and nothing i need new drivers yet.
I just checked my own NVidia Control panel settings there, and for some reason the "DSR - Factors" setting was set to 2x and "DSR - smoothness" at 33%!?? Does that mean I was killing my 3070 framerate running all my games at 2x1440p resolution all the time unknown to myself?? I did a reset to defaults, and the DSR is now set to OFF!
GT 710 is still being produced(or it was as of last year), doesn't mean anyone should expect it should have modern amenities Couldn't have said it better myself. It's like people want innovation, they want new features, but they are unwilling to accept the innovation and new features People who bought the last GTX cards knew what they were getting and what they were not getting, they knew nvidia was heavily putting their efforts and innovation into the RTX series, and the people who didn't know that are also the people who are extremely unlikely to even understand what DLDSR is, let alone use it. So there's no point at which i can understand being upset this doesn't go to what we already knew was less featureful GTX cards. In all honesty i feel like people only complain here because they feel entitled as gamers? cause this isn't abnormal for literally anything. You buy a samsung galaxy M series phone, you do not expect to get the support and features of the S series phone, or if there are new features down the line updated through software, you do not expect it'll fall back to the M series. You buy a LG OLED A2 tv series you would not expect it to have the same support and features of their LG OLED G2 series, and if future software comes out that improves the TV features, you would not expect it to go to the lower end. Not to say just because you would not EXPECT it to that it's impossible it wouldn't in any situation, but you would not EXPECT it to. This is life, there is nothing unique here just because it's gaming hardware and we are gamers.....
RSR will likely get expanded support in later drivers. The same thing happened when they introduced Radeon Image Sharpening.
The accompanying photo in this article shows that DLDSR 2.25x is almost as fast fps-wise as native resolution. Does this mean if you use DLDSR 1.78x instead, you will get faster frame rates than native resolution? Can you also use DLDSR together with DLSS?