Personally I feel like tensor cores are too expensive for the gaming space and fsr has made dlss somewhat redundant now Nvidia has to reinforce tensor cores usefulness.

What's going on nVidia? Why the hurry all of a sudden? Is it FSR that made you do this? 😎

2.0 already looks fantastic higher iq and improved motion quality is welcome,but what nvidia needs to work on is figuring out how to make a card with +8g vram

Here is a pack with most DLSS 2 versions It also includes 2.2.9.0 DLSS Library

icedman:

I feel like tensor cores are too expensive for the gaming space and fsr has made dlss somewhat redundant now

Is having dedicated silicon on your GPU expensive? Yes. Does FSR make DLSS redundant? No. The war just started, we can't say one wins over the other yet.

Devid:

Why the hurry all of a sudden?

FSR working on both teams might have shaken things a bit at the Green minion quarters. But I doubt this is something sudden, these things probably were in the working for months anticipating FSR. This is what good innovative companies do, try to silently have another card under their sleeve and always push with new tech. Not like Intel did with quad cores for who knows how many years...

cucaulay malkin:

what nvidia needs to work on is figuring out how to make a card with +8g vram

Still they sell everything, so no hurry.

FSR knocking on nVidia heels, i think Freesync vs Gsync deja vu all over again.

Maybe v3.0+ will improve on ghosting issues many games have with all DLSS versions, especially older.

DLSS 2.2 has a bit less ghosting than DLSS 2.0 and 21. Anyone can replace the nvngx_dlss.dll of a previous version, with a newer version, and improve image quality. https://cdn.videocardz.com/1/2021/06/NVIDIA-DLSS-2.2.9.0-comparison.jpg

lukas_1987_dion:

Maybe v3.0+ will improve on ghosting issues many games have with all DLSS versions, especially older.

Try the 2.2.6.DLL

AMD FSR Successfully Implemented in GTA V Thanks to Clever Mod Installation: Get the latest release Copy d3d11.dll and gta5_fsr.ini to Grand Theft Auto V root folder (e.g. \Steam\steamapps\common\Grand Theft Auto V) Go to ingame settings Advanced Graphics Settings and set Frame scaling mode: 0.5x - Performance Mode 0.667x - Balanced Mode 0.75x - Quality Mode 0.883x - Ultra Quality Mode Check gta5_fsr.ini file if you need more control over mod settings Compatibility: ENB FiveM RAGE.MP ✔️1 ❓ ❓ Rename this mod d3d11.dll to d3d11_fsr.dll and edit enblocal.ini file: [PROXY] EnableProxyLibrary=true InitProxyFunctions=true ProxyLibrary=d3d11_fsr.dll NarutoUA also provided a gameplay video. As you can see below, FSR provides a much sharper image than the original upscaling method of Grand Theft Auto V. [youtube=yN3tfoUUyos]

Devid:

What's going on nVidia? Why the hurry all of a sudden? Is it FSR that made you do this? 😎

FSR aside, UE5 features a brand new AA solution " Temporal Super Resolution " which allows you to drop the resolution drastically with little to no change in output pixel quality.

icedman:

Personally I feel like tensor cores are too expensive for the gaming space and fsr has made dlss somewhat redundant now Nvidia has to reinforce tensor cores usefulness.

DLSS give us free AA and at the same time can now give a higher quality image than the original in some situations, even though it also has the performance benefit of rendering at 66% of native. And the latest versions eliminate most temporal issues. This article constructively explains the many issues with FSR and potential ways to improve it in the future that are already implemented in one game. https://www.eurogamer.net/articles/digitalfoundry-2021-amd-fidelity-fx-super-resolution-fsr-performance-wins-but-what-about-image-quality Basically, FSR cannot reconstruct missing information from rendering at lower res, has temporal image quality issues and does not perform AA. The improvements listed are using the UE4 engine but might be possible to implement on other engines. IF NVidia cards were slower than AMDs you would have a point but not a solution. If prices were normal I'd happily pay for what DLSS now gives and the higher performance at UHD, though AMDs prices are just as crazy atm. I'm lucky I got my card before things went mental. I can understand wanting to use an FSR dependent card with current prices, I might have gone there too if I hadnt got lucky, but it takes nothing away from what Tensor cores can do that FSR cannot.

I see the benefits of DLSS, but I'm a native man at heart. I suspect these upscaling technologies will become mandatory as developers lazily use the headroom created by them instead of having to optimize.

Mufflore:

This article constructively explains the many issues with FSR and potential ways to improve it in the future that are already implemented in one game.

I used to enjoy DF videos, I mostly still do I suppose - but I'm really over all the pixel peeping, and 400% zooming-in in an attempt to show the blue / yellow pixel went missing... DLSS is clearly good, and FSR Quality and Ultra-quality is good enough. We just need something available in all demanding games so we can overcome some of the current silicon limitations.

beedoo:

I used to enjoy DF videos, I mostly still do I suppose - but I'm really over all the pixel peeping, and 400% zooming-in in an attempt to show the blue / yellow pixel went missing... DLSS is clearly good, and FSR Quality and Ultra-quality is good enough. We just need something available in all demanding games so we can overcome some of the current silicon limitations.

The entire point of DF's reviews are going into detail about how the technologies work, their pitfalls, the benefits etc. If they just showed two images and went "Looks good enough" and that was review, it would be doing a massive disservice to the industry. It's like MKBHD reviews of phones vs Anandtech - one gives a brief overview of the device, the other drills into the specifics and uses objective measurements to show why one technology is better than the other. Similarly, MKBHD used "pixel peeping" to describe people pointing out color calibration issues with screens (notably color shifting with the LG Pixel OLED screens) and he kind of just wrote it off like "its not that big of a deal" but if everyone did that, why would Google or in this case AMD with FSR, strive to improve the product? Edit: And honestly like even if you don't zoom in Ultra Quality is not good enough in most of the titles. I'm looking at some of these techpowerup comparison shots and they look horrible: https://www.techpowerup.com/review/amd-fsr-fidelityfx-super-resolution-quality-performance-benchmark/7.html https://www.techpowerup.com/review/amd-fsr-fidelityfx-super-resolution-quality-performance-benchmark/4.html _ Compare those to recent Doom ones: https://preview.redd.it/qce9pk98vj871.jpg?width=3200&format=pjpg&auto=webp&s=29ab20e53944d7110b8af04da824a592c827a482 https://preview.redd.it/8xcs8g98vj871.jpg?width=2400&format=pjpg&auto=webp&s=4a71aa9d3e66ad18b251b01737b32291d2bad8c4 https://preview.redd.it/gwnoel98vj871.jpg?width=3440&format=pjpg&auto=webp&s=86e24533eb6342123123a89a54ecbbfb11583f6c And those are high zoom. If you swapped the labels I probably wouldn't even be able to tell the difference between those.

There is another potential use for this. Game streaming kind of sucks because of the latency. Perhaps AI post-processing could allow multiple potential frames to be streamed simultaneously and then the user's input determines which frame becomes the active frame. Extremely good AI up-scaling would allow for a huge number of low-res frames to be sent simultaneously avoiding a lot of the bandwidth issues with this potential streaming solution.

icedman:

Personally I feel like tensor cores are too expensive for the gaming space and fsr has made dlss somewhat redundant now Nvidia has to reinforce tensor cores usefulness.

I disagree personally -- DLSS 2.1/2.2 is remarkably impressive in my experience and FSR is definitely not a replacement for the latest iterations of DLSS. Most outlet reviews I've seen for the initial FSR implementation place it as follows in the realm of upscaling/reconstruction techniques (where FSR is not really an image reconstruction technique, but rather a good upscaler going off the tests I've seen) -> [DLSS 2.2 > modern TAAU implementations such as the one found in UE4/UE5 > FSR > traditional bicubic/bilinear upscalers]. FSR is great for what it is, but it's basically just a good single frame spatial upscaler with some CAS sharpen. It's nice to have especially compared to traditional upscalers, certainly, but unless they move to a solution which can utilize more than the current frame, I'm skeptical it can match the quality of other options which pull from previous frames + current, use motion vectors, and for DLSS also uses the whole AI inference/comparison to "ground truth" source images. FSR needs to run on part of the GPU as well though of course it's quite inexpensive and what you gain is a lot more significant. It's nice FSR can run on any GPU and I agree that it's disappointing that the Tensor cores aren't more broadly used beyond DLSS (someone correct me if I'm wrong, iirc Nvidia originally said they could be used for RT denoising?), but DLSS itself in its current iteration is amazing.

XenthorX:

FSR aside, UE5 features a brand new AA solution " Temporal Super Resolution " which allows you to drop the resolution drastically with little to no change in output pixel quality.

I'm not sure yet exactly how much of an impact to image quality lowering the internal resolution would make for the revised UE5 TAAU solution as I haven't used it or really seen it tested yet (though I've seen the UE4 one tested), but TAAU does seem really cool, yeah.

beedoo:

I used to enjoy DF videos, I mostly still do I suppose - but I'm really over all the pixel peeping, and 400% zooming-in in an attempt to show the blue / yellow pixel went missing... DLSS is clearly good, and FSR Quality and Ultra-quality is good enough. We just need something available in all demanding games so we can overcome some of the current silicon limitations.

I still enjoy DF content, but I definitely acknowledge my perception is they've sorta shifted into more "filler" content. For example, I like some of their work, but do we really need 10+ minute videos on a minor patch that went live? Or the equivalent of a 2 hr microsoft teams work call? In fairness, sometimes there are some gems within those long calls and I suppose I can just treat it more as a podcast so perhaps I'm being overly harsh. I absolutely loved Alex's V-Sync video and Motion Blur video and Rich does some of the best CPU reviews around seems to me (though I am absolutely baffled they completely skipped reviewing Ryzen 5000 -- like, wtf? It was the biggest CPU launch of the year and they didn't review it). I wouldn't go so far as to call them shills for X brand like some other do on the board, but I would acknowledge that my perception is they often seem unwilling to be as critical or harsh on a given product as maybe they should be. Still, perhaps that's just my own arrogance, after all it's definitely possible they know more than I do given their testing experience and it can be hard to differentiate between whether they're just genuinely excited over new tech or whether their intentionally being lenient for business reasons. Perhaps it's a bit of both, but I still generally enjoy their content and reviews personally.

Could a game run with both fsr and dlss active