News Nvidia's RTX 50-series drivers feel half-baked, focus too much on MFG

[Elusive Ruse]

fake frames aren't real frames. Don't bench what isnt real.

What should he bench? What’s a real frame that your card renders then? Does it render real time footage of your driveway or revolution of the planet?
How about you articulate what is the actual issue you have with frame generation without regurgitating what you heard elsewhere.

 

[YSCCC]

What should he bench? What’s a real frame that your card renders then? Does it render real time footage of your driveway or revolution of the planet?
How about you articulate what is the actual issue you have with frame generation without regurgitating what you heard elsewhere.

raster, with maybe DLSS or FSR upscaling enabled is what should be realistically benched, FG makes things appear smoother but the latency to response still kept as the base FPS, it is a bonus info in a review, but when ppl are reading the FPS results in reviews, most are just looking at the numers and have a rough idea on how the stutter/lagginess to input, so the raw FPS is more important in action games.

 

[Elusive Ruse]

raster, with maybe DLSS or FSR upscaling enabled is what should be realistically benched, FG makes things appear smoother but the latency to response still kept as the base FPS, it is a bonus info in a review, but when ppl are reading the FPS results in reviews, most are just looking at the numers and have a rough idea on how the stutter/lagginess to input, so the raw FPS is more important in action games.

IIRC Jarred already benched rasterisation in his main review, there’s no reason to ignore or pretend FG and MFG don’t exist due to some silly bias.

Is power made by an electric car fake compared to a ICE? It used to be that turbo charged engines were considered gimmicks and not real like a naturally aspirated engine. Well I don’t have to tell you what happened after that.
If we raced a Ferrari 812, against a SF90, the screaming V12 would be smacked by the turbo charged V8 hybrid 10 out of 10 times.

 

[CelicaGT]

IIRC Jarred already benched rasterisation in his main review, there’s no reason to ignore or pretend FG and MFG don’t exist due to some silly bias.

Is power made by an electric car fake compared to a ICE? It used to be that turbo charged engines were considered gimmicks and not real like a naturally aspirated engine. Well I don’t have to tell you what happened after that.
If we raced a Ferrari 812, against a SF90, the screaming V12 would be smacked by the turbo charged V8 hybrid 10 out of 10 times.

The problem with FG and DLSS is that often they just don't work reliably. Raster benchmarks (as stated ad nauseam) give a perfect comparison of baseline performance that can be directly compared across generations and brands. FG and DLSS muddy the water and strictly speaking, cannot be used in a benchmark. I do appreciate the reviewer including them in regards to their proper function and impressions of image quality. FWIW I use DLSS in a few titles, as often it does a better job cleaning up aliasing, even though it does degrade overall quality somewhat (in my opinion).

 

[Mcnoobler]

Just out of interest. Do you have an RTX40/50 series card. Have you tried FG/MFG? Curious to know.

I have, for a couple years now on a 4090. DLSS FG was my favorite feature. Not just for getting beyond CPU bottlenecks/limitations either, but even being the reason early on that you could stay at a native resolution, over upscaling, and use the DLSS FG feature to get to the right level of smoothness. (Now with DLSS 4, upscaling seems to more acceptable for me). Great thing about FG also is no one is twisting your arm to use it, its optional. Turn it off and use more aggressive upscaling instead, all up to the gamer themselves individually.

Last time I had an issue with DLSS FG, of any kind, was Forza 5 at FG launch sometime in 2023 I believe. I mostly game on a Dualsense Edge with haptic feedback in Cyberpunk, just an amazing experience over 2 years that I would pay $1600 to do it all over again (I'm low to mid class too). It always seemed like latency was improved over all vs my PS5 which is the platform I came from. Yes since no one can give these GPUs proper praise, or its tech, I will from 2 years of personal use and say the experience has legitimately been amazing. Glad I'm alive to see what they come up with next.

 

[artk2219]

IIRC Jarred already benched rasterisation in his main review, there’s no reason to ignore or pretend FG and MFG don’t exist due to some silly bias.

Is power made by an electric car fake compared to a ICE? It used to be that turbo charged engines were considered gimmicks and not real like a naturally aspirated engine. Well I don’t have to tell you what happened after that.
If we raced a Ferrari 812, against a SF90, the screaming V12 would be smacked by the turbo charged V8 hybrid 10 out of 10 times.

Thats not the best analogy though, and this ones not necessarily much better

It can be like cooking with a solar oven vs a regular oven. On a warm day, with a bright sun, if the conditions are right, it will cook you an excellent meal. If its night, cloudy, rainy , windy, or extra cold, you're really going to be wishing you had access to a real oven. Your mileage varies much more with FG and MFG, its not always the same thing as what you can wring out with traditional rendering, and the results are always better if you have more power from a better starting source to begin with.

 

[DS426]

...

Will say Nvidia needs to go to a longer release window of new gpu's as this "bump" was lame. i'd of ratehr went another yr and got a meaningful improvement that wasn't just MFG which only works in some titles & has serious downsides.

Q1 2025 NV Earnings call, Jensen Huang said that they were moving to an annual AI GPU release cadence. So, if they can't even make a meaningful bump in two years on gaming GPU's, what does that tell you??

We said it when the 40 series launched: gaming is a noticeable 2nd place for green, and the gap is only growing wider.

 

[Gaidax]

Thats not the best analogy though, and this ones not necessarily much better

It can be like cooking with a solar oven vs a regular oven. On a warm day, with a bright sun, if the conditions are right, it will cook you an excellent meal. If its night, cloudy, rainy , windy, or extra cold, you're really going to be wishing you had access to a real oven. Your mileage varies much more with FG and MFG, its not always the same thing as what you can wring out with traditional rendering, and the results are always better if you have more power from a better starting source to begin with.

That's why as intelligent beings we can make a conscious and objective decision to use a solar oven on a warm day with a bright sun, and not otherwise. That's why it is an option.

There is this almost maniacal focus on worst case scenarios, but what about the best-case ones? Or even the good case?

I'm kind of tired of bringing this up again and again, but we all know that Frame Generation does not work well with low base FPS, been there done that - I get it. But it is a whole different story when you use Frame Gen with 50FPS baseline and even more so at 90FPS baseline.

So, let's not sit here pretending as if you need some perfect "solar oven" situation to use Frame Gen to a good effect, it's quite reverse, as a matter of fact - things need to be really bad to begin with for Frame Gen not to be usable.

 

[waltc3]

I'm waiting on the next AMD GPU release specs & reveal before I do anything...😉 Meanwhile, my 6900XT is running great with all of my games, and the newer AFMF2 drivers coming out now (25.1.1 is the latest and second official iteration of AFMF2) really made a noticeable difference at my res, which is 4k on everything.) I'm not as interested in this 5000 RTX series, even though I was slightly interested in the 4000 RTX series, but not interested enough to buy one at nVidia's ridiculous pricing. This looks like just a slight iteration on RTX 4k, to me. But I think only the cheaper 5ks will be available in any sort of quantity and that it will be awhile before the top end $2k-$3k + versions come out in any sort of quantity.

I thought the review here was fine, and of course I would expect nVidia to try and distract from the classical benchmarks to toss out a lot of somewhat complex "new" benchmarks in order to distract from the old standbys they don't want you to think about just yet...😉 Marketing 101. Looks like nVidia has a ways to go on driver optimization for the "new" architecture *cough*...😉 I see it as the RTX 4k architecture with a few hardware and software tweaks nVidia doesn't quite support, yet...IIRC, the RTX 4k was also released sort of half-baked, too. Apparently, though, people are happy with them now, so we will see. But anyway...

I'm looking for something really sweet from AMD for under $500...in about a month or so. I hope so!

 

[artk2219]

That's why as intelligent beings we can make a conscious and objective decision to use a solar oven on a warm day with a bright sun, and not otherwise. That's why it is an option.

There is this almost maniacal focus on worst case scenarios, but what about the best-case ones? Or even the good case?

I'm kind of tired of bringing this up again and again, but we all know that Frame Generation does not work well with low base FPS, been there done that - I get it. But it is a whole different story when you use Frame Gen with 50FPS baseline and even more so at 90FPS baseline.

So, let's not sit here pretending as if you need some perfect "solar oven" situation to use Frame Gen to a good effect, it's quite reverse, as a matter of fact - things need to be really bad to begin with for Frame Gen not to be usable.

Yes, but i think this is all mainly a reaction to all of the marketing material making it seem like FG and MFG are the only way to have more performance in the future, regardless of their downsides. That also feels wrong headed, its a nice option to have, and it most definitely helps in situations, like you said, its just a new option. It doesnt help that they're also trying to sell the idea that generated frames will be exactly the same as natively processed frames. We know thats not always the case, there is just alot of gas lighting going on from the marketing side. Which is compunded with a lack luster increase in games that dont, or cant, take advantage of any of those new features. Im sure we'll forget about all this hubub like we did with the RTX 2000 launch, and "just buy it" anyway, because at the end of the day, its whats available. That doesn't mean we shouldnt call attention to any flaws we find in the process though.

 

[JarredWaltonGPU]

That's why as intelligent beings we can make a conscious and objective decision to use a solar oven on a warm day with a bright sun, and not otherwise. That's why it is an option.

There is this almost maniacal focus on worst case scenarios, but what about the best-case ones? Or even the good case?

I'm kind of tired of bringing this up again and again, but we all know that Frame Generation does not work well with low base FPS, been there done that - I get it. But it is a whole different story when you use Frame Gen with 50FPS baseline and even more so at 90FPS baseline.

So, let's not sit here pretending as if you need some perfect "solar oven" situation to use Frame Gen to a good effect, it's quite reverse, as a matter of fact - things need to be really bad to begin with for Frame Gen not to be usable.

Amen! The "fake frames" crowd isn't trying to be objective and truthful, they're just yelling the same meme very loudly and pretending it's always correct. FG and MFG can and do break down at a certain point for sure.

Hello, Indiana Jones and the Great Circle, with Full RT enabled and DLSS Quality mode? That works, on an RTX 4080 Super, and gives about 60 FPS. Turn on framegen and you exceed the VRAM and it breaks. On the 5080, it just stopped responding. On the 4080 Super, it dropped to 18 FPS. But on the 4090, it went from 80 FPS to 110 FPS... which still isn't a great result.

That's one where I'd say the benefits of FG aren't high enough, at least with the old DLSS3 FG in that particular game. I feel like FG needs to give about a 40~50% boost to FPS, minimum, or it's a drawback. The increased latency overcomes the smoothing effect. And Nvidia (and AMD and Intel) know this and are working to make framegen algorithms run faster and look better.

In Black Myth Wukong, 4080 Super goes from 36 FPS to 58 FPS with framegen. Even though input sampling is now at 29 vs. 36, overall? I'd say it plays better that way. But what it really needs is a higher base FPS, like a 4090 going from 48 FPS to 77 FPS. Either one is a 60% increase and acceptable — still not great but better than playing without, depending on subjective feelings.

Now, what about the new and improved framegen? The 4080 Super goes from 32 FPS base to 60 FPS with framegen in Cyberpunk 2077 (RT-Overdrive enabled with DLSS Quality Transformers). That's an 88% improvement and definitely feels better. It still probably only feels like about 40~45 FPS to me, but for sure it feels better than 32 FPS. Because going from 32 input samples per second to 30 samples per second isn't a huge drop, but visually 60 looks much, MUCH smoother than 32.

TLDR: Old DLSS3 framegen was more problematic as it often didn't boost performance enough. FSR3 did better on performance boosting but was severely lacking in image quality. DLSS4 framegen seems to get both the performance boost and image quality "right." And it will still depend on the game and implementation. (Hogwarts Legacy with all the RT stuff enabled still sucks for baseline performance.)

 

[snemarch]

I wish the media industry would stop testing and reporting about framegen, full stop.

It's a crutch, and when Jensen, with a straight face, can spout insane nonsense like a 5070 outperforming a 4090, that's really all it should take for people to realize why this needs to be boycotted.

We all know it doesn't help gameplay experience if the source framerate is too low. If it's high enough, jeez, just get a variable-refresh monitor.

If we accept that framegen and ever-increasing power budgets are necessary to increase visual fidelity, innovation is dead. It's a bloody cop-out.

 

[Gaidax]

I wish the media industry would stop testing and reporting about framegen, full stop.

It's a crutch, and when Jensen, with a straight face, can spout insane nonsense like a 5070 outperforming a 4090, that's really all it should take for people to realize why this needs to be boycotted.

We all know it doesn't help gameplay experience if the source framerate is too low. If it's high enough, jeez, just get a variable-refresh monitor.

If we accept that framegen and ever-increasing power budgets are necessary to increase visual fidelity, innovation is dead. It's a bloody cop-out.

Adeptus Mechanicus vibes.

So, the innovation that does not conform to some dogma is heresy and not innovation?

C'mon man.

 

[Peter_Bollwerk]

Almost everything about this generation/launch feels half baked.

• the performance/spec gap between the 5090 and 5080 is larger than past generations
• low launch supply
• too much focus on fake frames and not on raw power

I can't help but feel this is what happens when Nvidia has no competition from AMD or anyone else.

 

[kyzarvs]

What should he bench? What’s a real frame that your card renders then? Does it render real time footage of your driveway or revolution of the planet?
How about you articulate what is the actual issue you have with frame generation without regurgitating what you heard elsewhere.

A real frame is one that can be influenced by the users' input. Frames that ignore it are just filler.

 

[thestryker]

On one hand I'm somewhat surprised by the driver stumble, but on the other not particularly. There has been high driver overhead with nvidia for a long time and as cards keep getting faster this was always going to become more apparent. Hopefully they'll work on the overall performance as they shift from NVCP to the App.

I feel like we need a different description for frame generation frame rates since they don't actually provide the same experience as just an increased frame rate. Either that or include latency for every result.

Much like upscaling I don't think frame generation belongs in a standard review (perhaps their own separate segment, but that seems like a lot of extra work with limited return on a per card basis). There's simply too much individual subjectivity involved with the technology.