Discussion Nvidia Blackwell in Q1-2025

Page 62 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hulk

Diamond Member
Oct 9, 1999
4,941
3,372
136
Some questions for those of you with 4090's.
Are you considering upgrading to the 5090 or another GPU in the near future?
Why or why not?

It seems as though the 4090 can provide quite good frame rates at 4k in pretty much every game available today so I'm wondering what people with 4090's might be considering an upgrade?

I don't know a lot about gaming and GPU requirements so I'm curious to hear from actual owners of these super powerful GPU's.
 

moonbogg

Lifer
Jan 8, 2011
10,667
3,197
136
I think the 5070Ti is the most appealing one. Should basically be a 4080 Super + maybe 5 or 10% for a slightly reduced price. It's still within that traditional high-end price category and should still offer good high-end performance. I'm still not about all these $1000+ cards.
 

gdansk

Diamond Member
Feb 8, 2011
3,768
6,020
136
I think the 5070Ti is the most appealing one. Should basically be a 4080 Super + maybe 5 or 10% for a slightly reduced price. It's still within that traditional high-end price category and should still offer good high-end performance.
That would be good. But I'm a bit concerned by the lack of compute. Is it even faster than the 4080? We'll see. It depends on Nvidia sandbagging the boost clocks more than usual.
 

Artorias

Platinum Member
Feb 8, 2014
2,184
1,486
136
While technology becomes generally more accessible it seems the GPU industry continues to move backwards.

Crypto got Nvidia drunk on all that money and now consumers are paying the price.

Lack of competition is killing the industry.

Another overpriced space heater sucking god knows how much energy, all ready to serve you fake AI frames because our hardware is not up to snuff with a CEO cutting costs to feed the bottom line.
 
Reactions: igor_kavinski

gdansk

Diamond Member
Feb 8, 2011
3,768
6,020
136
CEO cutting costs to feed the bottom line.
GB202 isn't cutting costs and it's the only one that increased MSRP?
It is disappointing that Nvidia chooses to market with multiple frame generation but I don't think it's so much about cost cutting.
At worst it is the first time Nvidia hasn't executed well in many years. Is it just another Turing? It remains to be seen but I'm still hopeful it's pretty good.
 

Hulk

Diamond Member
Oct 9, 1999
4,941
3,372
136
While technology becomes generally more accessible it seems the GPU industry continues to move backwards.

Crypto got Nvidia drunk on all that money and now consumers are paying the price.

Lack of competition is killing the industry.
Sounds like Intel from about 15 years ago.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,653
6,113
136
And this is why you cannot take whatever Jensen says at face value. Just because he's the CEO of the most valuable company in the world doesn't automatically mean whatever comes out of his mouth is automatically correct.

Yeah, predicting several future frames without an endpoint would not end well. They have a hard enough time doing one good one between, when the have a start point and an endpoint. They would get ridiculously out of sync on a regular basis rendering multiple predictive frames in a row, without an endpoint. All those resyncs would induce some kind of weird micro-stutter I bet.

Plus that DF hands on 50 series video from before, basically says that the big hit from latency comes from buffering the second frame, and going to 3x and 4x only increase latency a bit more. So it's behaving just like it always has with that big latency hit from turning it on (from the wait to buffer the second frame). If it was predicting future frames, the latency hit would be MUCH smaller, potentially negligible.
 
Reactions: Saylick

Hulk

Diamond Member
Oct 9, 1999
4,941
3,372
136
This is new..

“When we asked [Jensen Huang] how DLSS 4 multi frame generation works and whether it was still interpolating, Jensen boldly proclaimed that DLSS 4 "predicts the future" rather than "interpolating the past." That drastically changes how it works, what it requires in terms of hardware capabilities, and what we can expect in terms of latency.”
Hmm. So my first thought may have been correct. DLSS will run a few frames into the future using AI prediction. If the prediction is good enough and the time interval small enough there *shouldn't* be a visible jump from the last predicted frame to the first new computed frame. I did the math a few pages ago on how this might work.
 
Reactions: MoogleW

Golgatha

Lifer
Jul 18, 2003
12,310
790
126
Some questions for those of you with 4090's.
Are you considering upgrading to the 5090 or another GPU in the near future?
Why or why not?

It seems as though the 4090 can provide quite good frame rates at 4k in pretty much every game available today so I'm wondering what people with 4090's might be considering an upgrade?

I don't know a lot about gaming and GPU requirements so I'm curious to hear from actual owners of these super powerful GPU's.
I'm considering it.

Conditions - I can actually purchase one for $2000. So no AIB shenanigans (best bet here is FE or Zotac) and/or tariffs. Also, I want to see 4090 raster versus 5090 raster performance at 1440p to evaluate the performance increase I should expect to see. I would still purchase one at my local Micro Center on day one, but I'll leave it in the shrink wrap until I see the 1440p numbers benchmarked on a 9800X3D. I already have a 1000w power supply and my UPS has about 250w of headroom to spare, so I'm golden there.

Why - If performance increase is at parity or better with the cost minus the value I think I can get from selling my 4090, I will want the new shiny.
 
Reactions: blckgrffn

Heartbreaker

Diamond Member
Apr 3, 2006
4,653
6,113
136
Hmm. So my first thought may have been correct. DLSS will run a few frames into the future using AI prediction. If the prediction is good enough and the time interval small enough there *shouldn't* be a visible jump from the last predicted frame to the first new computed frame. I did the math a few pages ago on how this might work.

I don't care what Jensen said. It's not going to work that way. DF hands on, is reporting the same big latency hit when you turn it on for buffering the current frame to interpolate between the current and past, then insert them, then finally release the current frame (quite a bit later).

When third part reviewers get full access, they will show it still has the big latency hit...
 
Reactions: NTMBK and Saylick

Hulk

Diamond Member
Oct 9, 1999
4,941
3,372
136
Yeah, predicting several future frames without an endpoint would not end well. They have a hard enough time doing one good one between, when the have a start point and an endpoint. They would get ridiculously out of sync on a regular basis rendering multiple predictive frames in a row, without an endpoint. All those resyncs would induce some kind of weird micro-stutter I bet.

Plus that DF hands on 50 series video from before, basically says that the big hit from latency comes from buffering the second frame, and going to 3x and 4x only increase latency a bit more. So it's behaving just like it always has with that big latency hit from turning it on (from the wait to buffer the second frame). If it was predicting future frames, the latency hit would be MUCH smaller, potentially negligible.
The end point is the next computed frame.

Think of it this way. Assume every other frame is predicted so the GOP (group of frames) would be C-P-C-P... on and on. C is Computed and P is predicted.

If it takes half the time to predict a frame using AI compared to computing it and if the fps is 100, or 1ms per frame computationally then instead of taking 2ms to display two frames it now takes 1ms for the C frame and 0.5ms for the P frame or a total of 1.5ms vs 2ms. Now each frame only take 0.75ms to render and the new frame rate is 133.33fps, or 33.3% faster.

On the positive side, not much really changes in 1ms so if the prediction is good I doubt it would be noticeable. This assume good AI prediction. Also note that as native frame rate decreases then the time between frames increases and accurate prediction becomes more difficult. I would think at low frame rates perhaps 1 or maybe 2 frames would be predicted while at high native frame rates it might be possible to predict 3 or 4 frames between computed frames.

This is a lot of fun math to be done here at various native frame rates to see how this would work. All you have to do is decide/figure out how far into the future gameplay can be predicted. 1ms? 2ms? 4ms? One you pick a number and a native framerate, knowing the how much faster frames can be predicted rather than computed you can figure the "fakery" frame rate.

Maybe if I have the time I'll make up a little spreadsheet tonight with some numbers. Might be fun to play with. Good brain exercise to keep the neurons firing reliably.
 

Golgatha

Lifer
Jul 18, 2003
12,310
790
126
I don't care what Jensen said. It's not going to work that way. DF hands on, is reporting the same big latency hit when you turn it on for buffering the current frame to interpolate between the current and past, then insert them, then finally release the current frame (quite a bit later).

When third part reviewers get full access, they will show it still has the big latency hit...
Latency increases and possible graphical errors are why I personally won't ever take fake frames into consideration for my purchasing decision. I only care about pure raster and (as a distant 2nd) RT performance, as those are features which work with every game.
 
Reactions: Jaskalas and Elfear

gdansk

Diamond Member
Feb 8, 2011
3,768
6,020
136
Was Jensen actually referring to DLSS4 when combined with Reflex 2 which seems to be predicting some motion/warp?
 

Saylick

Diamond Member
Sep 10, 2012
3,798
8,666
136
The end point is the next computed frame.
This is where I think it's not true. I think DLSS 4 frame gen is strictly frame interpolation, where it's interpolating up to 3 intermediate frames instead of DLSS 3's 1. For the interpolation to work, it needs the before and after frame and the only way to get the after frame is to wait for it to be rendered.

I drew this up in 5 minutes based on how I think it works. Boxes with numbers are "real" frames. Boxes with "i" are interpolated frames. Interpolation can only be performed after you have the "real" before and after frame.
 

pj-

Senior member
May 5, 2015
499
276
136
I don't care what Jensen said. It's not going to work that way. DF hands on, is reporting the same big latency hit when you turn it on for buffering the current frame to interpolate between the current and past, then insert them, then finally release the current frame (quite a bit later).

When third part reviewers get full access, they will show it still has the big latency hit...

IDK why you are so insistent it's interpolation when nothing indicates that it is.

"we render 2 million pixels (1080p) and predict the other 32 million (4 frames of 4k)" - paraphrasing Jensen from the presentation




Not sure how they could be more clear that it's not interpolating between two known frames

To me it makes no sense for it to be interpolating between two already generated frames. That isn't impressive and the results should be basically perfect (which they are not). LCD tv's have been doing that for decades without the benefit of motion vectors and z buffers and whatever else DLSS would have access to.

The additional latency doesn't mean they are holding back a frame. It could be for frame pacing purposes or something. It's not magic and it's probably not going to be a feature I use, but I would bet money that it isn't interpolation
 
Reactions: MoogleW

GodisanAtheist

Diamond Member
Nov 16, 2006
7,479
8,253
136
This is where I think it's not true. I think DLSS 4 frame gen is strictly frame interpolation, where it's interpolating up to 3 intermediate frames instead of DLSS 3's 1. For the interpolation to work, it needs the before and after frame and the only way to get the after frame is to wait for it to be rendered.

I drew this up in 5 minutes based on how I think it works. Boxes with numbers are "real" frames. Boxes with "i" are interpolated frames. Interpolation can only be performed after you have the "real" before and after frame.
View attachment 114461



In all seriousness I think you nailed it. No matter how much leather tanning fumes someone might have huffed, no one believes NV is *telling the future* with DLSS4....
 
Reactions: Hulk

poke01

Diamond Member
Mar 8, 2022
3,040
4,031
106
Watched the new digital foundry video, RTX 40 series is also getting lots of improvements which is nice to see
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |