Nvidia GeForce RTX 5070 review: No, it’s not “4090 performance at $549”
- Thread starter JournalBot
- Start date
The time stamp is 12:47, the marketing is intentionally deceptive. Jensen claims on stage that it can predict the future.
In a nutshell, the second frame is rendered and held for a bit while the frame gen tech does its thing.
Last edited:
Upvote
12
(12
/
0)
David Mayer
Wise, Aged Ars Veteran
Thanks, this is really quite absurd, the tech to "predict the future" actually does exist and is used now for VR. Why TF are they deliberately adding latency?
Upvote
3
(5
/
-2)
Frame gen is using interpolation. From my understanding the order of Nvidia frame gen is basically this:
1. real frame is displayed on screen
2. next real frame is rendered and held in queue
3. 1 to 3 frames are generated between the last real frame and next real frame
4. generated frames are displayed on screen
5. real frame in queue displayed on screen
Repeat
Upvote
19
(19
/
0)
"Adding latency" is exactly what it does, which is why Nvidia Reflex needs to be enabled to reduce it again... It's why a lot of people don't like framegen. Digital Foundary goes into a lot of detail about this in several videos, but here's their review of the 5070. The image above (like many Nvidia presentations) is very misleading. What it should show at the end is "Rendered Frame 2." It IS interpolating between frames, NOT extrapolating and it adds significant latency - upwards of 60ms in some cases.
Upvote
21
(22
/
-1)
Upvote
6
(6
/
0)
Thanks for the kind advice.
In the future before ranting about linguistic trends please make sure you're factually correct.
Upvote
7
(14
/
-7)
It looks like it's in a pretty miserable state. I wouldn't necessarily extrapolate from any single game to larger trends about the industry.
Upvote
1
(1
/
0)
I would guess that we're more so seeing relatively linear improvements in graphics technology, while the tasks they're running up against are scaling exponentially. Especially when you start looking at things like ray tracing.
Upvote
4
(4
/
0)
David Mayer
Wise, Aged Ars Veteran
So don't trust communications directly from the source, got it.
Upvote
-5
(5
/
-10)
Yeah, it's a misconception nvidia purposefully cultivated, to make the idea of frame generation seem more appealing.
Jensen, at CES: "it [the 5090] can predict the future".
I mean, I can predict the future too, just give me a one second latency.
Upvote
11
(12
/
-1)
I was going to say that I hope Nvidia is ready to mend fences with the gaming community after the AI bubble bursts, but it looks like the economy (at least in the US) is headed for a major recession if not a depression. The cards become more expensive and more power-hungry every year. AAA gaming is becoming a rather expensive hobby.
Upvote
-3
(2
/
-5)
To be fair, it still makes sense as an option. Maybe that additional latency is a problem in certain games, but it isn't in all games.
Upvote
-1
(2
/
-3)
The reason nobody is saying those words is because the long-term hope+goal is that AI will improve. I don't think it's an accident that everyone is leaning on AI as a method of improving visual quality. It's an effort to return to the era when GPU gains were faster and you could get more immediate improvements.
Now, of course you are right that there are game-specific tradeoffs and visual errata for those improvements that may not exist otherwise. But DLSS and even FSR3 have advanced to the point that they're a useful way to boost performance on low-end systems, even if they come with some visual tradeoffs.
Some technologies, like MFG, really rely on high frame rates to start with. Others, like DLSS / FSR, can help boost lower frame rates without interpolating "fake" frames. But taken collectively, it's clear that the GPU industry has pivoted to AI in the hopes that it can be used to collectively improve rendering, even if not every experiment pans out. And not every experiment will. It's easy to forget now, but once upon a time there were a lot of companies competing to be "the future" of 3D graphics, and some of those approaches were quite off-the-wall. The original NV1 worked on quadrilaterals, not triangles. Voxel-based games had a bit of a moment at the dawn of 3D. There were companies like S3/Virge, with their own "Metal" API, 3dfx had Glide.
The usefulness of DLSS suggests to me that AI will have a permanent role in game rendering going forward, for boosting low-end GPU performance if nothing else.
Upvote
6
(8
/
-2)
I'm just hoping that the new card can help drop some of the prices further down the market. Obviously the 50 series cards aren't really having any effect due to a lack of availability.
Upvote
1
(2
/
-1)
I'm sure someone out there is both crazy and rich enough to try gaming on a DGX SuperPOD...
Upvote
4
(4
/
0)
I was looking into 5080 ti for a gaming PC I’m hoping would last 5+ years, but now I’m less sure. For some reason Maingear doesn’t offer 4080 or 4070
Upvote
-7
(0
/
-7)
I'm sorry you got tripped up on the extrapolation / interpolation thing. I see where you were coming from.
I would say in this case that Nvidia sacrificed technical accuracy on the altar of making more impressive top-line comparisons. "Predict the future" sounds so much better than "Frame interpolation," especially when some people are aware that they hate interpolated frames on TVs thanks to motion smoothing. Ditto for the "It's an RTX 4090, but an RTX 5070."
It was instantly obvious that the RTX 5070 would never and could never match the RTX 4090 in raw performance -- and it can't. It doesn't even come close.
Upvote
12
(13
/
-1)
araczynski
Ars Scholae Palatinae
I'm MORE than happy with the 4k performance my 3900 is kicking out. I can wait till the 6000 series (or AMD equivalent whenever).
Upvote
0
(1
/
-1)
Availability isn't "tricky" it's NONEXISTANT, especially for the "MSRP" model at $549. Don't expect to get one for less than $700...at which point, it will be thoroughly trounced by the 9070XT and even the 9070 will outperform this piece of garbage! This card is DOA. If it ever actually arrives.
Upvote
2
(3
/
-1)
50x0 in a nutshell, but also with driver problems, defective chips, higher power draw.
Sigh, hopefully AMD is better.
Upvote
6
(7
/
-1)
I think we have reached the point where NVidia only really appears to care about the 90 class cards, there are enough AI bro's to sell to by the truck load.
Used to be that the 60 and 70 class would come out close to the 80 release and now 60 & 70 are an after thought by the look of it.
Used to be that the 60 and 70 class would come out close to the 80 release and now 60 & 70 are an after thought by the look of it.
Upvote
4
(5
/
-1)
Pretty much. First party manufacturer claims have a vested interest in presenting their product in the best light possible, even when it's all smoke and mirrors. This is why third party independent reviewers and technical experts are so valuable in validating (and refuting) the numbers.
Upvote
11
(12
/
-1)
If AMD's new RT architecture can significant'y close the gap with Nvidia, this could be a worrisome development for Nvidia.
Ahh, who am I kdding, everyone will just ignore that AMD provides killer price/performance (barring RT where its anywhere from terrible to not half bad) and go buy the shit Nvidia is shoveling because Nvidia is so good at occupying that sweet sweet real estate in our heads.
Interested to see how the 9070s end up falling once the embargoes are all gone.
Upvote
5
(8
/
-3)
What game were you playing that "struggled" at 1080p Medium on an RTX 3080?
I would typically define struggling as "dips below 30 FPS." I have a hard time believing any game "struggles" by that definition on an RTX 3080 @ 1080p Med, but perhaps you have a higher threshold definition for that.
Upvote
25
(25
/
0)
Seems like everyone involved has an incentive to slow development and maximize distribution of thinner and thinner slices of improvement.
If it’s down to greed vs. the limits of material science, I’m guessing that greed is the factor limiting performance improvements.
Upvote
1
(2
/
-1)
I think its about where you're spending your R&D money. Gen over gen they're making huge gains in AI performance. Why would that necessarily mean that raster performance is at the point of diminishing returns?
AI is what is what is getting Nvidia the insane profit margins someone shared earlier, so of course they're dumping their money into developing that and keeping their lead in that domain. Raster/RT performance improvements could to be minimal to nonexistent because they don't need to improve it to sell their enterprise/AI cards and they don't need to do a damn thing to sell their consumer GPUs.
Upvote
4
(4
/
0)
Only at MSRP. Which we know is a fabrication. The very next chart in that video is comparing market pricing and the 5070 is more than halfway down the list, which is topped by the 7800XT and the 7700XT.
Edit: I was somewhat incorrect, it was two charts ahead and the 5070 is closer to 3/4s of the way to the bottom not halfway down. And keep in mind this is comparing current retail pricing in Australia in AUD. But, yea, the Nvidia cards are nowhere NEAR good value.
Last edited:
Upvote
18
(18
/
0)
Why not just display the real frame if it's already there? That's the definition of latency.
Upvote
6
(7
/
-1)
I guess for any playback that isn’t responsive to user input. Seems like a post-processing interpolation tech they got to scale to the point of working in ‘real-time’ that turned into a marketing point.
Your brain rebels against the concept because putting it front and center is scummy and moronic.
Upvote
1
(1
/
0)
6 figures? Isn't that a minimum of $100,000? How many are you buying?
Upvote
3
(3
/
0)
I find it very hard to believe a 3080 was struggling at 1080p medium. I have RX6800, a much less powerful card, running 1440p at mostly high settings in modern games without much issue.
If anything the reviews for the 5000 series have shown very little architectural improvements over the 4000 series since we just see higher power consumption and minimal improvements in frame rates
Upvote
13
(13
/
0)
You've realized that it is actually interpolation and the next step to realize what impact that has on latency. because it has to wait for 2 native frames before giving 4 frames to the user, it has to add 2 native frames of latency to the output.
Upvote
0
(3
/
-3)
I have a 6 year old 2070, play all games at 60hz ultrawide 1440p. I can almost always play at a high enough quality that a further increase in quality is marginal. I can't help but realize I want a 5070 simply due to FOMO and have money to burn. (For example, all I heard about Indiana Jones was how demanding it "is." But I was able to get a perfectly acceptable og DLSS 60fps even setting it slightly higher than the mid settings it gave me as default, and it was absolutely gorgeous (notably in the beginning artifact museum), with amazing shadows and lighting. FOMO! My 2070 would of course be absolutely ridiculed as nearly worthless by the videocardz mafia. Sad.)
Upvote
7
(8
/
-1)
The hilarious thing is Steve absolutely rips into the 5070. They cherry pick a single graph without any of the context.
Upvote
11
(11
/
0)
I was just watching that video. He mentions several times that we should wait for the 9070/9070 xt review tomorrow. I'm taking this advice. Actually, I have half a mind to take the 6th off and camp out at MicroCenter.
Upvote
7
(7
/
0)
Both Steves do, in fact. (Steve from Hardware Nexus and Steve from Hardware Unboxed)
Upvote
8
(8
/
0)