NVIDIA GeForce RTX 4090 Founders Edition

[Upgrayedd]

I thought 95C was supposed to be the new 65C?
Does that mean this 72C is the new 42C?

 

[mouacyk]

I mean, you can see what it can do in games that can somewhat take advantage of its strong suits. 100% over a 3080 is nuts.

At >2x the cost. Unremarkable.

 

[N/A]

Yup. By the way, Turing was 38% faster than Pascal. Ada is 45% faster than Ampere.

Where does this number come from again 45%?? It's Double the performance the 3090 in non CPU limited situations and this is without the new feature of frame insertion.

 

[TheinsanegamerN]

That should be credited to TSMC, not NVidia?

Intel's ARC GPUs are a prime example of the exact opposite. The node is not the only thing that matters, GPU arch design is just as important.

 

[Chrispy_]

@W1zzard: Holy shit, you have the ONLY DLSS 3.0 frame generation results on the internet right now.
Good job, nice review; Crack open a cold beer and put your feet up!

As for the 3090, it's a beast, as expected - but it's not the model that 99.9% of people are interested in and we don't know what AMD's offering will be. For now it's just the first data point in a very incomplete next-gen dataset and I cannot wait to see how the 4070 or any RDNA3 cards compare.

 

[ARF]

I really wish SLI and multi-GPU setups would have evolved.

RTX 4090 is your new "dual-GPU replacement".
The RTX 4080-16 will be nvidia's new single-GPU top dog.
While the RTX 4080-12 - well, who knows, there is no such animal

 

[clopezi]

@W1zzard: Holy shit, you have the ONLY DLSS 3.0 frame generation results on the internet right now.
Good job, nice review; Crack open a cold beer and put your feet up!

As for the 3090, it's a beast, as expected - but it's not the model that 99.9% of people are interested in and we don't know what AMD's offering will be, so for now it's just the first data point in a very incomplete next-gen dataset. I cannot wait to see how the 4070 or any RDNA3 cards compare.

It's a supperb work! For things like this, Techpowerup it's my first tech place to read

 

[Frick]

Would it be interesting to add 4K Vsync 60Hz to this?

For power draw I meant.

 

[Gameslove]

@W1zzard
4K monitor, you can trying to enable a dynamic super resolution (DSR) - 4K into 8K.

 

[Shazamy]

Thanks for an amazing review W1zzard, as always, much appreciated!
The DLSS FG video was an awesome and unexpected inclusion!
I can't believe how much more power efficient and quiet this card is, really impressed!
I was going to get the MSI water cooled 4090, but with these temps I may not have to. Looking forward to the custom design card reviews!

 

[BigMack70]

Where does this number come from again 45%?? It's Double the performance the 3090 in non CPU limited situations and this is without the new feature of frame insertion.

View attachment 265050

Um.... from the portion of the review conclusion that I've been criticizing (because it makes no sense from the data)?

I'll quote it again, for reference. Emphasis added.

For a majority of gamers, the "classic" raster performance is very important though—highest settings, RT off, DLSS off—so we made sure to extensively test this scenario using 25 games at three resolutions. The GeForce RTX 4090 achieves incredible performance results here: +45% vs RTX 3090 Ti. Yup, 45% faster than last generation's flagship—this is probably the largest jump Gen-over-Gen for many years. Compared to RTX 3080 the uplift is 89%—wow!—almost twice as fast. Compared to AMD's flagship, the Radeon RX 6950 XT, the RTX 4090 is 64% faster. Somehow I feel that after RDNA2, Jensen said to his people "go all out, I want to conclusively beat AMD next time."

I'm not critiquing the card. I'm critiquing a section of this review's conclusion that makes ZERO sense. The data in this review does not support these statements with respect to "classic" raster performance. The data in this review suggests that this card is right in the middle of the pack with regards to new GPU architectures launched over the past 10 years of GPU launches and is just fine.

This card's performance is excellent - as expected - without needing to use unwarranted hyperbole and unearned effusive praise in comparison to what previous architectures achieved at their respective launches. It sounds much more like marketing than product review.

 

[Gica]

The price is not high for a monster. You have everything you want for 4K@100+ FPS gaming, DLSS 3.0 and full enc/dec support. For Content Creation, the former king 3090Ti turns into a poor kitten in front of the new king. Here, at least 100% performance increase is expected.

 

[evernessince]

More bad information. The 1060 6GB that is being compared was $299. The 2060 6GB was $350. I noted this in my post. The 1060 4GB you speak of was $250.

View attachment 265046

View attachment 265048




For a 20% price increase you got +45% in 1080P and +67% in 4K. I paid $309 for my 2060 KO which is a few percent faster still.

The only bad information here is that you are implying that $350 is the price the card was broadly available for. $350 was the founder's edition price which almost no one saw. Aside from the scalping, AIB pricing started at $400+ as was usual when comparing founders edition to non-founders edition.

Mind you that's comparing the 6GB 2060 to the 6GB 1060. Zero increase in VRAM. The 12GB 2060 cost a whopping $600 at time of review on TechPowerUp.

You paid $309 for a 2060 KO, which launched March 3rd 2020, a whopping 14 MONTHS after the Jan 7th 2019 launch of the 2060. Suffice to say, it's extremely misleading to compare late generation prices to MSRP.

You admitted you own a 2060 and I believe your ownership is clouding your judgement in this case. Even when you take the best turning SKU value wise in the best possible light it's at best underwhelming.

On price yes, but that's not what he was comparing: he was comparing the % gains of the 2060 over the 1060 with the % gains of the 3060 over the 2060.

He was replying to my comment referencing performance per dollar. I have to assume he's replying on topic. Otherwise I could care less about off-topic arguments.

 

[Dimestore]

DLSS3 'Frame Generation' makes no sense to me. Take two frames and interpose a generated frame in between them. What happens? Your FPS goes up, sure, but your *latency* goes up as well! Latency is time between action and reaction, from the time you move the control to the time the movement hits your eyes. If you are generating an extra frame for every two frames that are not controlled by you, then you are pushing latency up by 1/3. Does this make sense to anyone else?

 

[clopezi]

DLSS3 'Frame Generation' makes no sense to me. Take two frames and interpose a generated frame in between them. What happens? Your FPS goes up, sure, but your *latency* goes up as well! Latency is time between action and reaction, from the time you move the control to the time the movement hits your eyes. If you are generating an extra frame for every two frames that are not controlled by you, then you are pushing latency up by 1/3. Does this make sense to anyone else?

Latency it's reduced in all tests, they have dedicated technology for it.

I don't know in competitive games, but on single player games, latency it's heavely reduced.

 

[terroralpha]

this review is worthless. why would you test a high end GPU on a budget CPU? you are knee capping the performance. a lot of the benches here, even those at 4K, are CPU limited. you guys couldn't spring for a 5800X3D? or an intel bench?

no one who is the market for this GPU is going to be using a 5800X system.

everyone else used ryzen 7000 or high end intel. but you guys published this junk

TWO TO FOUR TIMES FASTER!!!

in some case, yes, it does go 4x faster. the reason you don't see that here is because, as mentioned above, these benches are being run on a ryzen 5800

 

[W1zzard]

DLSS3 'Frame Generation' makes no sense to me. Take two frames and interpose a generated frame in between them. What happens? Your FPS goes up, sure, but your *latency* goes up as well! Latency is time between action and reaction, from the time you move the control to the time the movement hits your eyes. If you are generating an extra frame for every two frames that are not controlled by you, then you are pushing latency up by 1/3. Does this make sense to anyone else?

You are absolutely right, that's why DLSS 3 requires devs to add Reflex support to their game

 

[Dimestore]

Latency it's reduced in all tests, they have dedicated technology for it.

I don't know in competitive games, but on single player games, latency it's heavely reduced.

Where is this addressed? I see no mention of it. I only see reviews use frame rates as performance indicators.

 

[clopezi]

this review is worthless. why would you test a high end GPU on a budget CPU? you are knee capping the performance. a lot of the benches here, even those at 4K, are CPU limited. you guys couldn't spring for a 5800X3D? or an intel bench?

everyone else used ryzen 7000 or high end intel. but you guys published this junk



in some case, yes, it does go 4x faster. the reason you don't see that is because, as mentioned above, these benches are being run on a ryzen 5800

View attachment 265064

LTT numbers are strange. Those numbers without DLSS enabled are not seen on any other review...

 

[W1zzard]

@W1zzard: Holy shit, you have the ONLY DLSS 3.0 frame generation results on the internet right now.
Good job, nice review; Crack open a cold beer and put your feet up!

lol really? and I felt bad because I had only two games

 

[3x0]

Latency it's reduced in all tests, they have dedicated technology for it.

I don't know in competitive games, but on single player games, latency it's heavely reduced.

 

[TheoneandonlyMrK]

DLSS3 'Frame Generation' makes no sense to me. Take two frames and interpose a generated frame in between them. What happens? Your FPS goes up, sure, but your *latency* goes up as well! Latency is time between action and reaction, from the time you move the control to the time the movement hits your eyes. If you are generating an extra frame for every two frames that are not controlled by you, then you are pushing latency up by 1/3. Does this make sense to anyone else?

Yeah,. But more frames though
Sorry I couldn't help myself.

I agree with your sentiments though virtual high FPS isn't for me.

Wow did I just delete a rant, in short wtaf I don't like ai generated frames. or its equivalents

Secondly htaf, does a maximum gain of 40% or close max in 4k and 12% at 1080p turn into 100% or 2 to 4 time's the performance, just how niche is the use case that gain is on considering its a gaming card that's some clear hyping by Nvidia.
I do want to know though, where does 4x kick in

 

[evernessince]

Latency it's reduced in all tests, they have dedicated technology for it.

I don't know in competitive games, but on single player games, latency it's heavely reduced.

No latency is absolutely not reduced by DLSS's AI frame insertion. Given that it requires you to wait for the next frame to be rendered before creating the AI generated frame it will always carry a latency penalty until they remove that requirement.

This is why Nvidia require that Reflex be enabled with the frame insertion, to hide a portion of the latency penalty. If you disable frame insertion and enable reflex you will get lower latency.

DLSS3 'Frame Generation' makes no sense to me. Take two frames and interpose a generated frame in between them. What happens? Your FPS goes up, sure, but your *latency* goes up as well! Latency is time between action and reaction, from the time you move the control to the time the movement hits your eyes. If you are generating an extra frame for every two frames that are not controlled by you, then you are pushing latency up by 1/3. Does this make sense to anyone else?

I assume there's a sweet spot between 60 FPS and 144 FPS where it makes sense for some people to enable it. Above 144 FPS it doesn't make much sense to enable as smoothness is already very good and the latency penalty would outweigh any diminishing benefits above that frame-rate. Below 60 FPS and the latency penalty would be quite sizable, maybe for strategy games that aren't real time but even then it could still be annoying and those are the games least likely to benefit from additional motion smoothness.

Hopefully improvements come to the tech but until then I feel like it's going to be highly preferential as to whether it's worth enabling.

 

[Footman]

Yes very good but.....

Nice Review. I noticed Techspot remarked that the 4090 was cpu bound with the 5800X3D in their test machine. Your test machine is running a 5800X, so I wonder if the performance of the 4090 might be higher at lower resolutions with a faster cpu???

 

[GerKNG]

in some case, yes, it does go 4x faster. the reason you don't see that here is because, as mentioned above, these benches are being run on a ryzen 5800

you show us the LTT benchmark where they forgot to disable DLSS.
this is not a real result.