NVIDIA GeForce RTX 4090 Founders Edition

[junglist724]

This graphics card designed for a 4K, 8K gaming, not less. It ridiculous is using this powerful card for 1080p.
4K gaming 5800X quite enough here.

@W1zzard
Thanks for the review. Waiting also 5K, 6K, 8K test. And PCI-E 3.0 vs PCI-E 4.0?

Even with a 12900K/5800X3D/7700X you'll still be CPU limited at 4K in a lot of titles.

 

[evernessince]

Yes it's unremarkable. It's ordinary. Expected. Routine.

Don't believe me? Go look at the flagship card launches for every generation of GPUs. Historically, it's ordinary and unremarkable in every way. It's identical to Ampere, slightly higher than Turing, MUCH lower than Pascal. Higher or lower than Maxwell depending on if you consider the 780 or the 780 Ti the Kepler flagship. Lower than Kepler.

Anyone jumping up and down as if a 45% performance improvement is anything other than expected is drinking Jensen's Kool-Aid and needs to take a break from Nvidia's marketing materials.

To be fair, Turing was statistically the worst Nvidia GPU generation ever. Zero improvement in performance per dollar, only a small performance bump over last gen, and a single digit efficiency increase.

I hope to see the same people that are praising this 45% increase also praise AMD when they achieve the same.

I believe the CPU is limiting the performance the card can achieve, especially at 1080P.

It's worth a new round with the big dogs like 7700x/12900k and 5800x3d

No point until the 13900K and 7950X3D or other Z4 3d variants release.

 

[SOAREVERSOR]

To be fair, Turing was statistically the worst Nvidia GPU generation ever. Zero improvement in performance per dollar, only a small performance bump over last gen, and a single digit efficiency increase.

I hope to see the same people that are praising this 45% increase also praise AMD when they achieve the same.



No point until the 13900K and 7950X3D or other Z4 3d variants release.

People praise AMDs stuff but then all say they hope this will force nvidia to drop prices and when they don't they go out and buy nvidia anyways at the inflated price.

 

[AnotherReader]

Thanks for the timely review, @W1zzard. The performance, while great, is nothing unexpected; the uplift is within the historical norm and less than some other transitions in the past. Even the much panned Turing saw a similar uplift. Considering the node change and the vastly greater clock speeds, the performance increase is rather lacking. On the plus side, raytracing performance has improved again. To summarize:

Pros

• Acceptable performance increase over the 3090
• Greater increase in raytracing performance compared to rasterization

Cons

• If going from 82 SMs to 128 SMs at higher clock speeds gets us a nearly 65% increase, the 4080 16 GB is likely to trade blows with the 3090 Ti
• Increase in performance is far less than Nvidia's claims and the vast increase in resources and clock speed
• Lower performance per TFLOP than Ampere

 

[W1zzard]

why is the 4090 trailing behind all old gen cards at 1080p & 1440p res in games like Battlefield 5, Borderlands 3, Diviniti Original Sin II, Far Cry 6, Halo Infinite or Hitman 3

As mentioned in the conclusion, it seems the driver overhead is higher than on Ampere, so this higher CPU usage will eat into the CPU time available for the game and thus make it even more CPU limited

@W1zzard Have you noticed any fan wobble on your sample? I think Der8auer and TechYesCity had their fans wobble and exhibit strange fan noise at certain levels

No, runs perfectly smooth

Waiting also 5K, 6K, 8K test.

No plans, unless someone sends me a monitor Even then, not sure if I can add another resolution.

And PCI-E 3.0 vs PCI-E 4.0?

Very soon

 

[evernessince]

This GPU is a beast. Very impressive.
Good job Nvidia!

Agreed. Regardless of other factors Nvidia did a good job with this GPU.

 

[chowow]

Maybe good thing you tested 5800x my 5600 might not bottleneck this at 4k on ultra settings and VR gaming.

 

[Richards]

https://twitter.com/i/web/status/1579825383842537473

The card needs zen 4 and raptor lake to show its true power.. 5800x is bad

 

[AnotherReader]

I understand the constraints that @W1zzard was operating under, but this would have been a good time to switch the GPU test bed. If not going with Zen 4 or Alder Lake, replacing the 5800X with a 5800X3D would have been appreciated. That being said, it's unlikely to be a bottleneck in most games at 4K, and for this card, the 4K results are the most relevant ones.

 

[W1zzard]

You'll get your 13900K with RTX 4090 results soon enough, just be a bit more patient

 

[RandallFlagg]

To be fair, Turing was statistically the worst Nvidia GPU generation ever. Zero improvement in performance per dollar, only a small performance bump over last gen, and a single digit efficiency increase.

That's not accurate. The 2060 was 45% faster than its predecessor 1060. The MSRP increase with the same VRAM was +$50. There was a huge jump in the midrange with Turing. The top end is where it was lackluster.

With Ampere, the mid range didn't get much. The 3060 was only about 19% faster than the 2060, and basically tied the 2060 Super. On paper, the 3060 had a $20 price reduction, but we never saw that thanks to Covid and crypto.

1060 vs 2060 on release :

3060 vs 2060 on release :

 

[AnotherReader]

https://twitter.com/i/web/status/1579825383842537473

The card needs zen 4 and raptor lake to show its true power.. 5800x is bad

TechSpot only tested 12 games so their average can't be compared to TPU's. Even in GPU limited scenarios like CyberPunk at 4K, the ratio between the 4090 and the 3090 is almost the same as the average. Still a faster CPU seems to be called for with this new generation of GPUs.

That's not accurate. The 2060 was 45% faster than its predecessor 1060. The MSRP increase with the same VRAM was +$50. There was a huge jump in the midrange with Turing. The top end is where it was lackluster.

With Ampere, the mid range didn't get much. The 3060 was only about 19% faster than the 2060, and basically tied the 2060 Super. On paper, the 3060 had a $20 price reduction, but we never saw that thanks to Covid and crypto.

1060 vs 2060 on release :
View attachment 265041

3060 vs 2060 on release :
View attachment 265042

Even the flagship was pretty fast for Turing; it's the prices that were terrible. At 4K, we saw:

 

[vmarv]

I understand the constraints that @W1zzard was operating under, but this would have been a good time to switch the GPU test bed. If not going with Zen 4 or Alder Lake, replacing the 5800X with a 5800X3D would have been appreciated. That being said, it's unlikely to be a bottleneck in most games at 4K, and for this card, the 4K results are the most relevant ones.

Hardware Unboxed did the review with the 5800X3D

 

[TheDeeGee]

While not the claimed 4 times faster, the sometimes near double jump is impressive from one generation to the next.

I have to quote myself.

It is actually 4 times faster in 4K with DLSS ^^

 

[Fluffmeister]

Hey @W1zzard, it's must be terrible for you having all those big chunky 4090's taking up all that room at your no doubt comfortably sized labs, but still... feel free to send me your least favourite 4090 if only to free up some room of course.

 

[evernessince]

That's not accurate. The 2060 was 45% faster than its predecessor 1060. The MSRP increase with the same VRAM was +$50. There was a huge jump in the midrange with Turing. The top end is where it was lackluster.

With Ampere, the mid range didn't get much. The 3060 was only about 19% faster than the 2060, and basically tied the 2060 Super. On paper, the 3060 had a $20 price reduction, but we never saw that thanks to Covid and crypto.

1060 vs 2060 on release :
View attachment 265041

3060 vs 2060 on release :
View attachment 265042

No, Turing was terrible across the stack.

The comparison you are making, the 1060 vs the 2060, is a bad one because the 1060 had an MSRP of $250 whereas the 2060 had an MSRP of $399. That's an increase of $150 USD. I'm not sure where you are getting your +$50 figure but it was a lot more than that: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_200_series

Mind you in this price category even a $50 price increase is huge. For a $200 video card that's a 25% price increase. The price increase of $150 more than offsets any performance increases seen.

So yes even in the best scenario you can come up with for Turning you are still looking at a average to less than average gain at a large increase in price. In fact the 2060 doesn't even target the same market segment the price increase is so large.

Ampere is just a continuation of the trend of a lack of improvement in entry level / mid range video cards. Mind you I'm not sure if I'd quality $400 as mid range, it's in that middle ground where the price could warrant a high end tag.

 

[RandallFlagg]

TechSpot only tested 12 games so their average can't be compared to TPU's. Even in GPU limited scenarios like CyberPunk at 4K, the ratio between the 4090 and the 3090 is almost the same as the average. Still a faster CPU seems to be called for with this new generation of GPUs.


Even the flagship was pretty fast for Turing; it's the prices that were terrible. At 4K, we saw:

View attachment 265043

Both the 4K and the 1080P gains with 1080 Ti vs 2080 Ti pale compared to 1060 vs 2060 gains. At 1080P the 2060 got +45% - but at 4K, it got +67%.

4K

No, Turing was terrible across the stack.

The comparison you are making, the 1060 vs the 2060, is a bad one because the 1060 had an MSRP of $250 whereas the 2060 had an MSRP of $399. That's an increase of $150 USD. I'm not sure where you are getting your +$50 figure but it was a lot more than that: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_200_series

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.

Mind you in this price category even a $50 price increase is huge. For a $200 video card that's a 25% price increase. The price increase of $150 more than offsets any performance increases seen.

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.

 

[AnotherReader]

Both the 4K and the 1080P gains with 1080 Ti vs 2080 Ti pale compared to 1060 vs 2060 gains. At 1080P the 2060 got +45% - but at 4K, it got +67%.

4K
View attachment 265045



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 265047




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.

That's right, but I think the true successor for the 1060 was the 1660 Ti. The 2060 was too expensive to be considered a 1060 successor.

 

[R0H1T]

I have to quote myself.

It is actually 4 times faster in 4K with DLSS ^^

And you need to buy that new card to get DLSS 3, win-win

 

[HTC]

The comparison you are making, the 1060 vs the 2060, is a bad one because the 1060 had an MSRP of $250 whereas the 2060 had an MSRP of $399

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.

 

[mb194dc]

Great performance if you want RT. Can already game at 4k 60fps+ on prior gen. $1600 seems totally ludicrous price to me.

 

[RandallFlagg]

That's right, but I think the true successor for the 1060 was the 1660 Ti. The 2060 was too expensive to be considered a 1060 successor.

I heard a lot of people say this at the time, many even said it was just as fast as the 2060 which was false (2060 was 18%-20% faster).

The problem with comparing at a static price point is that you fail to factor in inflation. Think what you'd be looking at if you compared cars that way.

To illustrate - in the past 24 months in the US there has been over 14% general price inflation. So a $300 widget from 2020 would today cost $342. We are talking about a 1060 from 2016 vs a 2060 from 2018 vs 3060 from 2020. The same tier is not going to stay the same price. $299 - $350 is IMO a perfectly valid range for these cards.

 

[adilazimdegilx]

Impressive performance, glad to see it's efficient and runs quite cool even as FE card.
Shocked to see the CPU struggling to keep with this card on 4K! This got me hyped even more for DLSS3.
I'm not really sold about this new PCIex power connector though. Other reviewers had some serious concerns.
Great review as always, looking forward to other brands' cards and scaling charts.
Also what's about RE Village chart? 4090 has phenomenal gains there compared to to others?

 

[BigMack70]

To be fair, Turing was statistically the worst Nvidia GPU generation ever. Zero improvement in performance per dollar, only a small performance bump over last gen, and a single digit efficiency increase.

I hope to see the same people that are praising this 45% increase also praise AMD when they achieve the same.



No point until the 13900K and 7950X3D or other Z4 3d variants release.

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

Again, I ask - why is this review's conclusion authored in such a way that would suggest historically significant performance improvements worthy of effusive praise, when it's doing what is merely ordinary?

It's really misleading, and suggests far too much of Jensen's marketing kool-aid has been consumed.

 

[mouacyk]

The efficiency is the most surprising part.

That should be credited to TSMC, not NVidia?