Real Stagnation: Six Years of GeForce RTX 60 Class GPUs Tested

[Steve]

What has Nvidia delivered over the past six years? We've tested the mainstream RTX 2060, 3060, 4060, and new 5060 across modern games to find out how far the 60-class GPU lineup has actually come.

Read the full article here.

Real Stagnation: Six Years of GeForce RTX 60 Class GPUs Tested

 

[Burty117]

Wild how much the VRAM is a limit on Horizon: Zero Dawn, yet the GPU itself can quite comfortably run the game at its highest setting if it had enough VRAM as per the 3060 12GB metrics.

 

[Puiu]

Tl;dr the 5060 is 56% faster than the 2060... 6 years of "innovation" to get such a paltry jump in performance.

 

[takaozo]

4070 and 5070 are the real x60 series, based on bus and other specs. But Nvidia decided to price 70 series as 80 series.
So a step down in specs and a step up in price.

This is how you make your company worth $4T and your stakeholders happy.

 

[Dr Roboto]

If it has not become clear, Nvidia no longer prioritizes graphics and video games. Nvidia is an AI hardware company now. I really do appreciate TS for doing the hard work for readers to compare these products and the thorough analysis. However, the fact of the matter is that the age of significant graphics improvement at ever lower prices is gone. There are no great new products that drastically alter that price to performance ratio for graphics cards. Silcon now focuses on AI improvements and not rasterization or VRAM. At this point, let's just be happy that Nvidia still makes GPUs for gaming at all.

From Tom's Hardware - Nvidia revenue by segment: Graph of revenue
"Nvidia’s data center business grew from $10.61 billion in December 2021, which accounted for 39.43% of its revenue, to a whopping $115.19 billion some three years later, making up 88.27% of sales. In comparison, its gaming GPU business shrunk to $11.35 billion and 8.7% from $12.46 billion and 46.31% of the revenue share in the same time frame."

--> Data Center (AI) business is 10X larger than Gaming now! <--

 

[WhiteLeaff]

Apart from the 1xxx, the performance gains of other generations were not revolutionary. The problem now is that the transistor budget is used for RT and ML hardware, not just rasterization performance, limiting improvements.
GPUs stagnating at 8GB is another joke.

 

[Neatfeatguy]

Comparing the recent xx60 models to make the 5060 look better than it really is, well, it's deceiving.

In past the 60 model card was usually as fast or faster than the top end card from 2 previous generations.
RTX 3060 12GB = GTX 1080Ti
RTX 2060 > 980Ti by roughly 25%
GTX 1060 > GTX 780Ti by roughly 15%
GTX 960 = GTX 680
GTX 760 > GTX 580 by roughly 15%
GTX 660 = GTX 480
GTX 560 > GTX 285 by roughly 15%

Ever since the Ampere generation this has fallen to the wayside pretty hard.
RTX 4060 < RTX 2080Ti by roughly 25%
RTX 5060 < RTX 3090Ti by roughly 75%
(this comparison applies to pretty much every card from the last two generations aside from the 4090 and 5090. Everything else has been minimal gains, stagnant or non-existent when it comes to performance gains.)

This is how I compare generational jumps with GPUs (and CPUs) and so far the last two generations from Nvidia and AMD have been sub-par when comparing older generational gains.

 

[EvilRebar]

This complaining about generation-over-generation improvements misses that silicon is now wildly more valuable for AI uses over gaming ones. If it was up to.AMD and Nvidia shareholders, gaming GPUs wouldn't exist.

You can farm anger clicks by pointing out that GPUs are not linearly improving on price to performance, but that is unlikely to change.

 

[BadThad]

Interesting article Steven, thanks!

 

[CapNemo72]

For me, 1060 was legendary.

Would be interesting to see, if you could underclock 5060Ti 16Gb, to bring it down to where 5060 performance is, when VRAM is no issue, and then show just how much more than 8Gb would help it show full performance it can give.

I would personally like to see low profile 5060 16Gb or well 5060Ti 16Gb, even if it is a bit underclocked

 

[ScottSoapbox]

Comparing the recent xx60 models to make the 5060 look better than it really is, is a bit deceiving.

In past the 60 model card was usually as fast or faster than the top end card from 2 previous generations.
RTX 3060 12GB = GTX 1080Ti
RTX 2060 > 980Ti by roughly 25%
GTX 1060 > GTX 780Ti by roughly 15%
GTX 960 = GTX 680
GTX 760 > GTX 580 by roughly 15%
GTX 660 = GTX 480
GTX 560 > GTX 285 by roughly 15%

Ever since the Ampere generation this has fallen to the wayside pretty hard.
RTX 4060 < RTX 2080Ti by roughly 25%
RTX 5060 < RTX 3090Ti by roughly 75%
(this comparison applies to pretty much every card from the last two generations aside from the 4090 and 5090. Everything else has been minimal gains, stagnant or non-existent when it comes to performance gains.)

This is how I compare generational jumps with GPUs (and CPUs) and so far the last two generations from Nvidia and AMD have been sub-par when comparing older generational gains.

Yeah, but some of the decline is legitimate slowing of die shrink gains.

If you also compare halo cards from each gen, you will see the gains are slowing there as well even with increases in power consumption.

Sure Nvidia is taking advantage of their dominate position to offer smaller generational increases, but physics is slowing what is possible - which is why AMD is blowing right past them with their gains.

 

[ScottSoapbox]

If it has not become clear, Nvidia no longer prioritizes graphics and video games. Nvidia is an AI hardware company now. I really do appreciate TS for doing the hard work for readers to compare these products and the thorough analysis. However, the fact of the matter is that the age of significant graphics improvement at ever lower prices is gone. There are no great new products that drastically alter that price to performance ratio for graphics cards. Silcon now focuses on AI improvements and not rasterization or VRAM. At this point, let's just be happy that Nvidia still makes GPUs for gaming at all.

From Tom's Hardware - Nvidia revenue by segment: Graph of revenue
"Nvidia’s data center business grew from $10.61 billion in December 2021, which accounted for 39.43% of its revenue, to a whopping $115.19 billion some three years later, making up 88.27% of sales. In comparison, its gaming GPU business shrunk to $11.35 billion and 8.7% from $12.46 billion and 46.31% of the revenue share in the same time frame."

--> Data Center (AI) business is 10X larger than Gaming now! <--

It's true gaming is only 10% of their revenue, but Nvidia likes money so there is no way they would stop making gaming GPUs.

But you are right about it not being a priority. Gaming GPUs will improve as AI cards are improved which will give general architecture improvements and die shrink improvements, but anything gaming specific likely won't be improved moving forward until something changes in the market.

 

[Puiu]

It's true gaming is only 10% of their revenue, but Nvidia likes money so there is no way they would stop making gaming GPUs.

But you are right about it not being a priority. Gaming GPUs will improve as AI cards are improved which will give general architecture improvements and die shrink improvements, but anything gaming specific likely won't be improved moving forward until something changes in the market.

It's actually around 6-7% now and dropping.

 

[ScottSoapbox]

It's actually around 6-7% now and dropping.

Oh, well, it was 10% last earnings statement I looked at. Still, even at 1% it's basically easy additional money for Nvidia as long as AMD is happy to mirror whatever they do.

And Jensen knows that AI hype can't last forever so strategically it makes sense to stay in the gaming market (especially it that only involves minimal effort).

 

[casteve]

It's true gaming is only 10% of their revenue, but Nvidia likes money so there is no way they would stop making gaming GPUs.

But you are right about it not being a priority.

I figure it's a wafer allocation issue, in part. If TSMC had enough capacity, Nvidia would spend more time on/ship more GPUs. Since there's a limited amount of capacity, Nvidia will prioritize AI chips as they make more margin $'s per wafer.

 

[Daniel Sims]

To put it another way, the gap between the 2060 and the 5060 is similar to the gap between the 1060 and the 2060.

In previous generations, the xx60 card was often as good as the previous generation's xx80 card, but with Blackwell we're only just now getting an xx60 card that's similar to the 2080.

xx70 cards are also starting to see the same effects. In prior generations, xx70 cards were often as good as the previous generation's flagship. The 970 was close to the original Titan, the 1070 was comparable to the Titan X, the 2070 Super resembles the 1080 Ti, the 3070 is like a 2080 Ti with less VRAM, and the 4070 Super is kinda like a 3090. That's probably the basis behind Jensen's "5070 w/MFG=4090" declaration. Instead the 5070 renders frames a lot like a 4070 Super, and there's no way the 5070 Super will make up the difference.

 

[ScottSoapbox]

I figure it's a wafer allocation issue, in part. If TSMC had enough capacity, Nvidia would spend more time on/ship more GPUs. Since there's a limited amount of capacity, Nvidia will prioritize AI chips as they make more margin $'s per wafer.

Agreed. I know TSMC has been bringing more capacity online, but I don't know how long until that helps us. It seems like AMD is selling their GPUs as fast as they can make them but only reaching 20-30% midrange market share depending on which numbers you believe. When capacity catches up is when Nvidia will have to adjust pricing (somewhat) to limit AMDs gains. I'm hoping that timing is more Super refresh this year rather than next gen.

 

[Burty117]

Do you know what would also be interesting? A very similar article but with AMD's offerings.

5070XT vs 6070XT vs 7700XT vs 9070XT

Just for fairness and I’m also intrigued if they’ve had similar uplifts or if AMD has been improving more consistently.

 

[Dr Roboto]

It's true gaming is only 10% of their revenue, but Nvidia likes money so there is no way they would stop making gaming GPUs.

But you are right about it not being a priority. Gaming GPUs will improve as AI cards are improved which will give general architecture improvements and die shrink improvements, but anything gaming specific likely won't be improved moving forward until something changes in the market.

I was not implying they would stop supporting graphics/games, just that it is no longer their primary focus for R&D or Profit. They have done a 180 and now AI is the business driver not graphics, which was exactly the opposite just 3 short years ago. Crazy how things change so quickly in the tech world.

 

[agentnate009]

Nvidia would never have approached 4 trillion valuation with gaming GPUs only. AI and deep enterprise client pockets are enabling that level of valuation and stakeholder happiness.

 

[ZackL04]

4070 and 5070 are the real x60 series, based on bus and other specs. But Nvidia decided to price 70 series as 80 series.
So a step down in specs and a step up in price.

This is how you make your company worth $4T and your stakeholders happy.

Pc graphics cards are not where they are making their money…