GeForce RTX 4060 Ti vs. RTX 3060 Ti: 40 Game Benchmark

[Steve]

We're taking a second look at the GeForce RTX 4060 Ti, but not because we think it's a graphics card that you should buy, but rather because we think it's completely lacking and you should avoid it.

Read the full article here.

https://www.techspot.com/review/2704-nvidia-rtx-4060-ti-vs-rtx-3060/

 

[oasked]

Is this the first time in graphics card history where there has been no notable generational uplift in performance?

 

[Kam7r]

Ppl should realise Nvidia is an Evil company maybe as bad as Apple, by my work I had contact with some of their ppl back in 2008 or something... maybe earlier... since then I'm boycotting them, they treated us as subhumans for months , if those ppl were in front of me and not by phone at the time, I would have punched them... anyway... last GPU I bought for me from NV was the Geforce 4 ti4200... I don't say that AMD is better... far from it... but at least a little "less worse" I think even intel is not at that lvl even with their corruption oops sorry "lobbying" of OEMs... maybe it's time to stop barking like little dogos like a lot of you do, never acting ... and really "vote with your wallet" or maybe you are into BDSM or something and love lower your pants in front of them...

 

[Theinsanegamer]

I stand by my original statement.

The 4060ti is a 4050ti with a name change. It should have been a $200 card. Tops. No 8GB card in 2023 is worth more then that.

Ppl should realise Nvidia is an Evil company maybe as bad as Apple, by my work I had contact with some of their ppl back in 2008 or something... maybe earlier... since then I'm boycotting them, they treated us as subhumans for months , if those ppl were in front of me and not by phone at the time, I would have punched them... anyway... last GPU I bought for me from NV was the Geforce 4 ti4200... I don't say that AMD is better... far from it... but at least a little "less worse" I think even intel is not at that lvl even with their corruption oops sorry "lobbying" of OEMs... maybe it's time to stop barking like little dogos like a lot of you do, never acting ... and really "vote with your wallet" or maybe you are into BDSM or something and love lower your pants in front of them...

Hey, hot pro tip. Corporations are not your friend.

Signed - O.B.Vious.

 

[Kam7r]

I stand by my original statement.

The 4060ti is a 4050ti with a name change. It should have been a $200 card. Tops. No 8GB card in 2023 is worth more then that.

Hey, hot pro tip. Corporations are not your friend.

Signed - O.B.Vious.

I know very well .. but there are really bad corporations and those a little "less worse" like I said... the best we can do is to not encourage and support the worse ones...

 

[Theinsanegamer]

Is this the first time in graphics card history where there has been no notable generational uplift in performance?

Nope. The radeon HD 6000 series was a complete standstill. The geforce 9000s as well.

 

[mdenigma]

3DCenter recently published a result of a poll about the reception of the card. It was all-time record in terms of negative feedback: Basically a 98% dont like this card. Where usually 80% is already really, really bad card and 60% is a bad or overpriced card.

https://www.3dcenter.org/news/umfra...steindruck-zu-nvidias-geforce-rtx-4060-ti-8gb

 

[Neatfeatguy]

Nope. The radeon HD 6000 series was a complete standstill. The geforce 9000s as well.

Yep, the Nvidia 8800 GTS 512MB = 9800 GTX/GTX+

The same GPU was even carried across to the GTS 250 that came out after the 9xxx series.G92 was an awesome chip when it first showed it's face on the 8800 GT card, shortly followed by the 8800 GTS 512.

 

[Kam7r]

Yep, the Nvidia 8800 GTS 512MB = 9800 GTX/GTX+

The same GPU was even carried across to the GTX 250 that came out after the 9xxx series.G92 was an awesome chip when it first showed it's face on the 8800 GT card, shortly followed by the 8800 GTS 512.

it was kinda the same with the hd7970 / 7970ghz and the R9 280 /280x

 

[neeyik]

One might be tempted to point all of the blame for the unimpressive performance of the 4060 Ti squarely on the 128-bit memory bus, but there's more to it than that.

After all, the 4070 Ti had a bus half as wide as the 3090's and 46% less bandwidth, but in Steve's testing, it was 13% faster overall at 1440p and 6% faster at 4K (though slower in quite a few games). And on paper, it's only 10% to 13% (in terms of pixel fill rate, texture fill rate, and FP32 throughput) than the 3090, though the 3090 is going to be a little CPU-limited at 1440p.

The 4060 Ti, though, has a 36% uplift over the 3060 Ti, with regards to FP32 rate and texture fill rate, thanks to its 52% higher clock rate. It should be a lot better than this. However, the older card has a few tricks up its sleeve compared to the Ada Lovelace one. The first being the number of GPCs -- the 3060 Ti has 5 whereas the 4060 Ti has 3, and while the latter is clocked higher, the difference in count means the newer card rasterization rate is only 32% higher.

The AD106 also has six TPCs (each of which comprises a pair of SMs) per GPC, while the GA104 in the 3060 Ti has four. That means the ROPs in each GPC have to contend with fewer read/write requests than those in the 4060 Ti. Again, they're clocked 52% higher, but each set of ROP partitions has to deal with 50% more requests.

The much larger L2 cache helps to mitigate this problem, but as soon as there's a cache miss, then the memory controller count and bandwidth issue kicks in. This is true for the 4070 Ti, which is why the 3090 is better in a fair number of games at 4K, but as the number of TPCs in each GPC is the same in both cards, the older model is using its bandwidth far more often.

Had Nvidia used a heavily cut-down AD104 for the 4060 Ti and retained the memory sub-structure (I.e. six memory controllers), then it would have been a much better prospect against the 3060 Ti.

 

[takaozo]

Now this is what I call progress.
I say the same gen leap as the 8800GT to 9800GT.

 

[DSirius]

Thank you for this review.
Will Nvidia 4060ti win the worst product of the year prize? For sure, but let's not underestimate Nvidia's greed. They have time till the end of the year to come up with more worst products.

 

[Dr Roboto]

Out of random coincidence, I just posted these marked up charts at Tom's Hardware yesterday. I was curious about the gen over gen performance and they have these really nice charts. My markup is not perfect and you need to ignore the AMD cards.

Maybe some of you will find it helpful, but it really shows how bad Nvidia screwed mid and low tier GPU consumers.

The 4060 is really bad. At 1440p it cannot come close to the older 3070, not even the 3060 Ti. When you look at chart 1 (previous series one model up), you can see that most new gen cards equal the performance of the previous gen but only level up (new 60 = old 70). However, the 4060 is actually 25% slower than the 3070 (red line)!
The 4060 Ti fairs a bit better but still poor. It is barely faster than a 3060 Ti and it is noticeably slower than the 3070 Ti.
The 4070 is about the same as a 3080 (disappointing but acceptable).
The 4070 Ti is a fair bit faster than the 3080 Ti (not bad).
The 4080 is 15% faster than the 3090. Now that is the kind of generational upgrade we like to see, but not the 25% we wanted.

 

[grvalderrama]

I've already said it before, this generation of gpus - so far- should be avoid except for the 4090, which is an overall a fair product, not even a good one, but if one wants the best of the best, then you have no choice other than the 4090 at its premium price.
I won't be upgrading my 6700xt anytime soon, I guess, at least amd brings a superior product at a decent price. I hold No expectations in that regard.

 

[KIAg1978]

Nothing new here !!

 

[Theinsanegamer]

One might be tempted to point all of the blame for the unimpressive performance of the 4060 Ti squarely on the 128-bit memory bus, but there's more to it than that.

After all, the 4070 Ti had a bus half as wide as the 3090's and 46% less bandwidth, but in Steve's testing, it was 13% faster overall at 1440p and 6% faster at 4K (though slower in quite a few games). And on paper, it's only 10% to 13% (in terms of pixel fill rate, texture fill rate, and FP32 throughput) than the 3090, though the 3090 is going to be a little CPU-limited at 1440p.

The 4060 Ti, though, has a 36% uplift over the 3060 Ti, with regards to FP32 rate and texture fill rate, thanks to its 52% higher clock rate. It should be a lot better than this. However, the older card has a few tricks up its sleeve compared to the Ada Lovelace one. The first being the number of GPCs -- the 3060 Ti has 5 whereas the 4060 Ti has 3, and while the latter is clocked higher, the difference in count means the newer card rasterization rate is only 32% higher.

The AD106 also has six TPCs (each of which comprises a pair of SMs) per GPC, while the GA104 in the 3060 Ti has four. That means the ROPs in each GPC have to contend with fewer read/write requests than those in the 4060 Ti. Again, they're clocked 52% higher, but each set of ROP partitions has to deal with 50% more requests.

The much larger L2 cache helps to mitigate this problem, but as soon as there's a cache miss, then the memory controller count and bandwidth issue kicks in. This is true for the 4070 Ti, which is why the 3090 is better in a fair number of games at 4K, but as the number of TPCs in each GPC is the same in both cards, the older model is using its bandwidth far more often.

Had Nvidia used a heavily cut-down AD104 for the 4060 Ti and retained the memory sub-structure (I.e. six memory controllers), then it would have been a much better prospect against the 3060 Ti.

So it's kinda the same as always, the cards are one level too high in the product stack. As a 4050/ti, the AD106 is pretty impressive. As a 3060/ti, it's lackluster at best. The 4070 is really a 4060, and so on. Only the 4090 at the top is any sort of improvement.

 

[neeyik]

So it's kinda the same as always, the cards are one level too high in the product stack. As a 4050/ti, the AD106 is pretty impressive. As a 3060/ti, it's lackluster at best

Very much so. Nvidia hasn't really used such a low-level chip for a mainstream model before, and it really shows. At some point in the not-too-distant future, though, there are going to be significantly diminishing gains in the budget and mid-range sectors, as we go from generation to generation -- no matter what chips they're using or what SKU names they have.

While there is still room to continue improving component density, alongside clock and power scaling, the associated cost for developing such processors (as well as the increase in wafer prices) is eventually going to result in performance increases at the bottom end of the product lineup tapering off. Unless the market is willing to pay higher prices for larger chips, of course (which it's not at the moment).

 

[Julnor]

I suspect the 40 series is a limited run designed to make the 30 series look more appealing. Nvidia realized they just weren't going to be able to move 30 series stock even with nominal price discounts, and they didn't want to do massive price cuts. Therefore they made a terrible limited run series to trick people into thinking the 30s were a good deal and GPUs were no longer going to have major gains. This also has the added bonus of upselling people to the 4090 and thinking they can just sit on that for several years, like a console.

However once they achieve their goal, they'll accelerate the 50 series release which will actually have major performance uplift and they won't have to do such a hard sell for it, it will sell itself. Those will have normal amounts. Additionally those GPUs will probably get great reviews too.

 

[yRaz]

Yep, the Nvidia 8800 GTS 512MB = 9800 GTX/GTX+

The same GPU was even carried across to the GTS 250 that came out after the 9xxx series.G92 was an awesome chip when it first showed it's face on the 8800 GT card, shortly followed by the 8800 GTS 512.

Do you know why they released those cards un the 8800 name instead of the 9000 name? It was because they had a bug in the DX10 instruction set and they dumped them for cheap. I know, I had 2 of them in SLi

 

[mcnabney]

Out of random coincidence, I just posted these marked up charts at Tom's Hardware yesterday. I was curious about the gen over gen performance and they have these really nice charts. My markup is not perfect and you need to ignore the AMD cards.

Maybe some of you will find it helpful, but it really shows how bad Nvidia screwed mid and low tier GPU consumers.

The 4060 is really bad. At 1440p it cannot come close to the older 3070, not even the 3060 Ti. When you look at chart 1 (previous series one model up), you can see that most new gen cards equal the performance of the previous gen but only level up (new 60 = old 70). However, the 4060 is actually 25% slower than the 3070 (red line)!
The 4060 Ti fairs a bit better but still poor. It is barely faster than a 3060 Ti and it is noticeably slower than the 3070 Ti.
The 4070 is about the same as a 3080 (disappointing but acceptable).
The 4070 Ti is a fair bit faster than the 3080 Ti (not bad).
The 4080 is 15% faster than the 3090. Now that is the kind of generational upgrade we like to see, but not the 25% we wanted.

Tom's images distort the horizontal with the inclusion of AMD.

I can't paste my own charts, but the 80 tier GPUs from Nvidia are a very steady and linear progression from 1080 to 2080 to 3080 to 4080. A slightly less steep slope is also found for the 70 and 70ti tiers. It almost flattens for the 60ti tier going from 2060super to 3060ti to 4060ti. The 60 tiers are almost as bad as the 60ti.

The problem with the higher tiers is just pricing. The new generations are showing strong gains. The 60ti and below tiers are simply lesser cards (and pricing) than what they should be. The end result is the significant performance delta between 4060ti and 4070.

 

[Theinsanegamer]

Very much so. Nvidia hasn't really used such a low-level chip for a mainstream model before, and it really shows. At some point in the not-too-distant future, though, there are going to be significantly diminishing gains in the budget and mid-range sectors, as we go from generation to generation -- no matter what chips they're using or what SKU names they have.

While there is still room to continue improving component density, alongside clock and power scaling, the associated cost for developing such processors (as well as the increase in wafer prices) is eventually going to result in performance increases at the bottom end of the product lineup tapering off. Unless the market is willing to pay higher prices for larger chips, of course (which it's not at the moment).

True, a $200 mainstream chip is likely not going to happen anytime soon, but nvidia is taking the piss on the 4060ti. I'd bet good money that at #350, they'd be able to sell the 16GB version and still make a healthy product, and even $300 is likely profitable (but not with the margins nvidia wants).

$400 for 8GB is pure price scalping.

 

[axiomatic13]

Sadly if these don't sell they will be sold at a loss to HP, Dell, etc. to go into prebuilt systems aimed at parents who don't know what a real gaming PC is.

 

[maxxcool7421]

Just to add add a hint of conspiracy .. people who bought 3xxx cards likely bought them during the effed up scalping phases of 2019-2021 and have -0- motivation to buy again after spending 600$ on a 3060/3060ti .. So NV is putting out a crap card on purpose because they know only people who 100% NEED an upgrade from a 2xxx series card or integrated will bite in this low tier segment.. so why waste fab resources on a card they know will be low volume.

Big spenders / gamers / streamers are only going to buy 4070ti/4090/4090ti's anyways ..

 

[WhiteLeaff]

"Resident Evil 4, using the max preset, provides us with another example where the 4060 Ti lags behind the part that came 2.5 years before it for the same price – a significant misstep indeed. It was 10% slower at 1080p and 9% slower at 1440p."

?? The data in the charts show something else.

 

[Neatfeatguy]

"Resident Evil 4, using the max preset, provides us with another example where the 4060 Ti lags behind the part that came 2.5 years before it for the same price – a significant misstep indeed. It was 10% slower at 1080p and 9% slower at 1440p."

?? The data in the charts show something else.

I saw that some time ago, but was too busy at work to make a post about it. Looks like the numbers in the graph shows the 4060Ti easily beating the 3060Ti, but the caption under the graph says the 3060Ti beats the 4060Ti (as you quoted in your post).