4070TI reviews thread

[BFG10K]

NVIDIA France Accidentally Reveals GeForce RTX 4070 Ti Specs

With less than a week to go until the official launch of the GeForce RTX 4070 Ti, NVIDIA France has gone and spoiled things by revealing the official specs of the upcoming GPU. The French division of NVIDIA appears to have posted the full product page, but it has since then been pulled. That...

www.techpowerup.com

First official performance leak. A perfect opportunity to highlight the fraud pushed by nVidia with DLSS.

Performance looks fantastic, right? Look again, at the smallprint - 4070 is using frame generation.

I said exactly this would happen as soon as DLSS 3.0 was announced. 2560x1440 is a lie given it's actually ~1080p due to legacy DLSS. Since nVidia got away with it, the next step is frame interpolation lies. It's "faster", yo!

Adjusting for frame generation (i.e. halve the 4070's bars), it's actually barely faster than the 3080 in MFS, and slower in Warhammer 40K. All this from a card that will likely cost more.

Customers not seeing the smallprint and/or not understanding DLSS are having systematic fraud perpetuated on them.


Reviews

MSI GeForce RTX 4070 Ti Gaming X Review

MSI's GeForce RTX 4070 Ti Gaming X comes with a large triple-fan, triple-slot thermal solution that runs at whisper-quiet noise levels with low temperatures. In terms of performance the new card is able to match last generation's flagship, the RTX 3090 Ti, and offers new features like DLSS 3...

www.techpowerup.com

 

[Mopetar]

Doesn't mention a definite set price and that's ultimately what'll determine the value. Frankly, I'm not optimistic on that point though.

 

[adamge]

Fake frames are fake.

 

[GodisanAtheist]

Ad104 is a 300mm2 die it's pretty small.

So I'm thinking $799 or $899 for the additional poke in the eye.

 

[Kaluan]

Ad104 is a 300mm2 die it's pretty small.

So I'm thinking $799 or $899 for the additional poke in the eye.

From a technical standpoint, I'm looking at it's ~300mm2, the 285W TGP/2,6GHz clock and it's projected 3080Ti-3090 performance and it doesn't add up. This chip is likely overtuned way outside it's efficiency v/c curve. Not unlike 3070Ti, which was a efficiency disaster, especially coming from it's brother and Ampere efficiency flagship, the 3070.

This SKU should've been a 3080/GA102-esque cut of the AD103, especially for that money. But nah, nVidia wants to milk the market until it breaks, again.

 

[jpiniero]

From a technical standpoint, I'm looking at it's ~300mm2, the 285W TGP/2,6GHz clock and it's projected 3080Ti-3090 performance and it doesn't add up.

The raw specs are virtually identical to the 3090 Ti.

 

[Timorous]

The raw specs are virtually identical to the 3090 Ti.

192 bit and 12GB is nothing like a 3090Ti in raw specs.

 

[Saylick]

I think it's supposed to fall in between a 3090 and the 3090 Ti. As for the power draw, Nvidia has actually been somewhat conservative in their TDP figures for the 4080/4090, so even with a 285W TDP I expect it to be closer to 250W.

 

[TESKATLIPOKA]

If someone doesn't like a relatively high TDP, which 285W is, then under volt and limit power to 200W. I don't think you will lose more than 10% of performance.

 

[jpiniero]

192 bit and 12GB is nothing like a 3090Ti in raw specs.

I am of course talking about fill rate and compute power. So the gap between the two in games should pretty much depend on how much the memory bandwidth hurts.

 

[moonbogg]

Gamers were happy to buy the 8gb 3070ti for $900, so the pricing here is simply following consumer behavior. I think this card should be $450, but that's just me I'm sure.

 

[Mopetar]

192 bit and 12GB is nothing like a 3090Ti in raw specs.

The larger cache offsets the memory bandwidth to some degree, similar to AMD having much smaller busses due to the inclusion of infinity cache. The 12 GB of memory is probably a bigger concern going forward, especially if it's going to be a long term card. Let's just be thankful it wasn't 10 GB or 8 GB.

 

[jpiniero]

NVIDIA GeForce RTX 4070 Ti 3DMark performance leaks out - VideoCardz.com

NVIDIA GeForce RTX 4070 Ti in 3DMark We have some 3DMark data to share featuring the soon-to-be-released RTX 4070 Ti graphics card. The RTX 4070 Ti will trade places with RTX 3090 Ti GPU in 3DMark tests. The data we collected shows that RTX 4070 Ti will be 20% to 30% slower than RTX 4080 […]

videocardz.com

Well, here you go. 3dmark scores are right around the 3090 Ti. Videocardz doesn't claim a source so I'm guessing it's a reviewer. It is true though that the 7900 XT is faster in the raster tests.

 

[SteveGrabowski]

Gamers were happy to buy the 8gb 3070ti for $900, so the pricing here is simply following consumer behavior. I think this card should be $450, but that's just me I'm sure.

I miss when the 70 card was $330, and this is a 70 card even if they want to call it a 70 Ti. Nvidia is dead to me with their greed. 12GB on a $900 card is gross.

 

[PJVol]

3dmark scores are right around the 3090 Ti.

Which puts it somewhere inbetween 3090 and 3090ti in gaming.

 

[igor_kavinski]

I hope game developers start increasing texture sizes more and more so Nvidia cards other than the 4090 just keel over and look obsolete in such game benchmarks. Nvidia did this with RTX so AMD should do it too. Convince game developers to put 7900 XT and 7900 XTX specific graphics settings presets that punish anything with less than 20GB RAM.

 

[TESKATLIPOKA]

I hope game developers start increasing texture sizes more and more so Nvidia cards other than the 4090 just keel over and look obsolete in such game benchmarks. Nvidia did this with RTX so AMD should do it too. Convince game developers to put 7900 XT and 7900 XTX specific graphics settings presets that punish anything with less than 20GB RAM.

Then you will kill both N32 and N33, not just Nvidia. Not to mention, how to convince them when you are the minor player in the gaming industry.

 

[igor_kavinski]

Then you will kill both N32 and N33, not just Nvidia.

It's just a preset in the games' graphics settings. Point being, if your card can't run that preset, you feel bad and upgrade to a better card or wait three to four years to afford the next mainstream card that will finally run that preset

That's what Nvidia's been doing with RTX and now they make all the 3090 Ti users feel bad about their dog slow RTX performance

 

[GodisanAtheist]

I hope game developers start increasing texture sizes more and more so Nvidia cards other than the 4090 just keel over and look obsolete in such game benchmarks. Nvidia did this with RTX so AMD should do it too. Convince game developers to put 7900 XT and 7900 XTX specific graphics settings presets that punish anything with less than 20GB RAM.

-16GB IMO.

N32 will be fine, as will N21 (throw prior gen top end a bone) but 4070ti and below get left out in the cold, as well as 3080ti and below.

 

[Saylick]

Bleh, so 3090 Ti performance at best for over $799, but likely over $899 since nothing launched this generation actually lists for MSRP. That's absolutely theft in broad daylight for essentially a sub-300mm2 die. Just pure dog crap, and that's insulting to the dog that crapped it.

As @SteveGrabowski mentioned earlier in this thread, it's a 70-class die through and through (forget the Ti modifier), which fits the bill because the 70-class card of the latest generation has historically offered flagship last-gen performance at half the cost.

For example:
- GTX 1070 @ $379 vs. GTX 980 Ti @ $650
- RTX 2070 Super @ $500 vs. GTX 1080 Ti @ $700 (Turing in general was inflated due to RTX tax)
- RTX 3070 @ $500 vs. RTX 2080 Ti @ $1200
- RTX 4070 Ti @ $900 vs. RTX 3090 Ti @ $2000 (BS MSRP to begin with)

The major issue this time is that the MSRP of flagship Ampere was just inflated out the wazoo and Nvidia doesn't get the memo that the crypto boom is over. If you assume a 3090 Ti should sell for no more than $1200, which is only possible because 4080s sell for much higher than it's $1200 retail price, using historical trends means the RTX 4070 should be a $600 card tops.

Honestly, I feel like building PCs as an enthusiast isn't even exciting anymore because PCs are already plenty fast if you are reasonable with your goals, but it doesn't help that GPUs are stupid expensive. I miss the days of looking forward to upgrading hardware because of the performance gains AND that an upgrade didn't cost you an arm and a leg. Many people did the whole "buy fastest GPU for same amount of money every 2 generations" and got to step-up their PC for a uniform, consistent level of financial investment on a yearly basis. Today? Forget about it. That strategy today requires you to pour more and more money because the perf/$ needle isn't even moving, as shown by the MSRP of each GPU class going up with each subsequent generation.

I was fortunate enough to snag a good condition used 3080 Ti for $650 from someone on this forum earlier this month, and it will likely be years before I buy another GPU. I see 4K as the highest resolution I'll ever play PC games at, and it can do 60 fps comfortably. What more do I need?

 

[Timorous]

As @SteveGrabowski mentioned earlier in this thread, it's a 70-class die through and through (forget the Ti modifier), which fits the bill because the 70-class card of the latest generation has historically offered flagship last-gen performance at half the cost.

680 was GK104 and 294mm with $500 MSRP
780 was a refresh so used GK110 at 594mm with $650 MSRP for about 24% faster than 680 performance.
980 was GM204 and 398mm with $550 MSRP for about 11% faster than 780Ti performance.
1080 was GP204 and 315mm with $600 MSRP for about 20% faster than 980Ti performance.
2080 was TU104 and 545mm with $700 MSRP for about 1080Ti equaling performance.
3080 was GA102 and 628mm with $700 MSRP for about 18% faster than 2080Ti performance.
4080 is AD103 and 379mm with $1,200 MSRP for about 16% faster than 3090Ti performance.

So the x80 using the 104 die is nothing new. The issue is the horrible pricing structure NV have gone with that totally breaks their historic pricing.

if they wanted to add a tier between the 102 and 104 dies it needed to be larger than 359mm² and offer better performance.

It would mean everything below 4090 goes up a tier which is fine if an AD104 equipped 4070 does what prior x80 parts have done such as beating the old top part by 10-20% at around $700.

 

[jpiniero]

So the x80 using the 104 die is nothing new. The issue is the horrible pricing structure NV have gone with that totally breaks their historic pricing.

That's what Moore's Law being dead does for ya.

 

[Timmah!]

680 was GK104 and 294mm with $500 MSRP
780 was a refresh so used GK110 at 594mm with $650 MSRP for about 24% faster than 680 performance.
980 was GM204 and 398mm with $550 MSRP for about 11% faster than 780Ti performance.
1080 was GP204 and 315mm with $600 MSRP for about 20% faster than 980Ti performance.
2080 was TU104 and 545mm with $700 MSRP for about 1080Ti equaling performance.
3080 was GA102 and 628mm with $700 MSRP for about 18% faster than 2080Ti performance.
4080 is AD103 and 379mm with $1,200 MSRP for about 16% faster than 3090Ti performance.

So the x80 using the 104 die is nothing new. The issue is the horrible pricing structure NV have gone with that totally breaks their historic pricing.

if they wanted to add a tier between the 102 and 104 dies it needed to be larger than 359mm² and offer better performance.

It would mean everything below 4090 goes up a tier which is fine if an AD104 equipped 4070 does what prior x80 parts have done such as beating the old top part by 10-20% at around $700.

103 is the new 104 size-wise, its actually even worse than that, since iirc historically 104 used to be 66 percent or 2/3 of the big 102 chip. Ada103 is not even 60 percent of 102 and 104 is now even less than that.

 

[Mopetar]

That's what Moore's Law being dead does for ya.

Moore's law isn't really dead, but if you tell people it is they might just believe you and not be as upset over the outrageous prices.

 

[igor_kavinski]

Yes, folks! Spend your hard earned money now. Tomorrow may never come and if it does come, you won't get the performance you are expecting coz Moore's law doesn't work anymore!