The GeForce RTX 4070 Ti is Here
In today’s video card review, we take a look at the brand-new ASUS TUF Gaming GeForce RTX 4070 Ti 12GB GDDR6X OC Edition video card, based on NVIDIA’s new GeForce RTX 4070 Ti. The GeForce RTX 40 Series from NVIDIA was announced on September 20th, 2022 but the first video card to launch in this series was the GeForce RTX 4090 on October 11th, 2022. The GeForce RTX 4090 was followed by the launch of the GeForce RTX 4080 on November 15, 2022. Both video cards are very extreme, high-end enthusiast GPUs that sit above $1,200 MSRP.
Today, the next step down from the GeForce RTX 4080 is launching, the GeForce RTX 4070 Ti, which is now under a thousand dollars, coming in with an official MSRP of $799.99. However, this comes with a caveat, this round there are no Founders Edition video cards from NVIDIA, instead, this will be a partner-only (add-in-board partner) AIB launch with only AIB cards available. This means price-wise; it will have a varying degree of pricing with custom SKUs from each AIB. Hopefully, we will see some ‘reference spec’ $799.99 models available, which is something to look for.
The specific video card we have for review today is the ASUS TUF Gaming GeForce RTX 4070 Ti OC video card (TUF-RTX4070Ti-O12G-GAMING), which has an MSRP of $849.99. This video card is a custom build from ASUS, and sits below the STRIX version, but offers a factory overclock and customized cooling that is built for overclocking. Before we look at the ASUS TUF GeForce RTX 4070 Ti Gaming OC, let’s see what the specs are on the new GeForce RTX 4070 Ti.
NVIDIA GeForce RTX 4070 Ti
Let’s get the elephant, nay Blue Whale, out of the room and acknowledge that the GeForce RTX 4070 Ti is the GeForce RTX 4080 12GB re-named. NVIDIA un-launched the GeForce RTX 4080 12GB, renamed it GeForce RTX 4070 Ti, and gave it a new sticker price.
So then, what is the GeForce RTX 4070 Ti? The GeForce RTX 4070 Ti is based on NVIDIA’s AD104 die using the Ada Lovelace architecture found in the new GeForce RTX 40 Series graphics cards. The AD104 die sits just below the AD103 die, which the GeForce RTX 4080 is based on, while the GeForce RTX 4090 is based on the AD102 die. Therefore, the move to calling the AD104 an “xx70” series, was the right one, as this is a different die altogether. This iteration of the AD104 die constitutes the full spec of that die with 60 SMs. You can read all about the Ada Lovelace architecture in our GeForce RTX 4090 review.
The GeForce RTX 4070 Ti is built on TSMC 4N NVIDIA Custom Process and has 35.8 billion transistors. It has 5 Graphics Processing Clusters, 60 Streaming Multiprocessors (SMs), 7,680 CUDA Cores, 240 4th Gen Tensor Cores, 60 3rd Gen RT Cores, 240 Texture Units, and 80 ROPs with 49,152 KB of L2 Cache. It has a boost clock set at 2610MHz. It has 12GB of GDDR6X memory running at 21GHz on a 192-bit memory bus providing 504GB/s of memory bandwidth. The Total Graphics Power (TGP) is 285W. Minimum Power Supply 700W. It will run on either 2x PCIe 8-pin cables or 300W or greater PCIe Gen 5 cable.
As with all other Ada Lovelace GPUs, it supports AV1 decoding and encoding and NVIDIA Broadcast. It also supports DLSS 3 Frame Generation and NVIDIA Reflex. The list of DLSS 3-supported games is growing.
| GeForce RTX 4090 | GeForce RTX 4080 | GeForce RTX 4070 Ti | |
|---|---|---|---|
| GPU Codename | AD102 | AD103 | AD104 |
| Architecture | Ada Lovelace | Ada Lovelace | Ada Lovelace |
| Process | TSMC 4N NVIDIA | TSMC 4N NVIDIA | TSMC 4N NVIDIA |
| Transistors | 76.3 Billion | 45.9 Billion | 35.8 Billion |
| L2 Cache Size | 73728 KB | 65536 KB | 49152 KB |
| GPCs | 11 | 7 | 5 |
| TPCs | 64 | 38 | 30 |
| SMs | 128 | 76 | 60 |
| CUDA Cores | 16384 | 9728 | 7680 |
| RT Cores | 128 (3rd Gen) | 76 (3rd Gen) | 60 (3rd Gen) |
| Tensor Cores | 512 (4th Gen) | 304 (4th Gen) | 240 (4th Gen) |
| ROPs | 176 | 112 | 80 |
| Texture Units | 512 | 304 | 240 |
| GPU Boost | 2520 MHz | 2505 MHz | 2610 MHz |
| VRAM | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X |
| Memory Interface | 384-bit | 256-bit | 192-bit |
| Memory Clock | 21 GHz | 22.4 GHz | 21 GHz |
| Memory Bandwidth | 1008 GB/s | 716.8 GB/s | 504 GB/s |
| TGP | 450W | 320W | 285W |
In terms of power, the GeForce RTX 4070 Ti should use even less power than a GeForce RTX 3080, yet perform at an RTX 3090 Ti or above.
Discussion (19 replies)
Join Discussion →Aaaand, it's a dud. The 7900 XT now actually makes sense.
Sort of want, but will wait until under 500. 1080ti still gets job done.
Ok, so this is only tangential to the topic... but at what point do you get to call it either a monopoly or price fixing?
Now, I do not think that AMD and nVidia are colluding to raise consumer prices. I think nVidia is doing it because they can get away with it, and AMD is just playing along with "just slightly less than" pricing because their shareholders would oust them outright if they didn't.
There is no government limit on the margin a company can ask, or the profit they can pull in. That's kinda how capitalism works. But I sure hate this feeling that we are just getting utterly taken advantage of. I guess it's just a hobby, I don't have to play along - I just hate seeing it happen before my eyes.
Nvidia has enormous pricing power because of RTX & DLSS3
At the premium price points ($500 plus) people most likely go by added features rather than outright raster performance
What is surprising is that: even in below $500 prev gen GPUs, higher price Ampere cards outsell equivalent but much cheaper RDNA 2 cards
If more people purchase Navi 22 cards (RX 6750 XT, 6700 XT, RX 6700) instead of Ampere, than Nvidia might get the message !!
DLSS3 itself isn't a real selling point. It exists, and it has a purpose - but not for most gaming scenarios. DLSS2 is still the main DLSS technology and remains more competitive overall, especially in situations where RT is used with heavy loads overall.
If nvidia pushes DLSS3 as hard as they pushed DLSS, we'll see support in most upcoming games.
Besides, 100% performance increase does sound like a big selling point.
Thanks, @Brent_Justice for the very thorough review which answered nearly every question I could have about this card.
Well, it seems the card lies between the RTX3090 and 3090Ti, but men the thing gets bandwidh starved at 4k
I'm really impressed just becauese looking at the specs I really didn't think it would do that good.
Great review!
I figured Nvidias "faster than 3090ti" claims were going to be all fluff, but it does at least get close most of the time, and actually succeeds in others.
Still, not quite sure I'd be happy with the value proposition $799 is a lot for a mid range card.
If I could make a request, if it isn't too much of a main in the ass, Minimum/1%/0.1% framerates are very relevant and would add a lot of information to your reviews. I'd even argue that minimum frames are MORE relevant that average frames.
Sure?
DLSS 1.0 was pretty bad - but showed the promise of intelligent upscaling.
DLSS 2.0 has more or less delivered on that promise. My opinion, of course; I find DLSS 2.0 to be more than close enough to justify any perceived degradation in visuals. Granted I have no problem using FSR / FSR 2.0 where exclusively available, whether I'm gaming on an AMD GPU or that's just what the game in question supports.
DLSS 3.0 is 'frame doubling'. This works great for pre-recorded media (if you like the effect itself), but for real-time applications it means that there's a definite insertion of additional input latency attached with usage. So, for DLSS 3.0 to even be useful in a particular situation, performance already has to be high enough for real-time interaction. Whereas DLSS 2.0 taking you from an arbitrary 20FPS to 35FPS, or 60FPS to 90FPS and so on can significantly impact the quality of the experience, DLSS 3.0 (that is, the 'frame generation' part of it) needs you to be running at 60FPS+ in the first place.
And there's a place for that - many games that aren't as user-response dependent can benefit, thinking anything turn-based or perhaps RPG and adventure-style games. Some simulators as well.
But for anything where optimizing for input latency is already a concern, DLSS 3.0 is a non-starter.
No one should be! But it's still the cheapest entry point for the level of technology Nvidia is bringing to the table.
BTW seems to me the cards are unnecessarily big. It seems the PCB is about 2/3 the size of the card.
I wonder if they reused parts from its bigger brothers to save R&D and inventory costs.
It's a tradeoff. It costs more per unit, but it saves having to develop new parts.
Yeah, I'm sure. There are already about 50+ games announced that will support DLSS3. If someone know how to shove new technology down your throat, that's nvidia. BTW thanks to nvidia reflex, latency is actually lower than without DLLS3. Not saying its going to be great, but by now DLSS already gained high praise and is considered a must have for many gamers, DLSS3 at least deserves a chance.
Anyway AMD is also having its own frame interpolation/frame creation technology coming soon. We'll see.
I haven't owned a Nvidia GPU newer than my 2016 Pascal Titan X, so I ahve not personally played with DLSS yet.
Can all titles that support DLSS3 also run DLSS2? In other words is it backwards compatible or an option?
Because I too tend to think - from what I have read and seen, through not through personal experience) that DLSS2 was the pinnacle of the tech.
Pretty sure the answer is yes. Brent has a write-up on it in his 4090 review that also goes a little more in-depth about the frame generation feature. It's about midway down the first page.
https://www.thefpsreview.com/2022/10/11/nvidia-geforce-rtx-4090-founders-edition-review/