Skip to main content

Nvidia GeForce RTX 4070 Ti Super review: a new 4K/1440p contender

Extra VRAM is welcome, performance gains are variable.

Nvidia's Super refresh line of updated RTX 40-series graphics cards continues with the launch of the RTX 4070 Ti Super - an ungainly-named arrival that seeks to replace the non-Super counterpart, offering a range of upgrades: more memory bandwidth, more CUDA cores and 16GB of framebuffer memory - for the same money. On paper, the specs look good, but in reality, Nvidia needs to strike a balance here: we can't expect the card to be within striking distance of the existing RTX 4080, because the specs suggest that 4080 Super won't be much better. Push too hard on 4070 Ti Super and there'll be no market for the next Super.

The specs do look good, however. The RTX 4070 Ti Super is based on the same AD103 silicon as the RTX 4080, automatically giving it an advantage over the RTX 4070 Ti non-Super, which used the lower-performing AD104. The Ti Super has 8488 CUDA cores - 10 percent more than its predecessor and 87 percent of the 4080's complement. Meanwhile, the new card has a 100MHz boost clock advantage over the RTX 4080.

AD103 silicon also means that the 70-series class gets a 256-bit memory bus, up from the 192-bit interface in the non-Super predecessor and an impressive 33 percent increase to bandwidth. And there's more - not only does AD103's 256-bit interface enable 16GB of memory, it means users get the improved media block with dual video encoders. AD104 on the non-Super just got a single encoder. I actually use an RTX 4080 in my workstation for Adobe Premiere work and just a little gaming: the RTX 4070 Ti Super works just as well.

RTX 4070 Ti Super Max Turbo Hyper Fighting Edition: the Digital Foundry video review, presented for your viewing pleasure.Watch on YouTube
4080 Super 4080 4070 Ti Super 4070 Ti 4070 Super 4070
Processor AD103 AD103 AD103 AD104 AD104 AD104
CUDA Cores 10240 9728 8448 7680 7168 5888
Boost Clock 2.56GHz 2.51GHz 2.61GHz 2.61GHz 2.48GHz 2.48GHz
Mem Allocation 16GB 16GB 16GB 12GB 12GB 12GB
Mem Interface 256-bit 256-bit 256-bit 192-bit 192-bit 192-bit
Mem Bandwidth 736GB/s 717GB/s 672GB/s 504GB/s 504GB/s 504GB/s
TGP 320W 320W 285W 285W 220W 200W
US MSRP $999 $1199 $799 $799 $599 $599
UK RRP £969 £1199 £769 £799 £579 £589
Release date 31/1/24 Out 24/1/24 Out 17/1/24 Out

For review today, we don't have the Founders Edition card, because there isn't one. Similar to RTX 4070 Ti, Nvidia has handed this over to third-party board manufacturers to produce their own models. European press were gifted the MSI Ventus 3x model (which didn't run to reference spec at launch and requires a VBIOS update) and I also received the Asus TUF Gaming version. Both are large, relatively cheap-looking cards that lack anything like the aesthetic refinement of a Founders model.

However, their sheer size and bulk - enough to register on DRADIS - do at least ensure very quiet operation and good temperature management. The 12VHPWR socket is back on both models for power delivery, while it's the usual assortment of HDMI 2.1 and DisplayPort 1.4a video outputs on the rearof the MSI model, while the Asus (helpfully!) gains an additional HDMI port.

Testing in this review was performed using the Asus TUF Gaming version of the RTX 4070 Ti Super, as until that arrived, it was unclear whether the VBIOS update for the MSI model was working correctly. Both Asus and MSI models should be available at the $799/£769 MSRP and both should be operating to Nvidia's reference specs.

Power analysis throws up little in the way of surprises. We use Nvidia's PCAT interposer hardware sitting between the graphics card and its power sources (PCIe slot and auxiliary inputs). We measure average power draw along with frame-rate. When dividing power in watts by frames per second, you get joules per frame - and the lower that is, the better.

I've tested here using the Asus board - but I did run the MSI Ventus 3x numbers beforehand (pre-VBIOS update) and it seems that the TUF Gaming card is drawing more power - a little more than its 285W TGP spec limit. Even so, there's little to write home about at this point, as we all know that the Ada Lovelace cards are very efficient. The gains are most pronounced in ray tracing applications, where I found the 4070 Ti Super to have a distinct advantage over the non-Super and leaves the RX 7900 XT (AMD's closest competitor) way behind.

The gulf is wider in Forza Horizon 5, where Nvidia tends to have a performance (and therefore efficiency) advantage, with the Ti Super sipping just a little more power than the non-Ti version. Meanwhile, in Hitman 3 with RT disabled, AMD's performance advantage comes to the fore and so as you can see, in best case scenarios, AMD can actually be more efficient than Nvidia.

RX 7900 XT RTX 4070 Ti RTX 4070 Ti Super
Control, 4K, High RT 303.2W/31.0fps - 9.8 Joules Per Frame 277.3W/34.3fps - 8.1 Joules Per Frame 298.7W/38.7fps - 7.7 Joules Per Frame
Forza Horizon 5, 4K, Extreme, RT Off, 4x MSAA 309.1W/96.1fps - 3.2 Joules Per Frame 219.1W/101.4fps - 2.2 Joules Per Frame 245.8W/108.0fps - 2.3 Joules Per Frame
Hitman 3, 4K, Max, RT Off 312.1W/165.4fps - 1.9 Joules Per Frame 280.2W/132.3fps - 2.1 Joules Per Frame 298.5W/136.4fps - 2.2 Joules Per Frame

With the preliminaries out of the way, it's time to get into the benchmarks proper. As usual, we're using a high-end rig that pairs a Core i9 13900K with 6000MT/s G.Skill DDR5 and an Asus ROG Maximus Hero Z690 motherboard, minimising CPU bottlenecking to show GPU differences more clearly.

Nvidia GeForce RTX 4070 Ti Super Analysis