Nvidia GeForce RTX 4070 Ti Super review: More VRAM and bandwidth, slightly higher performance

Nvidia's RTX 4070 Ti Super, in the form of the Asus TUF Gaming reference clocked card, provides plenty of weight to arguments from Nvidia that modern graphics cards don't necessarily need 16GB of memory and a 256-bit interface to perform well. There are certainly cases where the extra memory helps, but they're relatively rare.

Across our standard 15-game test suite, the biggest performance advantage we saw compared to the 4070 Ti 12GB card was only 15%. Even 1% lows weren't ever more than 15% higher than the 4070 Ti Super's predecessor. But we did find at least one example in our bonus gaming tests where the added memory and bandwidth helped more: Alan Wake 2, with full path tracing, running at 4K using DLSS Quality upscaling, ran 23% faster, on both average and 1% low fps results.

That doesn't mean other games, particularly future games, might not want more than 12GB of VRAM. Plus, if you're spending $800 on a graphics card, it's really difficult to justify not having at least 16GB of memory these days. The vanilla 4070 Ti may perform well enough in general, but it always felt overpriced and underspec'd relative to other GPUs.

As we've said before, we would have preferred an additional 4GB of VRAM (and a 64-bit wider interface) on nearly every RTX 40-series GPU. The RTX 4070 Ti Super basically gives us exactly that by opting for the AD103 chip rather than the AD104. By using a relatively cut-down chip with only 83% of the cores enabled, there's still a decent gap between the 4070 Ti Super and the 4080 — and soon its 4080 Super replacement — which also gives Nvidia something to do with chips that couldn't meet the 76 SMs requirement of the 4080.

Nvidia's still making a sizeable profit on these cards, though. With a die size of 379mm^2, Nvidia should get around 150 potential chips per 300mm wafer from TSMC. Even at a price of $20,000 per wafer — which is almost certainly far more than what Nvidia is paying (we've seen estimates of more like $14~$15 thousand) — that would still only be $133 per chip in pure silicon costs. Call it another $200 or so for the PCB, cooling, and other components, and that's still less than half of the MSRP.

This isn't a generously priced graphics card, in other words, and by raising the MSRP, Nvidia probably took a more sizeable cut from its AIB partners. Still, it's certainly better than paying the same $800 for the RTX 4070 Ti — which is probably why the base price on those cards has fallen to around $740, and we've seen sales push the price as low as $720, which is still arguably too high.

Is the RTX 4070 Ti Super worth the asking price? That depends on what you want to do with it. Relative to AMD's RX 7900 XT, even at its current promotional pricing starting at $709, you can certainly make arguments in favor of Nvidia's GPU. It's more power efficient, is potentially better equipped for future games (if ray tracing adoption picks up), offers access to Nvidia's proprietary DLSS features, including frame generation, and you get superior AI performance.

If all you care about is rasterization performance, AMD's 7900 XT comes out ahead and offers a better value. And there are hundreds of new rasterization-only games released every year. But if you value any of those other 'extras' — even if you only think you might want to try them — Nvidia has cards at every price point that are worth a look.

Ultimately, the RTX 4070 Ti Super provides some worthwhile improvements over its non-Super predecessor. If you're in the market for a high-end Nvidia GPU and you haven't upgraded in a few years, it's a great card. Just don't be surprised when next-generation GPUs come out in a year or so that have even more new features, improved performance, and just maybe not a massive generational price increase. (We can dream about that last one, right?)

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content