Introduction

Say hello to the NVIDIA GeForce RTX 4070 SUPER Founders Edition graphics card. The RTX 4070 SUPER is the new gateway to high-end PC gaming, announced at the 2024 International CES earlier this month; and available starting tomorrow, January 17. It is part of a three-product GeForce RTX 40 SUPER series, which form a mid-life refresh for the GeForce RTX Ada family, although this round tidies things up for NVIDIA toward the upper half of its lineup. NVIDIA releases a new gaming graphics architecture roughly once every two years; and during the RTX 20-series Turing, it gave its lineup a similar refresh, debuting the SUPER brand. It skipped doing so with the RTX 30-series Ampere, because it really wasn’t hard for anyone to sell a high end GPU around that time, thanks to the crypto-mining boom.

The new GeForce RTX 4070 SUPER is recommended by NVIDIA for maxed out gaming at 1440p, including with ray tracing; although as we’ve seen in our RTX 4070 review that this category of GPU very much can game at 4K, including ray tracing; in some cases you might just need to tone down the eye-candy just a little bit, or engage DLSS. With the RTX 40-series, NVIDIA debuted DLSS 3 Frame Generation, which makes it only that much easier to nearly double frame-rates in supported games, making a stronger case for cards like the RTX 4070 SUPER that we’re reviewing today.

As with the RTX 20-series SUPER, the new brand extension SUPER in the RTX 40-series doesn’t denote any new features or technological changes; but is a case of NVIDIA increasing performance at given price-points. This is done by enabling more of the available shaders on the silicon, or even switching to a larger die altogether. NVIDIA debuted the RTX 4070 and RTX 4070 Ti roughly a year ago; both based on the 5 nm AD104 silicon. While the RTX 4070 Ti maxed out all 60 streaming multiprocessors available on the silicon, along with all its 80 ROPs, and 48 MB of L2 cache; the RTX 4070 is significantly cut down, enabling just 46 out of 60 SM (three quarters); just 64 ROPs, and just 36 MB of L2 cache. The RTX 4070 SUPER is a significant uplift over this, and you’ll see why.

NVIDIA carved the GeForce RTX 4070 SUPER out of the AD104 silicon by enabling 56 out of 60 SM, all 80 ROPs, and all 48 MB of L2 cache. This is almost a RTX 4070 Ti, spare for 4 SM. The 56 SM available work out to 7,168 CUDA cores, 224 Tensor cores, 56 RT cores, and 224 TMUs, an over 21% increase in SIMD resources over the RTX 4070, and not counting the performance impact from the extra 16 ROPs and a larger cache. The memory sub-system is the same as RTX 4070 Ti, you get 12 GB of 21 Gbps GDDR6X memory across a 192-bit memory interface. What happens to the RTX 4070 Ti now? Well, NVIDIA has retired it from its product stack, the remaining cards will sell out at slightly discounted prices, while the RTX 4070 Ti SUPER replaces it at its price, going on sale next week. RTX 4070 non-Ti remains as-is, and will sell for $550.

The GeForce RTX 4070 SUPER is based on NVIDIA’s latest Ada Lovelace graphics architecture, which, in addition to generational performance increase, introduces new features, and energy efficiency improvements, as it taps into the 5 nm foundry process. Ada debuts a new generation CUDA core which, besides increased IPC and support for new math formats, supports shader execution re-ordering, benefiting ray tracing workloads. The 3rd generation RT core, besides improvements to the ray intersection performance, adds support for displaced micro-meshes, which increases the complexity of ray traced objects. The Optical Flow Accelerator component assists in the generation of entire alternate frames entirely using AI, which is why DLSS 3 is exclusive to the RTX 40-series.

The new GeForce RTX 4070 SUPER Founders Edition is the de facto reference design by NVIDIA. It is differentiated from the regular (non-SUPER) RTX 4070 Founders Edition with a few aesthetic updates. The card now sports a predominantly black appearance, and looks like jewellery. The metal outer frame is now matte black, the X-shaped main frame has diamond cut edges that contrast with glossy black; while the backplate gets a stylized RTX 4070 SUPER logo. The card draws power from a 16-pin 12VHPWR power connector, which is needed, as NVIDIA has increased the total graphics power (TGP) of the RTX 4070 SUPER to 225 W, up from 200 W of the RTX 4070, to support those extra shaders. This also means that custom cards will lose the single 8-pin PCIe power connector, and switch over to the newer 16-pin connector. NVIDIA has set $600 as the baseline price for the RTX 4070 SUPER, which is also what the Founders Edition is being offered at. This is the same price the RTX 4070 was launched at, which now gets pushed down to $550—it’s not being retired from the product stack, unlike the RTX 4070 Ti.