Nvidia’s bold 5070 claims could set buyers up for disappointment

Nvidia’s presentation at CES 2025 was a bumper one, as the tech giant unveiled the complete details about the upcoming release of the Blackwell architecture 50 series.

Among all of that juicy new info was plenty of impressive performance promises, and these new cards are undoubtedly a marked step up from their equivalents in the previous generation.

The revamped architecture is geared towards allowing AI to make significant improvements to the overall experience, prompting CEO Jensen Huang to suggest that Blackwell is “the most significant computer graphics innovation since we introduced programmable shading 25 years ago.”

This is all very exciting, but it was another assertion from Huang during his keynote presentation that drew all of the attention.

When talking about the upcoming release of the 5070 – the budget option when compared with the 5080/5090 – Huang suggested that those who buy the card would be getting “RTX 4090 performance at $549.”

A bold claim, but is it true? Well, after looking at the currently available data: yes and no.

What’s under the hood? 4090 vs 5070

NVIDIA

To start this conversation as objectively as possible, a side-by-side comparison of the two cards doesn’t reflect particularly well on the 5070. Though the Blackwell architecture is objectively better than the Ada Lovelace foundations of the 4090 (with support for Display Port 2.1), that’s largely where the outright hardware advantages end.

The 4090 has a monstrous 24GB of GDDR6X memory compared to the 5070’s 12GB of GDDR7. In terms of parallel processing, the 4090 also vastly outstrips its rival, packing 16,384 CUDA cores, while the 5070 has just 6,144. It also has double the bus width and almost double the bandwidth, and over three times as many TMUs and ROPs, leaving the 5070 with virtually no tangible advantages on paper.

So, what is this debate all about? Well, it comes down to a heady mix of AI, true performance vs perceived performance, and a hint of obfuscation on Nvidia’s part.

DLSS 4 is doing the heavy lifting

NVIDIA

Jensen Huang’s claims are based on the theoretical output offered by DLSS 4, a major step up from the already pioneering DLSS 3.5 that the previous generation of cards uses. DLSS 4 has a new Multi Frame Generation technique that generates 3 additional frames for every traditionally rendered one.

Obviously, the difference this makes to the user is significant, dialing them up from 30 FPS to 120 FPS with the technology active. The addition of neural rendering adds even more to that tapestry, allowing them to reasonably claim parity with the 4090.

Equally, Nvidia’s CEO also put the 5070’s performance at twice the level of its equivalent in the previous generation. With all being equal, playing a game with DLSS 4 on the 5070 will give roughly twice the performance of the same setup with a 4070 on DLSS 3.5. 

The major issue here is that many games don’t support DLSS 4 at this stage, rendering those performance gains moot. Without DLSS 4, you’re only looking at a performance increase of around 30%.

Whether you favor the budget offering of the 5070 or the titanic output of the 4090 likely depends on your perspective, but don’t be fooled into thinking these cards are one and the same.

On the face of it, the 5070 has the potential to be the best budget card of all time within the context of its immediate competitors, but it doesn’t have the hardware under the hood to compete with the 4090 in a straight shootout.

In any case, the 5090 is the new monster on the scene, so if you have a couple of thousand dollars to burn, then that’s the one to go with.