Nvidia made a big splash with the official announcement of its upcoming GeForce RTX 50-series Blackwell GPUs during the CES 2025 keynote. And while the Halo RTX 5090 certainly looks like an absolute monster, for many people it’s the mainstream $549 RTX 5070 that will be the star of the show. The RTX 4070 has been one of the best graphics cards since its launch, and now its replacement is on the way.

Nvidia claims the 5070 will offer “RTX 4090”-level performance at about a third the price and just over half the performance. But how do they really compare and how does the 5070 compare to the existing RTX 4070? Let’s find out, and we’ve added estimates for a few bits and pieces for now, but most of the specs are correct.

First, let’s be clear: the idea that the RTX 5070 can keep up with the RTX 4090 in all workloads looks like very rose-colored glasses. It’s obvious that Nvidia is putting a lot of emphasis on AI with Blackwell, relying on DLSS 4 and other neural rendering techniques to make up the difference. However, for many existing games, pure specs still matter – unless there is a driver-side solution that enables higher performance without the need for patches and updates.

The RTX 5070 has 48 SMs compared to the 4070’s 46 SMs. That’s not a big change at all and a far cry from the 4090’s 128 SMs. The total calculation of the FP32 graphics is 31 TFLOPS on the 5070, 29 TFLOPS on the 4070 and 83 TFLOPS on the 4090. It is extremely heavy to believe that the 5070’s performance will generally come anywhere close to that of the 4090 without using DLSS 4 and related technologies.

VRAM is also something to consider. The 4090 has 24GB, compared to half that amount on the 4070 and 5070. There aren’t many games where 12GB isn’t enough, but Indiana Jones and the Great Circle definitely exceeds 12GB at 4K with full RT and no upscaling . There are likely to be more games coming that could allow for more than 12GB of VRAM usage at higher resolutions and settings.

But this is where “RTX Neural Material” could come into play. This appears to be enabling neural texture compression, which Nvidia discussed back in 2023 and has been fully implemented in a game. Will it work with? any Game? Probably not, but we would love to see a driver-side solution that makes such a feature a reality. Without NTC or RTX neural materials, the 12GB will definitely prevent the 5070 from being able to compete with a 4090.

The bandwidth must also be taken into account. The RTX 4090 features 21Gbps GDDR6X on a 384-bit interface, compared to the 5070’s 28-32Gbps GDDR7 on a 192-bit interface. We know that the 4090 has a bandwidth of 1008 GB/s. The 5070 should end up with a bandwidth between 672 GB/s and 768 GB/s. The same applies here: without NTC or neural materials it will not be able to keep up with higher resolutions.

AI workloads like LLMs also enjoy a lot of VRAM capacity. Quantization only gets you so far, and the neural compression of LLMs doesn’t matter (as far as we know). The RTX 4090 with 24GB VRAM can easily load larger LLMs than the 5070, which will only rival the 4070 in terms of AI model sizes.

The situation is different when we look at the computing power of AI. We know that the RTX 50 series will support the FP4 number format, but just as importantly, it appears to have twice the processing power per tensor core as the RTX 40 series. While that’s not enough processing power to outperform the 5070, the theoretical performance is “only” about 25% slower. And if something can use FP4 on the 5070 while the 4090 has to use FP8, then it might run better on the 5070. But even the INT8 TOPS prefer the 4090.

The real highlight is of course the price. There are many gamers who simply can’t afford a $1,599 graphics card – not to mention the shortage-induced $2,000-plus prices we’re currently seeing for the 4090. A $549 GPU, even if slower in most games, is a completely different ball game. Nvidia’s xx70-class GPUs have traditionally been the favorite for mainstream gamers, and the 5070 seems to continue that pattern. Even if it doesn’t outperform the 4090, it should end up being extremely successful if it can consistently deliver performance close to the RTX 4080’s levels.

DLSS 4 | New multi-frame generation and everything improved – YouTube

Keep looking