Nvidia is launching souped-up versions of its RTX 40 GPUs: the Super series. These super cards are faster than their predecessors and should support not only gamers but also AI workloads.

Nvidia announces the GeForce RTX 40 Super series at CES. In normal times we would position these cards as GPUs for better gaming performance, but today we don’t live in normal times. Everything has to have an AI focus, and that goes for these super cards too. According to Nvidia, these offer “the ultimate way to experience AI on PCs”. That’s a shame, because that’s what Intel also claims with its Core Ultra processors.

This AI focus is provided by the Tensor Cores on the cards along with Nvidia TensorRT software. The TensorRT LLM library for Windows is an open source library designed to accelerate inference with LLMs using Nvidia cards. For example, image generation with the new Super cards would be significantly faster compared to the RTX 3080 Ti. Nvidia’s focus is very interesting as Intel wants to draw similar conclusions with its CPUs. Jensen Huang clearly disagrees with this view and sees a prominent role for his GeForce cards.

Three GPUs

Nvidia is launching three new cards:

• The GeForce RTX 4080 Super has 836 AI TOPS (Tensor operations per second) and supports 4K gaming with ray tracing. The card is said to be 1.4 times faster than the RTX 3080 Ti. Nvidia announces a recommended retail price of $999. We don’t know a euro price yet.
• The GeForce RTX 4070 Ti Super is aimed at gamers with a 1440p screen, but can also handle 4K resolutions. You’ll pay $799 for this GPU.
• The GeForce RTX 4070 Super is a light version of the Ti, but still has 20 percent more cores than the RTX 4070. The card is faster than an older RTX 3090 and uses less. It will retail for $599.

When it comes to GPUs, Nvidia competes with the newly introduced Radeon RX 7600 XT. This GPU is cheaper than the RTX 4070 Super, but is also said to have a lot of AI capabilities on board. However, hardware is only part of the equation. Software optimization is more important and Nvidia has an advantage over both Intel and AMD thanks to its dominance. The TensorRT-LLM library can make a significant contribution to this.