As for AMD, we now know a bit more about the port of FSR 4.1 to RDNA3 GPUs, the much-anticipated RX 7000 series. Broadly speaking, the company promises that there will be no loss in visual quality compared to its RX 9000 series, for one thing. On the other hand, the company had to convert its AI models to support INT8. As a reminder, the brand’s RX 7000 series is not compatible with this type of instruction.
FSR 4.1: AMD’s upscaling will run using INT8 instructions!
No loss in visual quality!
AMD’s new cards are fully compatible with FP8 instructions, which is not the case for the RX 7000 series. Therefore, to make theupscaling technology work on this older generation of cards, AMD must convert its AI models to run in INT8, which is entirely feasible. Additionally, the company promises there will be no loss in visual quality compared to FSR 4.1 running on the RX 9000 series, which is always a plus.
That said, the question remains whether the performance gains will be similar to those offered on the RX 9000 series. Indeed, early tests of the technology on the RX 7000 series seemed to indicate lower gains. That said, those were modifications created by modders, whereas this is work done by AMD on the software stack, so we’ll have to wait and see.
Things are looking hot for RDNA 2!
While RDNA 3 might be fine, for RDNA 2 (the RX 6000 series), porting FSR 4.1 looks like it’ll be a challenge. In fact, these cards don’t have a dedicated AI accelerator, so the calculations will have to be done on the GPU itself, directly via the Compute Units and Stream processors. Naturally, the impact on performance will be more significant, and the gains have the potential to be more limited. The company needs to further optimize its software and refine its AI models even more. In fact, it’s no coincidence that support for the technology on the RX 6000 series isn’t expected until early 2027.
