If you’ve already devoured our coverage of AMD’s CPU-side announcements from CES, then you’ve seen the boatload of processors the Reds have cooked up. Side GPUthings are a little calmer, notably due to a total absence of a complete range, and performance figures still absent… However, RDNA4 and two graphics cards are announced, the AMD Radeon RX 9070 et AMD Radeon RX 9070 XT : enough to seriously raise the hype. Let’s see it all together!
RX 9070 and RX 9070 XT
Also subject to leaks, the RX 9070 and RX 9070 XT are officially announced, but we feel that the thing is sensitive. Indeed, no mention is made of performance, only of the RDNA4 microarchitecture, and even then, with a grain of salt. So don’t expect to be able to extract any data that will allow you to know in advance whether this generation is going to be worth it! We are therefore content with qualitative descriptions: RDNA 4 was designed with AI in mind (the opposite would have been surprising given the times!). The chip appears monolithic on the 3D renderings made available by AMD, and is based on 4 nm (N4P?) from TSMC. Internally, the CUs are reviewed to improve the IPC and the increase in frequency – we felt the WorkGroup Processors (WGP) bringing together two CUs in dual issue of RDNA 3 fragile on this point, we bet that this is simplified -, the AI finally benefits from dedicated matrix units worthy of the name (legacy of CDNA?), Ray Tracing is also upgraded to catch up on the delay on competition; It remains to be seen what happens in practice with regard to new generation of greens. Finally, media encoders and decoders are also better, but we don’t know why or how. Joy!
Finally, the most informative probably comes from the nomenclature slide, see:
If we can feel sorry for the alignment with the competitor (and get lost in conjecture for the following year: will we be entitled to the RX 10070?), the placement of the cards is not entirely innocent . The RX 9070 and 9070 XT are in fact placed between the RX 7800 There still remains the unknown of prices: you don’t need to be Irma to assume that the rumors will still be around in the weeks to come! Finally, note that AMD plans a release date in the first quarter of 2025, with designs available from 9 partners: Acer, ASUS, Sapphire, XFX, ASRock, GIGABYTE, PowerColor, Vastarmor and Yeston. After the loss of MSI last year, it’s up to Biostar to bow out, replaced by Acer: it’s up to you to draw your own conclusions!
The FSR 4, yes, but not for everyone!
Expected since the first days of NVIDIA’s DLSS, FSR 4 (FidelityFX Super Resolution 4), that is to say AMD technology allowing an image to be displayed in a definition higher than that calculated, finally integrates machine learning. Hooray! Enough to imagine a significant gain in quality, whether it is scaling or image generation, especially since Lisa Su’s firm is unveiling an AMD Anti-Lag at the same time. 2 supposed to be able to reduce latency even more effectively in the case where generation is active.
Purely by chancethis integration of machine learning goes hand in hand with the addition of matrix units and a restriction of functionality to RDNA 4 cards only, well, well! This is going to make more than one aficionado of the system that is usually open to all reds cringe… but which is nevertheless justified on a material level. AI requires a lot of computing, and typical graphics cores aren’t that efficient at the task. NVIDIA had accustomed us to the thing, and it is fair game that AMD does not put its own finances into optimizing its technology on Tensor Cores of the competitor! In addition, AMD communicates on an integration of FSR4 on titles already FSR 3.1, without however specifying whether an update of the game code is necessary: we should however have more information as the launch date progresses. is getting closer.
Finally, AMD could not stand still in the face of the competitor’s AI software ecosystem. This is howAdrenalin— the pilots, that is — is equipped with AI capabilities. Why not integrate all of this into a dedicated application? Certainly not to lighten the weight of the beast! In any case, the bousin will now be able to use your GPU to generate images, summarize texts or even serve as a chatbot by answering questions relating to your AMD hardware. For this last feature, we do not know if an internet connection will be required to benefit from it, but it is likely.
We cannot deny that this presentation leaves us somewhat unsatisfied as the details are both enticing and imprecise. For marketing, this is therefore a successful objective, but on a technical level, the information is poor, if not non-existent! See you in a few weeks for a real architectural dive?
