When a manufacturer prepares to fundamentally rebuild its GPU architecture, it’s worth taking a closer look. This is exactly the case with AMD’s upcoming RDNA 5 generation (rumored to be “UDNA”), at least if the latest leaks and roadmap rumors are to be believed. After the tactical withdrawal from the direct high-end duel with Nvidia – think of the RX-9000 series, which was clearly positioned below the RTX-90 class – there are now many indications of a renewed offensive. But not a hasty head-through-the-wall action, but a strategically well-considered architectural evolution that should make AMD competitive again on several fronts.
The flagship of the upcoming GPU series is said to have 50 percent more compute units (CUs) than the previous Navi 48, so we are talking about a massive boost in terms of raw performance. At the same time, a significantly wider memory bus (384 to 512 bits) and an expansion of the VRAM to up to 32 GB are also on the cards. The whole thing is to be manufactured on the basis of the N3E process at TSMC, which suggests both efficiency gains and potentially higher clock rates. What is particularly noteworthy is that this could be AMD’s first attempt to establish a genuine chiplet design in the GPU segment, a technology that has already been adapted from the CPU sector with great success.
But that’s not all. RDNA 5 is also set to introduce new functional blocks: The so-called Radiance Cores, dedicated hardware for ray tracing and pathtracing, should finally close the gap that AMD has so far left open compared to Nvidia in terms of RT performance. There is also talk of neural arrays, i.e. logically networked CU groups that act as AI units and are specially designed for tasks such as upscaling or neural rendering. The third new addition is the Universal Compression Engine, a hardware algorithm for low-loss data compression that is designed to effectively reduce memory bandwidth. All in all, the overall picture is quite coherent, at least on paper.
A look at the presumed SKU configurations shows that AMD doesn’t just want to be in the upper echelons this time. While the top model is to be equipped with 96 CUs, a massive memory interface and up to 32 GB of VRAM, the mid-range is also being considered with 40 and 24 CUs as well as VRAM configurations between 8 and 24 GB. Even the entry-level segment is getting a refresh: 12 CUs with a 128-bit memory connection are planned here, sufficient for affordable gaming PCs if the pricing is right. RDNA 5 differs fundamentally from RDNA 3 in this breadth in particular and shows that AMD has obviously learned from old mistakes. Of course, the big question remains: How much does the fun cost? There are no official price details, but if you extrapolate the effort and the predicted specifications, the flagship is likely to land somewhere between 1,000 and 1,500 US dollars. By way of comparison, the RX 7900 XTX came onto the market with an RRP of 999 dollars. For Europe, with import sales tax and margin, significantly higher street prices are likely. This will be exciting, especially if Nvidia relaunches its 80 class in parallel. Should AMD still score points in terms of price-performance, RDNA 5 could become a real “comeback chip”.
The launch is currently planned for the second quarter of 2026. Until then, we are sure to see more leaks, “accidentally” published slides and the odd cryptic statement from AMD executives. Meanwhile, the codename circus is already underway: internally, there is talk of “Alpha Trion” (consumer GPUs), “Ultra Magnus” (Xbox SoC) and “Orion Pax” (PlayStation chip), a clear indication of the target groups. Let’s hope that the Transformers hype is not symbolic of an architecture that delivers more appearance than reality. The bottom line remains: RDNA 5 is a bet, with high stakes. Technologically ambitious, strategically well thought out, but dependent on many unknowns. Anyone familiar with AMD’s GPU path to date knows that not every big success has been a big success. But this time the signs are better: a broader line-up, useful features and, provided the drivers deliver what the hardware promises, competitive RT performance at last. Whether this is enough to stand up to Nvidia remains to be seen. But it won’t be boring.
Source: Wccftech
20 Antworten
Kommentar
Lade neue Kommentare
Mitglied
Mitglied
Mitglied
Urgestein
Urgestein
Veteran
Mitglied
Urgestein
Mitglied
Mitglied
Urgestein
Mitglied
Urgestein
Urgestein
Mitglied
Urgestein
Veteran
Mitglied
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →