Rumors say AMD’s next-gen graphics cards are slightly different in specs than previously expected due to the new leak – and while there’s a bit of a disappointment about some of the claims made in this new leak, there’s really nothing to worry about. Indeed, Nvidia will likely be worried about some of the information that has appeared here …
Before we get into why this is happening, let’s take a look at the leak itself that comes from Angstronomics (via VideoCardz (opens in a new tab)), a publication about the semiconductor industry we’ve heard about before. (Even if it’s not one of the more regular sources that reveal information on Twitter, we find Angstronomics reliable enough – plus a major YouTube leak more or less confirms these seriously detailed details, and we’ll come back to that.)
Angstronomics Breaks The Line On RDNA 3! We detail the key specifications of Navi31, Navi32 and Navi33 which were finalized in 2020 and have not changed since then! We also highlight some architectural changes including OREO, WGP, cache and die sizes! https: // t.co/n2qB7KiiBgAugust 12, 2022
The key points of this speculation include that AMD focuses on “area, area, area,” meaning smaller chips that are still strong enough to meet Team Red’s performance goals for RDNA 3, and also focuses on performance per watt ( performance). Of course, the company has already boasted of achieving a 50% increase over RDNA 2 using the latter (the same RDNA 2 leap that made the original RDNA architecture).
Add your own spice to it, but Angstronomics tells us the Navi 31 – the flagship of the RX 7000 graphics cards – will, rumor has it, be one Graphics Chiplet Die (GCD) and six Memory Chiplet Dies (MCD) modules, with 12 288 cores (known as ALUs). This main GCD will apparently be 308mm² in size, and the other big change from previous speculation is the size of the Infinity Cache, which is apparently 96MB for the top dog model (the trimmed Navi 31, possibly the version without the XT, will use 80MB).
The rumor mill mentioned much larger sizes like 192MB; and indeed the existing Navi 21 uses 128MB Infinity Cache, so would actually be smaller. Angstronomers believe AMD experimented with twice the amount of cache in the stack for the Navi 31, but that failed to provide enough additional performance to make it worth doing.
Hence, some slight disappointment with these supposedly specs, and a second point that raised little concern is the alleged Navi 32 configuration, which Angstronomics assures will run with 7,680 cores instead of 8192 as previously thought. The Infinity cache is apparently set to 64MB, or maybe 128MB for the 3D stacked model (but again it looks like AMD may have rethought this idea and then ditch it as it doesn’t make sense from a cost point of view the extra performance obtained) .
As for the Navi 33, the mid-range offering, it will be much smaller if Angstronomics is for the money, with a chip rumored to be around 400mm², but actually weighs around 200mm². It is rumored to work with 32MB of Infinity Cache.
As for the cooling system of the flagship RDNA 3 graphics card, AMD is believed to retain a similar three-fan system to the existing flagship, but it will be slightly taller. According to the informant, the GPU will use a pair of 8-pin PCIe power connectors.
Analysis: Are you worried about all of this? Well, maybe Nvidia should be worried …
As mentioned, some people are concerned about the elements of this supposed spec and some of the bits that seem to be a bit shortened – or significantly in the case of the Infinity Cache – compared to the previous vines chatter. However, there’s no need to worry, and indeed, potentially Nvidia should be worried, and here’s why.
Interestingly, Moore’s Law is dead (opens in a new tab) (MLID), a prominent leak from YouTube, quickly got hooked on the topic, and after talking to multiple sources, it basically endorses everything Angstronomics presents here. That said, there is one key difference, namely that with the Navi 32 GPU, MLID still hears that it will be 8,192 cores rather than the 7,680 cores mentioned in this new leak. Of course, MLID still admits that its sources can be wrong, so it could go both ways (and 7,680 cores are certainly doable).
The MLID sources largely confirm the accuracy of the specs and the smaller die sizes in question here, and the reduced burden on the Infinity cache – but here’s the key thing on that front. MLID notes that the performance estimates it hears haven’t really changed, and that it still expects AMD to uphold rumors of performance leaps we’ve heard of in the past – enough to compete with the Nvidia RTX 4000 series – even though the specs have been modified along the lines above.
Even if the performance turns out to be a bit weaker than Nvidia, there shouldn’t be much of a difference that MLID theorizes. And we must bear in mind that considering the aforementioned smaller matrix sizes, AMD will be cheaper to produce these RDNA 3 GPUs – and therefore price them to attacking Nvidia RTX 4000 models. This could be the most exciting thing about next-gen cards if AMD starts aggressively pricing. Nvidia could be in trouble – especially considering its more power-hungry GPUs (at least coupled with rumors) could spell other complications, such as a PSU update.
Not to mention that Nvidia is reportedly having a hard time deciding on the Lovelace next-gen launch schedule, which could allow AMD to be there first with RDNA 3 graphics cards and most of all with a sales advantage over these other factors. Of course, we shouldn’t get carried away by any of these rumors, but we can imagine this latest leak to worry Nvidia on several fronts.
Oh, and as for Intel – Angstronomics mentions that the Navi 33 outperforms the flagship Arc Alchemist GPU while being more energy efficient and costs less than half as much, which of course would be a bit of a torpedo to sink Team Blue’s desktop GPUs (which are already struggling bad even though it is, without competing new generation products).