AMD’s high-end graphics cards are rumored to be a little different than previously expected in terms of specs, going through a new leak – and while there’s a bit of disappointment with some claims made in this new spill, there’s really not nothing to worry about. In fact, it’s likely that Nvidia is worrying about some of the information released here…
Before we understand why this might be the case, let’s examine the leak itself, which comes from Angstronomics (via VideoCardz (opens in new tab)), a publication covering the semiconductor industry we’ve heard about. (Even though it’s not one of the more regular sources releasing information on Twitter, we find Angstronomics reliable enough – plus a major YouTube leaker more or less backs up these seriously in-depth details, and we’ll come back to this one.)
Angstronomics draws the line in RDNA 3! We’ve detailed key specs for Navi31, Navi32 and Navi33 that were finalized in 2020 and haven’t changed since! We also highlight some architectural changes, including OREO😋WGP, Cache and Array Sizes! https://t.co/n2qB7KiiBgAugust 12, 2022
Key points to this speculation include that AMD is focusing on “area, area, area”, meaning smaller chips that are still powerful enough to hit Team Red’s performance goals for RDNA 3 and also a focus in performance per watt (efficiency). Of course, the company has already boasted that it has achieved a 50% boost over RDNA 2 with the latter (the same leap that RDNA 2 made over the original RDNA architecture).
Apply your own spice liberally to all of this, but Angstronomics tells us that the Navi 31 – the flagship model for RX 7000 graphics cards – will go as rumored with one Graphics Chiplet Die (GCD) plus six Memory Chiplet Dies (MCDs), with 12,288 cores (known as ALUs). This flagship GCD will apparently be 308mm² in size, and the other big change from previous speculation mentioned is the size of the Infinity Cache which is apparently 96MB for the top dog model (a scaled-down Navi 31, presumably the non-XT version, will use 80MB).
The rumor mentioned sizes much larger than that, like 192 MB; and indeed the existing Navi 21 employs 128MB of Infinity Cache, so it would be smaller than that. Angstronomics reckons that AMD experienced double the cache stack for the Navi 31, but that didn’t work to provide enough additional performance to be worth it.
Hence some of the minor disappointments surrounding these specs, and the second point that has caused a bit of concern comes with the alleged Navi 32 configuration, which Angstronomics claims will work with 7,680 cores instead of 8,192 as previously believed. Infinity Cache is apparently set to ship at up to 64MB and maybe 128MB for a 3D stacked model (but then again, it looks like AMD may have pondered this idea and then abandoned it due to it not making sense in terms of cost for the extra performance gained).
As for the mid-range Navi 33 offering, this will be much smaller if Angstronomics is on the money, with the chip previously rumored to be perhaps up to 400mm² in size, but actually weighing in at over 200mm². It will supposedly run with 32 MB of Infinity Cache.
Regarding the RDNA 3 graphics card cooling system, AMD is believed to keep a similar three-fan system to the existing flagship, but it will be slightly louder. The GPU will use a pair of 8-pin PCIe power connectors, according to the leaker.
Analysis: Worried about all this? Well, maybe it’s Nvidia that should be worrying…
As mentioned, some people are fretting over elements of that supposed spec, and some of the bits and pieces that appear to have been scaled back slightly – or a lot in the case of Infinity Cache – compared to previous vine conversations. However, there’s no need to really worry, and in fact, it’s potentially Nvidia that should be worrying, and here’s why.
Interestingly, Moore’s law is dead (opens in new tab) (MLID), a prominent YouTube leaker, quickly jumped into the conversation, and after speaking to multiple sources, he pretty much backs up everything Angstronomics presents here. That said, there is one important difference, namely that for the Navi 32 GPU, MLID is still hearing that it will be 8,192 cores instead of the 7,680 cores mentioned in this new leak. Of course, the MLID still admits that its sources might be wrong, so it can still go either way (and 7,680 cores is certainly doable).
MLID sources largely support the accuracy of the smaller specs and array sizes mentioned here, and smaller Infinity Cache payloads – but here’s the main thing on that front. MLID notes that the performance estimates he’s been hearing haven’t really changed, and that he still expects AMD to keep up the rumors of performance leaps we’ve heard in the past – enough to compete with Nvidia’s RTX 4000 lineup – even if the specifications have been adjusted along the lines above.
Even if the performance ends up a little weaker than Nvidia’s, there shouldn’t be much difference, MLID theorizes. And we have to remember that because of the mentioned smaller die sizes, it will be cheaper for AMD to manufacture these RDNA 3 GPUs – and therefore price them to attack the Nvidia RTX 4000 models. high-end cards, if AMD is aggressive with pricing, Nvidia could be in trouble – mainly because their more power-hungry GPUs (at least going by rumor) could mean other complications like a power supply upgrade.
Not to mention that Nvidia is reportedly having a hard time deciding on the next-gen Lovelace release schedule, which could allow AMD to come first with RDNA 3 graphics cards, taking advantage of a head start for sales. these other factors. We shouldn’t get carried away by any of these rumors, of course, but this latest spill should be worrying for Nvidia on a number of fronts, we imagine.
Oh, and as for Intel – Angstronomics mentions the Navi 33 beating the Arc Alchemist GPU, being more energy efficient and costing less than half, which would obviously be a torpedo to sink Team Blue’s desktop GPUs (which are already struggling badly, even as it is, without rival next-gen products).