For a while now, AMD’s Radeon division has been utilizing some fairly fishy names for its pre-release GPUs. No, we imply that actually—the names mix a colour and a kind of fish. For the RDNA 2 GPUs comprising the Navi 2x sequence, we had Sienna Cichlid, Navy Flounder, Dimgrey Cavefish, and Beige Goby.
The newest leaks counsel that the corporate is constant that pattern with its upcoming graphics processors, each the RDNA 3-based Navi 3x household in addition to its built-in graphics primarily based on the identical structure. Over at Angstronomics, leaker SkyJuice has posted up an in depth breakdown of the Navi 3x household that features a few new particulars.
Probably the most attention-grabbing a part of Angstronomics’ new data is that, apparently, RDNA 3 really has much less Infinity Cache than RDNA 2. That is stunning, because it goes in opposition to what we thought was the case earlier than. In keeping with SkyJuice, RDNA 3 can have simply 16MB of cache on every MCD, as a substitute of 32MB as presumed earlier than. 32MB per MCD would have given the RDNA 3 elements the identical cache allotment as RDNA 2, and clearly, 16MB per MCD is half of that.
The leak goes on to say that there are certainly variations that embody an extra 16MB of Infinity Cache by way of a stacked cache die (ala 3D V-Cache), however that “the efficiency profit is restricted given the price enhance.” This appears odd to us, because the tendency of higher-end Radeon 6000-series elements to lose extra efficiency than anticipated at excessive resolutions is well-documented, and it has lengthy been assumed that this was as a consequence of overrunning the Infinity Cache. If that’s certainly the case, then that implies that next-gen Radeons will nonetheless battle with this challenge.
Fascinatingly, the publish on Angstronomics explains that an RDNA3 Workgroup Processor (WGP) is definitely barely smaller in space than an RDNA2 WGP because of the elimination of legacy cruft and a few adjustments to the structure. That is notably spectacular as a result of RDNA3 WGPs are twice as vast as RDNA 2 WGPs. Meaning they’ve twice as a lot FP32 compute energy as a consequence of a doubling of the variety of ALUs within the WGP. This element helps to elucidate why the Navi 31 and Navi 32 Graphics Compute Cube (GCDs) appeared surprisingly small within the renders we noticed earlier than.
Over at Japanese {hardware} weblog Coelacanth’s Dream, the Coelacanth factors out a patch to the Linux ALSA challenge that provides help for AMD’s “Pink Sardine” platform. What on the earth is that? Effectively, that is the identify for the purple staff’s forthcoming Phoenix APUs. Savvy AMD followers may most likely deduce this, as the corporate’s Cezanne APUs had been code named “Inexperienced Sardine,” whereas the AMD Ryzen 6000 cellular processors code named Rembrandt are often known as “Yellow Carp.”
The Coelacanth additionally factors out that one other APU code identify was revealed not too long ago, and that was “Sabrina.” Sabrina appears to be the code identify for AMD’s low-power Mendocino APUs. That one’s not a color-and-fish code identify as a result of it is the moniker for the APU product itself, not strictly for its graphics portion.
The Navi 3x household of RDNA 3-based graphics processors is predicted to launch later this 12 months, presumably in October or November. Whereas the desktop Ryzen 7000 processors will doubtless launch comparatively quickly (in mid-September), the APUs primarily based on Phoenix usually are not anticipated till early subsequent 12 months.