AMD Navi 3X GPUs and Phoenix APU to ship with RDNA3 architecture
LLVM patches confirm AMD is preparing at least four RDNA3 devices.
A good indicator that AMD is preparing a new graphics architecture, has always been a trace of open-source project development. In most cases, those are Linux drivers or development tools that are later used by the new hardware. In this case, we have AMD adding four new GFX11 GPU IDs added to the LLVM project, which is a set of compiler tools for various programming languages.
AMD is preparing at least four new IDs: GFX1100, GFX1101, GFX1102, GFX1103. In the past, some IDs were shared among graphics processors, but this has changed over the past few years, and now each GPU usually gets its own ID. Mostly, because they have design changes that prevent the use of the same tool sets or preset values. This will be particularly important once AMD releases its dual-chiplet Navi 31 and Navi 32 GPUs. Supposedly those two are included in this patch:
AMD Phoenix with RDNA3?
Furthermore, Navi 33 (GFX1102) and Phoenix APU (GFX1103) are expected to be among those new devices. This would also corroborate the rumors that Phoenix might indeed feature RDNA3 graphics. In other words, Ryzen 7000 laptop series, which might launch early next year (possibly around CES 2023), would already ship with new architecture, just a few months after AMD is (hopefully) unveiling its desktop GPUs.
Update: apparently AMD has confirmed through other patches that GFX1103 is indeed an APU. It does not confirm Phoenix, but as the author claims, there are not that many other devices that could fit into those specs:
|RUMORED AMD Ryzen APU Specifications|
|Package||BGA-1140 (25x32mm)||BGA-1140 (25x32mm)||BGA-1140 (25x32mm)||TBC|
|Die Size||156 mm2||175 mm2||~208 mm2||TBC|
|GPU Architecture||GFX9 (Vega)||GFX9 (Vega)||RDNA2||RDNA3|
|GPU Configuration||8 Compute Units||8 Compute Units||12 Compute Units||16 Compute Units|
|Memory Support||DDR4 / LPDDR4||DDR4 / LPDDR4||DDR5/ LPDDR5||DDR5 / LPDDR5|
|CPU Cores||8C / 16T||8C / 16T||8C / 16T||TBC|
Source: LLVM via @Kepler_L2