AMD ‘confirms’ Navi 32 with 60 Compute Units, Navi 33 has 32
The rumor from Angstronomics on possible lower core count for Navi 32 GPU has now been indirectly confirmed by AMD.
AMD Navi 31 (RDNA3) GPU for Radeon RX 7900 Series, Source: AMD
The site alleged that Navi 32 GPU has 30 Workgroup processors instead of the previously rumored 32. This means that the CU (Compute Unit) count is not 64 but 60. Exactly this number is now mentioned by AMD developers for ROCm software in a file called “performance.hpp” published 2 days ago.
ROCm Software Platform, Source: AMD/Github
Furthermore, the same core refers to Navi 33 GPU with CU count of 32, which matches previous rumors as well. Given what we already know about RDNA3 architecture, this means that full Navi 32 GPU offers 3840 Stream processors while Navi 33 has 2048 respectively.
Not much has changed since August leaks, the rumored specs, especially from Angstronomics have been very accurate. The only change that we had to implement to our spec chart is higher memory speed (from rumored 18 Gbps to 20 Gbps) and PCIe Gen4 support instead of Gen5.
A bigger question is where will Navi 32 and 33 actually land in terms of Radeon SKUs. The Radeon RX 7900 desktop series appear to be exclusively based on Navi 31 GPU, while Navi 32 may very well be launched as Radeon RX 7800 desktop GPUs and possibly as laptop RX 7900M GPU.
Therefore, it is expected that AMD will talk more about Navi 32 GPU at CES 2023 in January. At the event the company very likely to reveal its Radeon RX 7000M series alongside new Ryzen 7000 laptop.
| RUMORED Specs of AMD NAVI 3X (RDNA3) GPUs | |||
|---|---|---|---|
| VideoCardz.com | |||
| Architecture | AMD RDNA3 | AMD RDNA3 | AMD RDNA3 |
| Codename | “Plum Bonito” | “Wheat Nas” | “Hotpink Bonefish” |
| Fabrication Node | TSMC N5/N6 | TSMC N5/N6 | TSMC N5/N6 |
| GPU Package | Multi-Chip-Module (MCM) | Multi-Chip-Module (MCM) | Monolithic |
| Chiplet Configuration | 1x GCD + 6x MCD | 1x GCD + 4x MCD | – |
| Transistors | 58B | TBC | TBC |
| GCD Size | 300 mm² | ~200 mm² | – |
| MCD | 6x ~37 mm² | 4x ~37 mm² | – |
| Full GPU | 522 mm² | ~348 mm² | ~203 mm² |
| Shader Engines | 6 | 3 | 2 |
| Shader Arrays | 12 | 6 | 4 |
| RDNA Workgroups (WGP) | 48 | 30 | 16 |
| Compute Units | 96 | 60 | 32 |
| Stream Processors | 6144 | 3840 | 2048 |
| FP32 Cores | 12288 | 7680 | 4096 |
| Memory & Bus | GDDR6 384-bit | GDDR6 256-bit | GDDR6 128-bit |
| Memory Bandwidth | 960 GB/s (20 Gbps) | 640 GB/s (20Gbps) | 320 GB/s (20Gbps) |
| Infinity Cache | 96 MB | 64 MB | 32 MB |
| PCIe Interface | PCIe Gen4 | PCIe Gen4 | PCIe Gen4 |
| Release | December 2023 | TBC | TBC |
| Series | Radeon RX 7900 desktop | Radeon RX 7800 desktop Radeon RX 7900 mobile |
Radeon RX 7600 desktop Radeon RX 7700 mobile |
Source: Github (ROCmSoftwarePlatform) via @Kepler_L2