The CU count of AMD's next-generation Navi 3x GPUs based on the RDNA3 graphics architecture has been reduced. The Navi 31 GPU would get 12288 stream processors spread among 48 workgroup processors, 12 shader arrays, and 6 shader engines. This is still a 1.4x gain over the Navi 21.
With half the number-crunching power of the Navi 31, the Navi 32 silicon is the series' second biggest. Each WGP has 8 SA and 4 SE stream processors. The Navi 32 silicon powers RX 6700 successors. The Navi 33 has half the muscle of the Navi 32 and a third of the muscle of the Navi 31. That's 4,096 streams processed by 16 WGP, 4 SA, and 2 SE. However, we've heard reports about AMD doubling their Infinity Cache memory. RDNA3 should also enhance ray tracing performance as much of the pipeline uses fixed-function hardware.
As always, please do consider this to be what it is, a rumor. The below table courtesy goes to videocardz.
Rumored Specs AMD RDNA3 GPUs | |||
---|---|---|---|
AMD NAVI 31 | AMD NAVI 32 | AMD NAVI 33 | |
Node | TSMC 5nm + 6nm | TSMC 5nm + 6nm | TSMC 6nm |
Architecture | AMD RDNA3 | AMD RDNA3 | AMD RDNA3 |
GPU Package | Multi-Chip-Module (MCM) | Multi-Chip-Module (MCM) | Monolithic |
Estimated GPU Size | ~800 mm² (or more) | TBC | ~440 mm² |
Graphics Dies | 2 | 2 | 1 |
Shader Engines | 6 | 4 | 2 |
Shader Arrays | 12 | 8 | 4 |
RDNA Workgroups (WGP) | 60 -> 48 | 40 ->32 | 20 ->16 |
Compute Units | 120 ->96 | 80 ->64 | 40 ->32 |
Stream Processors | 15360 ->12288 | 10240 ->8192 | 5120 ->4096 |
GPU Clock | ~3.0 GHz | TBC | TBC |
Memory & Bus | GDDR6 256-bit | GDDR6 192-bit | GDDR6 128-bit |
Infinity Cache | 256 - 512MB | 256 - 384MB | 128 - 256MB |
Release | Q3/Q4 2022 | Q3/Q4 2022 | Q1 2023 |
Rumored SKU | Radeon RX 7900XT | Radeon RX 7700XT | Radeon RX 7600XT |
FP32 Performance | ~ 92 -> 73TFLOPS | TBC | TBC |
Rumored specs of RDNA3 architecture has changed a little - up to 12288 Stream Processors