AMD has included the latest information regarding future cache sizes RDNA 3 “Navi 3X” GPUs Under Linux patches.
AMD’s next-generation RDNA 3 GPUs for the Navi 3X lineup to double the size of the compute cache and shadow array
Posted in FreeDesktop Linux Repository Discovered by AMD’s Aaron Liu Koalakanth Dream And the Kepler 2In this article, we will see the first details regarding cache sizes in upcoming RDNA 3 GPUs such as the recently leaked Navi 31, Navi 32 and Navi 33 chips.
Coming to the details, the AMD RDNA 3 (GFX11) GPU lineup will have double the L0 vector cache per compute or CU and also double the GL1 data cache (RDNA L1 cache per Shader or SA array. According to new information , the Vector log file per SIMD will increase to 192KB versus 128KB on RDNA 2, the L0 Vector/Texture cache will increase from 16KB to 32KB per CU, and the GPU L1 data cache per Shader array will increase from 128 KB to 256 KB while the L2 data cache will remain the same as RDNA 2.
There are also cache sizes included in AMD’s Navi 33 and Phoenix APUs which will also feature an RDNA 3 graphics core but in a monolithic package. L0 Vector/Texture is increased from 16KB to 32KB while the L1 data (graphics) cache is increased from 128KB to 256KB. The log file size remains the same on Navi 33 GPUs and Phoenix APUs.
| Cache information | Yellow carp (Rembrandt) | RDNA 3 (GFX11 NAVI 31/32) | Phoenix (GC 11.0.1, GFX1103) |
|---|---|---|---|
| L0 Vector Register File per SIMD | 128 KB | 192 KB | 128 KB |
| L0 vector data (per CU) | 16 KB | 32 KB | 32 KB |
| Inst scalar L1. (per WGP) | 32 KB | 32 KB | 32 KB |
| L1 standard data (per WGP) | 16 KB | 16 KB | 16 KB |
| GL1 history (per SA) | 128 KB | 256 KB | 256 KB |
| L2 . data | 2048 KB (2 MB) | 2048 KB (2 MB) | 2048 KB (2 MB) |
| L3 (mol) | Unavailable | yes | Unavailable |
Coelacanth-Dream also states that all RDNA 3 “Navi 3X” GPUs come with VODP instructions (Wave32 version) and WMMA (Wave Matrix Multiply-Accumulate) support and the performance for each WGP has been greatly improved. The increased GL1 cache is said to improve pixel processing performance and is among the many changes that AMD is bringing within its RDNA 3 Navi 3x GPU family.
AMD confirmed That its RDNA 3 GPUs are coming later this year with a massive performance boost. The company’s senior vice president of engineering, Radeon Technologies Group, David Wang, said the next generation of Radeon RX 7000 series GPUs will deliver over 50% performance per watt lift versus current RDNA 2 GPUs. Some of the key features of RDNA 3 GPUs highlighted by AMD will include:
- 5nm process node
- Advanced chiplet packaging
- Researched computing unit
- Enhanced graphics pipeline
- The next generation of AMD Infinity cache
- >50% Perf/Watt vs RDNA 2
AMD will re-engineer the computing modules inside RDNA 3 to deliver improved ray tracing capabilities. Although there is no mention of what these capabilities are if we are to guess, we would say that they definitely speak volumes about the performance and set of advanced features in the RDNA 3 GPU core of the Radeon RX 7000 graphics cards. AMD Radeon RX 7000 graphics cards will be launched later this year and will deliver a huge leap in gaming performance So stay tuned for more information in the coming weeks.
AMD RDNA 3 Navi 3X GPU Configurations (Initial)
| GPU name | Navi 21 | Navi 33 | Navi 32 | Navi 31 | Navi 3X |
|---|---|---|---|---|---|
| Code name | Sienna Cichlid | Hotpink Bonfish | wheat people | Bloom Bonito | to be announced later on |
| GPU operation | 7 nm | 6 nm | 5 nm / 6 nm | 5 nm / 6 nm | 5 nm / 6 nm |
| GPU Package | congener | congener | MCM (1 GCD + 4 MCD) | MCM (1 GCD + 6 MCD) | Million cubic meters (TBC) |
| GPU Template Size | 520 mm 2 | 203 mm2 (GCD only) | 200 mm2 (GCD only) 425 mm2 (with MCDs) |
308 mm2 (GCD only) 533 mm2 (with MCDs) |
to be announced later on |
| Shader Engines | 4 | 2 | 4 | 6 | 8 |
| GPU WGPs | 40 | 16 | 30 | 48 | 64 |
| SP per WGP | 128 | 256 | 256 | 256 | 256 |
| Calculation units (per template) | 80 | 32 | 60 | 96 | 128 (per GPU) 256 (total) |
| cores (per die) | 5120 | 4096 | 7680 | 12288 | 8192 |
| cores (total) | 5120 | 4096 | 7680 | 12288 | 16384 |
| Memory bus | 256 bit | 128 bit | 256 bit | 384 bit | 384 bit x2? |
| memory type | GDDR6 | GDDR6 | GDDR6 | GDDR6 | GDDR6 |
| Memory capacity | up to 16 GB | up to 8 GB | up to 16 GB | up to 24 GB | up to 32 GB |
| Memory speed | 16-18 Gbps | to be announced later on | to be announced later on | 20 Gbps | to be announced later on |
| Memory bandwidth | 512-576 GB/sec | to be announced later on | to be announced later on | 960 GB/sec | to be announced later on |
| infinity cache | 128 MB | 32 MB | 64 MB | 96/192 MB | to be announced later on |
| Flagship SKU | Radeon RX 6900 XTX | Radeon RX 7600 XT? | Radeon RX 7800 XT? Radeon RX 7700 XT? |
Radeon RX 7900 XT? | Radeon Pro |
| TBP | 330 W | ~ 150 watts | ~ 250 watts | ~ 350 W | to be announced later on |
| release | Fourth Quarter 2020 | Fourth quarter 2022? | Fourth quarter 2022? | Fourth quarter 2022? | 2023? |
“Alcohol maven. Evil bacon lover. Wannabe social media geek. Travel guru. Amateur introvert. Pop culture nerd.”