AMD has listed down the most recent info relating to the Cache sizes of their upcoming RDNA 3 “Navi 3X” GPUs inside Linux patches.
AMD’s Subsequent-Gen RDNA 3 GPUs For Navi 3X Lineup To Double The Cache Dimension For Compute Models & Shader Array
Revealed over on the FreeDesktop Linux repository by AMD’s Aaron Liu and found by Coelacanth-Dream and Kepler_L2, we get to see the primary particulars relating to the cache sizes in upcoming RDNA 3 GPUs such because the Navi 31, Navi 32, and Navi 33 chips which have been not too long ago leaked.
Coming to the small print, the AMD RDNA 3 (GFX11) GPU lineup will function double the L0 vector cache for every Compute Unit or CU and likewise double the GL1 knowledge cache (RDNA L1 cache for every Shader Array or SA. As per the brand new info, the Vector Register File per SIMD will enhance to 192KB vs 128KB on RDNA 2, the L0 Vector/Texture cache will enhance from 16 KB to 32 KB per CU, the GPU L1 Information Cache per Shader Array will enhance from 128 KB to 256 KB whereas the L2 Information Cache will stay the identical as RDNA 2.
There are additionally cache sizes listed for AMD’s Navi 33 and Phoenix APUs which may also function an RDNA 3 graphics core however in a monolithic package deal. The L0 Vector/Texture dimension is elevated from 16 KB to 32 KB whereas the L1 Information Cache (Graphics) is elevated from 128 KB to 256 KB. The Register file dimension stays untouched on Navi 33 GPUs & Phoenix APUs.
| Cache Data | Yellow Carp (Rembrandt) | RDNA 3(GFX11 Navi 31/32) | Phoenix (GC 11.0.1, GFX1103) |
|---|---|---|---|
| L0 Vector Register File per SIMD | 128KiB | 192KiB | 128KiB |
| L0 Vector Information (per CU) | 16KiB | 32KiB | 32KiB |
| L1 Scalar Inst. (per WGP) | 32KiB | 32KiB | 32KiB |
| L1 Scalar Information (per WGP) | 16KiB | 16KiB | 16KiB |
| GL1 Date (per SA) | 128KiB | 256KiB | 256KiB |
| L2 Information | 2048KiB (2MiB) | 2048KiB (2MiB) | 2048KiB (2MiB) |
| L3 (MALL) | N/A | Sure | N/A |
Coelacanth-Dream additionally states that every one RDNA 3 “Navi 3X” GPUs include VODP (Twin-Situation Wave32) directions, WMMA (Wave Matrix Multiply-Accumulate) assist and the efficiency per WGP has been improved quite a bit. The elevated GL1 cache is claimed to enhance pixel processing efficiency and is amongst the numerous adjustments that AMD is bringing inside its RDNA 3 Navi 3x GPU household.
AMD confirmed that its RDNA 3 GPUs can be coming later this yr with an enormous efficiency uplift. The firm’s Senior Vice President of Engineering, Radeon Applied sciences Group, David Wang, mentioned that the next-gen GPUs for Radeon RX 7000 collection will provide over 50% efficiency per watt uplift vs the present RDNA 2 GPUs. Among the key options of the RDNA 3 GPUs highlighted by AMD will embody:
- 5nm Course of Node
- Superior Chiplet Packaging
- Rearchitected Compute Unit
- Optimized Graphics Pipeline
- Subsequent-Gen AMD Infinity Cache
- >50% Perf/Watt vs RDNA 2
AMD can be rearchitecting the compute items inside RDNA 3 to ship enhanced raytracing capabilities. Though there is no point out of what these capabilities are if we have been to guess, we might say it is positively speaking about efficiency and a set of superior options on the RDNA 3 GPU core for Radeon RX 7000 graphics playing cards. The AMD Radeon RX 7000 graphics playing cards are going to launch later this yr and provide an enormous leap in gaming efficiency so keep tuned for more information within the coming weeks.
AMD RDNA 3 Navi 3X GPU Configurations (Preliminary)
| GPU Identify | Navi 21 | Navi 33 | Navi 32 | Navi 31 | Navi 3X |
|---|---|---|---|---|---|
| Codename | Sienna Cichlid | Hotpink Bonefish | Wheat Nas | Plum Bonito | TBD |
| GPU Course of | 7nm | 6nm | 5nm/6nm | 5nm/6nm | 5nm/6nm |
| GPU Package deal | Monolithic | Monolithic | MCM (1 GCD + 4 MCD) | MCM (1 GCD + 6 MCD) | MCM (TBD) |
| GPU Die Dimension | 520mm2 | 203mm2 (Solely GCD) | 200mm2 (Solely GCD) 425mm2 (with MCDs) |
308mm2 (Solely GCD) 533mm2 (with MCDs) |
TBD |
| Shader Engines | 4 | 2 | 4 | 6 | 8 |
| GPU WGPs | 40 | 16 | 30 | 48 | 64 |
| SPs Per WGP | 128 | 256 | 256 | 256 | 256 |
| Compute Models (Per Die) | 80 | 32 | 60 | 96 | 128 (per GPU) 256 (Complete) |
| Cores (Per Die) | 5120 | 4096 | 7680 | 12288 | 8192 |
| Cores (Complete) | 5120 | 4096 | 7680 | 12288 | 16,384 |
| Reminiscence Bus | 256-bit | 128-bit | 256-bit | 384-bit | 384-bit x2? |
| Reminiscence Sort | GDDR6 | GDDR6 | GDDR6 | GDDR6 | GDDR6 |
| Reminiscence Capability | Up To 16 GB | Up To eight GB | Up To 16 GB | Up To 24 GB | Up To 32 GB |
| Reminiscence Pace | 16-18 Gbps | TBD | TBD | 20 Gbps | TBD |
| Reminiscence Bandwidth | 512-576 GB/s | TBD | TBD | 960 GB/s | TBD |
| Infinity Cache | 128 MB | 32 MB | 64 MB | 96/192 MB | TBD |
| Flagship SKU | Radeon RX 6900 XTX | Radeon RX 7600 XT? | Radeon RX 7800 XT? Radeon RX 7700 XT? |
Radeon RX 7900 XT? | Radeon Professional |
| TBP | 330W | ~150W | ~250W | ~350W | TBD |
| Launch | This autumn 2020 | This autumn 2022? | This autumn 2022? | This autumn 2022? | 2023? |