AMD RDNA 3 “Navi 3X” GPUs feature double the cache per compute unit and shader array

AMD has listed the newest info relating to the cache sizes of their upcoming “Navi 3X” RDNA 3 GPUs within the Linux patches.

AMD’s Subsequent-Technology RDNA 3 GPUs for the Navi 3X Line Double the Cache Dimension for Compute Items and Shader Matrix

Posted to the FreeDesktop Linux Repository by Aaron Liu of AMD and Found by Coelacanth-Dream and Kepler_L2we see the primary particulars relating to cache sizes in upcoming RDNA 3 GPUs such because the just lately leaked Navi 31, Navi 32 and Navi 33 chips.

As for the main points, the AMD RDNA 3 (GFX11) line of GPUs will function double the L0 vector cache for every compute unit or CU and likewise double the GL1 information cache (RDNA L1 cache for every Shader Array or SA Based on the brand new info, vector register file per SIMD will improve to 192 KB from 128 KB on RDNA 2, L0 vector/texture cache will improve from 16 KB to 32 KB per CU, GPU L1 information cache per Shader Array will improve 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 will even embody an RDNA 3 graphics core however in a monolithic package deal. L0 vector/texture measurement elevated from 16 KB to 32 KB, whereas L1 information (graphics) cache elevated from 128 KB to 256 KB. Register file measurement stays unchanged on Navi 33 GPUs and APUs Phoenix.

Cache Data Yellow Carp (Rembrandt) RDNA 3 (GFX11 Navi 31/32) Phoenix (GC 11.0.1, GFX1103)
L0 Vector Register File by SIMD 128KB 192KB 128KB
L0 vector information (per CU) 16KB 32KB 32KB
L1 Scalar Inst. (by WGP) 32KB 32KB 32KB
L1 scalar information (per WGP) 16KB 16KB 16KB
GL1 date (by SA) 128KB 256KB 256KB
L2 information 2048KiB (2MiB) 2048KiB (2MiB) 2048KiB (2MiB)

Coelacanth-Dream additionally says that every one RDNA 3 “Navi 3X” GPUs include VODP (Twin-Subject Wave32) directions, WMMA (Wave Matrix Multiply-Accumulate) assist, and efficiency by WGP has been considerably improved. The elevated GL1 cache is meant to enhance pixel processing efficiency and is among the many many adjustments AMD is making to its RDNA 3 Navi 3x household of GPUs.

AMD has confirmed that its RDNA 3 GPUs will arrive later this yr with an enormous efficiency enhance. The corporate’s Senior Vice President of Engineering, Radeon Applied sciences Group, David Wang, mentioned next-generation GPUs for the Radeon RX 7000 sequence will ship over 50% efficiency per watt over current RDNA 2 GPUs . A few of the key options of RDNA 3 GPUs highlighted by AMD will embody:

  • 5nm course of node
  • Superior chip packaging
  • Redesigned computing unit
  • Optimized graphics pipeline
  • Subsequent Technology AMD Infinity Cache
  • >50% efficiency/watts in comparison with RDNA 2

AMD will revamp the compute models inside RDNA 3 to offer enhanced ray tracing capabilities. Whereas there isn’t any point out of what these capabilities are if we needed to guess, we would say it is undoubtedly efficiency and a sophisticated function set on the RDNA 3 GPU core for playing cards Radeon RX 7000 graphics. AMD Radeon RX 7000 graphics playing cards are going to be launched later this yr and can ship an enormous leap ahead in gaming efficiency, so keep tuned for extra info within the coming weeks.

AMD RDNA 3 Navi 3X GPU Configurations (Preliminary)

GPU identify Navi 21 Navi 33 Navi 32 Navi 31 Navi 3X
Code identify Siena cichlid Vivid pink bonefish Wheathouse Bonito with plums To be decided
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 (to be decided)
GPU die measurement 520mm2 203mm2 (solely GCD) 200mm2 (solely GCD)
425 mm2 (with MCD)
308mm2 (solely GCD)
533 mm2 (with MCD)
To be decided
Shader engines 4 2 4 6 8
WGP-GPU 40 16 30 48 64
SP by WGP 128 256 256 256 256
Calculation models (per matrix) 80 32 60 96 128 (per GPU)
256 (Complete)
Cores (per matrix) 5120 4096 7680 12288 8192
Hearts (Complete) 5120 4096 7680 12288 16,384
Reminiscence bus 256 bit 128 bit 256 bit 384 bit 384 bit x2?
Reminiscence kind GDDR6 GDDR6 GDDR6 GDDR6 GDDR6
Reminiscence capability As much as 16 GB As much as 8 GB As much as 16 GB As much as 24 GB As much as 32 GB
Reminiscence velocity 16-18 Gbps To be decided To be decided 20 Gbps To be decided
Reminiscence bandwidth 512-576 GB/s To be decided To be decided 960 GB/s To be decided
Infinite Cache 128 MB 32 MB 64 MB 96/192 MB To be decided
Flagship SKU Radeon RX 6900 XTX Radeon RX 7600XT? Radeon RX 7800XT?
Radeon RX 7700XT?
Radeon RX 7900XT? Radeon Professional
PAD 330W ~150W ~250W ~350W To be decided
Launch This fall 2020 This fall 2022? This fall 2022? This fall 2022? 2023?

#AMD #RDNA #Navi #GPUs #function #double #cache #compute #unit #shader #array

Leave a Comment