AMD Navi 23 Radeon RX GPUs Feature 32 MB Infinity Cache, RDNA 2 Powered Van Gogh APUs & CDNA 2 Powered Instinct MI200 Without Infinity Cache


In the latest Linux Kernel Patch, AMD has confirmed the Infinity Cache sizes of its upcoming RDNA 2 Radeon RX GPUs based on the Navi 23 SKU. The company has also confirmed that neither RDNA 2 powered Van Gogh nor CDNA 2 powered Instinct accelerators will be featuring its Infinity Cache technology.

AMD RDNA 2 Navi 23 GPU Powered Radeon RX GPUs To Feature 32 MB Infinity Cache, A Quarter of Navi 21's Infinity Cache Pool

The information was spotted by Coelacanth-Dream (via Kepler_L2) and confirms cache sizes & GPU configurations of several RDNA 2 and CDNA 2 GPUs. The AMD Infinity Cache was first introduced on the Radeon RX 6000 series graphics cards featuring the RDNA 2 architecture. The Navi 21 SKU, codenamed Big Navi, was the first to get it but it was followed by the Navi 22 SKU recently & now we are also going to get a third SKU in the form of Navi 23.

AMD Entry-Level Navi 24 ‘RDNA 2’ Radeon RX GPUs Rumored For Q1 Launch, 120W Super-Clocked Design Aimed at RTX 3050 Ti & Intel ARC

AMD RDNA 2 Powered Navi 23 GPU For Radeon RX 6000 Graphics Cards

The patch confirms the Infinity Cache size which is listed as a level-3 cache (L3) for all three RDNA 2 GPUs that will be coming to desktop PCs. The Navi 21 GPU 'Sienna Cichlid' features the biggest size of 128 MB Infinity Cache followed by the Navi 22 GPU 'Navy Flounder' which features 96 MB of L3 or Infinity Cache. The AMD Navi 23 GPU 'Dimgry Cavefish' is going to be the next GPU in the lineup of Radeon RX 6000 series graphics cards and is going to feature 32 MB of Infinity Cache. That's a quarter of the L3 size featured on the Big Navi GPU.

It was previously expected that Navi 23 Radeon RX 6000 GPUs would feature 64 MB of Infinity Cache but that rumor is debunked now. The specifications of the Navi 23 GPU suggest a total of 2048 stream processors & 32 raytracing units inside the GPU package. The GPU will also be packed with 32 MB of Infinity Cache which will help it boost its bandwidth by a huge margin. For memory, we are looking at up to 8 GB GDDR6 capacity that will operate across a 128-bit bus interface and offer pin speeds of up to 16 Gbps. This should bring the total bandwidth to 256 GB/s.  The Navi 23 GPU will also be the smallest of the Navi 2X family with a die size reported at around 236mm2.

AMD Radeon RX 6000 Series "RDNA 2" Graphics Card Lineup:

Graphics CardAMD Radeon RX 6600AMD Radeon RX 6600 XTAMD Radeon RX 6700AMD Radeon RX 6700 XTAMD Radeon RX 6800AMD Radeon RX 6800 XTAMD Radeon RX 6900 XTAMD Radeon RX 6900 XT Liquid CooledAMD Radeon RX 6900 XTX
GPUNavi 23 (XL)Navi 23 (XT)Navi 22 (XL?)Navi 22 (XT?)Navi 21 XLNavi 21 XTNavi 21 XTXNavi 21 XTXHNavi 21 XTXH
Process Node7nm7nm7nm7nm7nm7nm7nm7nm7nm
Die Size237mm2237mm2336mm2336mm2520mm2520mm2520mm2520mm2520mm2
Transistors11.06 Billion11.06 Billion17.2 Billion17.2 Billion26.8 Billion26.8 Billion26.8 Billion26.8 Billion26.8 Billion
Compute Units283236406072808080
Stream Processors179220482304256038404608512051205120
TMUs/ROPs112/64128/64TBA160/64240 / 96288 / 128320 / 128320 / 128320 / 128
Game Clock2044 MHz2359 MHzTBA2424 MHz1815 MHz2015 MHz2015 MHz2250 MHzTBA
Boost Clock2491 MHz2589 MHzTBA2581 MHz2105 MHz2250 MHz2250 MHz2345 MHz2435 MHz
FP32 TFLOPs9.0 TFLOPs10.6 TFLOPsTBA13.21 TFLOPs16.17 TFLOPs20.74 TFLOPs23.04 TFLOPs24.01 TFLOPs24.93 TFLOPs
Memory Size8 GB GDDR6 + 32 MB Infinity Cache?8 GB GDDR6 + 32 MB Infinity Cache12 GB GDDR6 + 96 MB Infinity Cache?12 GB GDDR6 + 96 MB Infinity Cache16 GB GDDR6 +128 MB Infinity Cache16 GB GDDR6 +128 MB Infinity Cache16 GB GDDR6 +128 MB Infinity Cache16 GB GDDR6 +128 MB Infinity Cache16 GB GDDR6 +128 MB Infinity Cache
Memory Bus128-bit128-bit192-bit192-bit256-bit256-bit256-bit256-bit256-bit
Memory Clock14 Gbps16 Gbps16 Gbps?16 Gbps16 Gbps16 Gbps16 Gbps18 Gbps18 Gbps
Bandwidth224 GB/s256 GB/s384 GB/s384 GB/s512 GB/s512 GB/s512 GB/s576 GB/s576 GB/s
Price$329 US$379 USTBA$479 US$579 US$649 US$999 US~$1199 US~$1199 US

AMD CDNA 2 Powered Aldebaran GPU For Instinct MI200 HPC Accelerators

AMD Ryzen Threadripper 5975WX ‘Chagall’ HEDT CPU With 32 Zen 3 Cores Benchmarked, 10% Faster Than 64 Core Threadripper 3990X

The other GPU is Aldebaran which is the codename for the chip powering AMD's Instinct MI200 accelerator. According to the data, the GPU carries 16 KB of L1 cache per CU which makes up 2 MB of the total L1 cache considering that the GPU will be packing 128 Compute Units. The GPU also carries 8 MB of shared L2 cache but carries 14 CUs per Shader Engine compared to 16 CUs per SE in the previous Instinct lineup. Regardless, it is stated that each CU on Aldebaran GPUs will have a significantly higher computing output.

Based on the naming convention itself, it looks like MI200 will be twice as fast as Instinct MI100 but to achieve this, AMD is expected to go all out with its MCM design. Do note that AMD's CDNA 2 GPU will be fabricated on a brand new process node & are confirmed to feature a 3rd Generation AMD Infinity architecture that extends to Exascale by allowing up to 8-Way coherent GPU connectivity.

AMD Radeon Instinct Accelerators 2020

Accelerator NameAMD Radeon Instinct MI6AMD Radeon Instinct MI8AMD Radeon Instinct MI25AMD Radeon Instinct MI50AMD Radeon Instinct MI60AMD Instinct MI100AMD Instinct MI200AMD Instinct MI300
GPU ArchitecturePolaris 10Fiji XTVega 10Vega 20Vega 20Arcturus (CDNA 1)Aldebaran (CDNA 2)TBA (CDNA 3)
GPU Process Node14nm FinFET28nm14nm FinFET7nm FinFET7nm FinFET7nm FinFETAdvanced Process NodeAdvanced Process Node
GPU Dies1 (Monolithic)1 (Monolithic)1 (Monolithic)1 (Monolithic)1 (Monolithic)1 (Monolithic)2 (MCM)4 (MCM)?
GPU Cores23044096409638404096768014,080?28,160?
GPU Clock Speed1237 MHz1000 MHz1500 MHz1725 MHz1800 MHz~1500 MHzTBATBA
FP16 Compute5.7 TFLOPs8.2 TFLOPs24.6 TFLOPs26.5 TFLOPs29.5 TFLOPs185 TFLOPsTBATBA
FP32 Compute5.7 TFLOPs8.2 TFLOPs12.3 TFLOPs13.3 TFLOPs14.7 TFLOPs23.1 TFLOPsTBATBA
FP64 Compute384 GFLOPs512 GFLOPs768 GFLOPs6.6 TFLOPs7.4 TFLOPs11.5 TFLOPsTBATBA
Memory Clock1750 MHz500 MHz945 MHz1000 MHz1000 MHz1200 MHzTBATBA
Memory Bus256-bit bus4096-bit bus2048-bit bus4096-bit bus4096-bit bus4096-bit bus8192-bitTBA
Memory Bandwidth224 GB/s512 GB/s484 GB/s1 TB/s1 TB/s1.23 TB/s~2 TB/s?TBA
Form FactorSingle Slot, Full LengthDual Slot, Half LengthDual Slot, Full LengthDual Slot, Full LengthDual Slot, Full LengthDual Slot, Full LengthDual Slot, Full Length / OAMTBA
CoolingPassive CoolingPassive CoolingPassive CoolingPassive CoolingPassive CoolingPassive CoolingPassive CoolingTBA

AMD RDNA 2 Powered Van Gogh APUs For Ultra-Low-Power Notebooks

The Patch also confirms that the hotly anticipated RDNA 2 and Zen 2 powered Van Gogh APUs for ultra-low-power notebooks won't be featuring Infinity Cache. The Van Gogh APUs do feature a 128 KB GL1 cache, 1 MB of L2 cache but the L3 cache is non-existent.

It makes sense for an entry-tier RDNA 2 iGPU which will be relying on LPDDR4X or even LPDDR5 memory for its primary memory bandwidth requirements. These APUs will be fabricated on the 7nm process node but we don't have an exact figure of what the maximum CU count on the RDNA 2 chip would be.

cache info
RDNA L1 (GL1) L2 (TCC) L3 / Infinity Cache (MALL)
Raven / Picasso
Renoir / Lucienne / Cezanne
--- 1M ---
Raven2 --- 128K 2 ---
Vega20 --- 4M ---
Arcturus --- 8M? ---
Aldebaran --- 8M ---
Navi10 / Navi12 128K 4M ---
Navi14 128K 2M ---
Van Gogh 128K 1M ---
Sienna Cichlid 128K 4M 128M
Navy Flounder 128K 3M 96M
Dimgrey Cavefish 128K 2M 32M