AMD confirmed that its RDNA 3 GPUs shall be coming later this 12 months with an enormous efficiency uplift. The corporate’s Senior Vice President of Engineering, Radeon Applied sciences Group, David Wang, stated that the subsequent-gen GPUs for Radeon RX 7000 sequence will supply 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

Players ought to count on fluid 4K gaming to be buttery easy on these graphics playing cards and with the FidelityFX suite providing subsequent-gen FSR, SAS, and SAM assist, we would even see playable 60 FPS at 8K decision.

AMD Navi 31 “Plum Bonito” GPU Configuration

The AMD Navi 31 GPU, the flagship RDNA 3 chip, would energy the subsequent-gen fanatic playing cards such because the Radeon RX 7900 XT graphics card. We have now heard that AMD will drop CU (Compute Models) in favor of WGP (Work Group Processors) on its subsequent-gen RDNA 3 GPUs. Every WGP will home twin CU (Compute Models) however with twice the SIMD32 clusters versus simply 2 on every CU inside RDNA 2. Rumors are that AMD has the choice to pick out between Samsung and TSMC for the 6nm die.

• AMD Navi 31: 12288 Cores, 384-bit Bus, 192 MB Infinity Cache, 308mm2 GPU Die @5nm
• AMD Navi 21: 5120 Cores, 384-bit Bus, 128 MB Infinity Cache, 520mm2 GPU Die @7nm

Based on the newest data, the AMD Navi 31 GPU with RDNA 3 structure is predicted to supply a single GCD with 48 WGPs, 12 SAs, and 6 SEs. This can give out a complete of 12,288 SPs or stream processors. This is a rise of two.4x in cores in comparison with the 5120 SPs featured on the Navi 21 GPU. The GPU or the Navi 31 GCD is claimed to measure 308mm² and can come packaged on TSMC’s 5nm course of node.

The Navi 31 GPU may even carry 6 MCD’s which can characteristic 16 MB Infinity Cache per die and are additionally prone to carry the 64-bit (32-bit x 2) reminiscence controllers that can present the chip with a 384-bit bus interface. Whereas this equals 96 MB of Infinity Cache which is decrease than the 128 MB featured on the present Navi 21 GPUs, there’s additionally a 3D-Stacked resolution within the works which was identified lately and that might double the Infinity Cache with 32 MB (16 MB 0-hello + 16 MB 1-hello) capacities for a complete of 192 MB of cache. This can be a 50% improve versus the present Navi 21 design and it additionally makes Navi 31 the primary GPU with each, chiplet and 3D stacked designs. These chiplets or MCD’s shall be fabricated on TSMC’s 6nm course of node and measure 37.5mm² every.

The outlet studies that there was a good larger 288 MB (16 MB 0-Hello + 16 MB 1-Hello + 16 MB 2-Hello) MCD package deal deliberate however that was scrapped attributable to value-profit causes. There are additionally talks of a reduce-down variant of the Navi 31 GPU which can characteristic 42 WGPs or 10,752 cores and 5 MCDs for 80MB Infinity Cache throughout a 320-bit bus interface.

(*3*)

Now, that is going to end in larger energy draw and AMD appears to have confirmed this a lot that their subsequent-technology graphics card lineup will characteristic larger energy consumption however they may nonetheless be a extra environment friendly possibility than what NVIDIA has to supply. The AMD Radeon RX 6950 XT already has a TBP of 335W so for a >2x efficiency acquire. The playing cards are anticipated to retain their twin 8-pin plug enter for energy and have an up to date triple-fan cooling design which is barely taller than the one at the moment in use.

AMD Navi 32 “Wheat Nas” GPU Configuration

The AMD Navi 32 GPU codenamed “Wheat Nas” can also be one of many two MCM GPUs featured within the RDNA 3 lineup. The GPU will characteristic a single GCD (Graphics Compute Dies) and 4 MCDs (Multi-Cache Die). The GPU or GCD is claimed to measure 200mm² whereas every MCD will retain the identical measurement at 37.5mm².

• AMD Navi 32: 7680 Cores, 256-bit Bus, 64 MB Infinity Cache, 200mm2 GPU Die @5nm
• AMD Navi 22: 2560 Cores, 192-bit Bus, 96 MB Infinity Cache, 335mm2 GPU Die @7nm

The GCD shall be geared up with 3 Shader Engines and every Shader Engine has 6 Shader Arrays (18 in complete). The GPU will characteristic a complete of 30 WGP or 60 Compute Models for a complete of 7680 cores. This rounds as much as >50% extra cores than the Navi 31 flagship.

Every MCD also needs to characteristic 2 reminiscence join hyperlinks (32-bit). That is a complete of 8 32-bit reminiscence controllers for a 256-bit bus interface. Every MCD will retain the 16 MB 0-Hello capability however there doesn’t appear to be any 1-Hello variant which implies that the chip could have simply 64 MB most configs. That Navi 32 GPU is claimed to launch in 2023 and would be the highest chip for cellular and mid-tier chip for desktops. The GPU can be first rolled out to the cellular section and is likely to be introduced at CES 2023 which makes best sense since we could have Phoenix Vary and Dragon Vary laptops with Zen 4 cores prepared by then.

AMD Navi 32 “Hotpink Bonefish” GPU Configuration

The AMD Navi 33 GPU codenamed “Hotpink Bonefish” will begin the monolithic section throughout the RDNA 3 household. The GPU will characteristic a single die. The die is similar to the flagship Navi 21 GPU and is predicted to make the most of a 6nm course of node for fabrication with a die measurement of 203mm².

• AMD Navi 33: 4096 Cores, 128-bit Bus, 32 MB Infinity Cache, 335mm2 GPU Die @6nm
• AMD Navi 23: 2048 Cores, 128-bit Bus, 32 MB Infinity Cache, 237mm2 GPU Die @7nm

The Navi 33 GCD is predicted to characteristic 2 Shader Engines and every Shader Engine has 2 Shader Arrays (2 per SE / 4 in complete). This rounds as much as 16 WGP’s or 32 Compute Models for a complete of 4096 cores which is larger than the Radeon RX 6800 “Navi 21 XL” GPU. The GPU will come packaged with 32 MB of Infinity Cache, the identical quantity because the Navi 23 GPU, and throughout a 128-bit huge bus. Similar to the Navi 32 GPUs, the Navi 33 chips shall be centered first on the cellular lineup with early graphics efficiency placing them forward of the Intel Arc Alchemist lineup at lower than half the fee & whereas pulling much less energy.

The outlet additionally talks about two new options coming to AMD RDNA 3 Navi 3x GPUs for Radeon RX 7000 sequence graphics playing cards corresponding to:

One of many options within the RDNA 3 graphics pipeline is OREO: Opaque Random Export Order, which is simply one of many many space saving methods. With gfx10, the pixel shaders run out-of-order, the place the outputs go right into a Re-Order Buffer earlier than shifting to the remainder of the pipeline in-order. With OREO, the subsequent step (mix) can now obtain and execute operations in any order and export to the subsequent stage in-order. Thus, the ROB will be changed with a a lot smaller skid buffer, saving space.

The Reminiscence Hooked up Final Stage (MALL) Cache blocks are every halved in measurement, doubling the variety of banks for a similar cache quantity. There are additionally modifications and additions that improve graphics to MALL bandwidth and cut back the penalty of going out to VRAM.

by way of Angstronomics

The AMD Radeon RX 7000 “RDNA 3” GPU lineup primarily based on the Nav 3x GPUs is predicted to launch later this 12 months with studies pitting the flagship Navi 31 launch first adopted by Navi 32 and Navi 33 GPUs.

AMD RDNA 3 Navi 3X GPU Configurations (Preliminary)

(*31*)

(*33*)

GPU Title	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 Velocity	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?