Properly, it appears like Micron has formally commenced the mass manufacturing of its next-gen GDDR6X reminiscence chips providing speeds of as much as 24 Gbps.

Micron’s 24 Gbps GDDR6X Memory Enter Mass Manufacturing, Coming To NVIDIA’s Next-Gen GPUs Later This 12 months

Micron introduced earlier this yr that they have been readying their model new 24 Gbps GDDR6X reminiscence modules for future graphics playing cards. The announcement got here at a time when NVIDIA’s GeForce RTX 3090 Ti had simply entered the marketplace for which Micron had produced 16 Gb DRAM modules rated at as much as 21 Gbps speeds.

Now, Micron goes to supply even sooner pin speeds of 24 Gbps in a 16 Gb density which signifies that as soon as once more, we might be 2 GB VRAM capacities and as much as 24 GB capacities throughout a 384-bit bus interface. Similar to the GeForce RTX 3090 Ti which featured all of its reminiscence modules on the entrance of the PCB, the next-gen GeForce RTX 4090 would function its 12 GDDR6X modules on the entrance, permitting for extra cooling in comparison with options that function reminiscence modules on the again of the PCB such because the RTX 3090 (Non-Ti).

Though the complete 24 Gbps speeds aren’t anticipated to be utilized by the NVIDIA GeForce RTX 4090 (rumored to supply as much as 21 Gbps die speeds), the 24 Gbps GDDR6X reminiscence modules can provide as much as 1.152 TB/s of bandwidth which is a rise of 14% versus the 1.008 TB/s supplied on present 21 Gbps GDDR6 reminiscence modules.

Following are the bandwidth numbers you possibly can anticipate with a 24 Gbps DRAM resolution:

• 512-bit resolution – 1.5 TB/s
• 384-bit resolution – 1.1 TB/s
• 320-bit resolution – 960 GB/s
• 256-bit resolution – 768 GB/s
• 192-bit resolution – 576 GB/s
• 128-bit resolution – 384 GB/s
• 92-bit resolution – 276 GB/s
• 64-bit resolution – 192 GB/s

The NVIDIA GeForce RTX 4090 goes to be amongst the primary next-gen graphics card to make the most of the newest GDDR6X reminiscence modules when it launches later this yr. On condition that these chips are probably going to be downclocked to hit the facility targets, there might be an enormous overclocking potential for fanatics to benefit from.

AMD then again is more likely to rely upon its companion, Samsung, for its next-gen RDNA 3 lineup. Samsung can be engaged on a 24 Gbps GDDR6 reminiscence module which is predicted to hit mass manufacturing quickly.

(*24*)

GRAPHICS MEMORY	GDDR5X	GDDR6	GDDR6X
Workload	Graphics	GraphicsAI Inference Accelerator	GraphicsAI Inference Accelerator
Platform (Instance)	TitanX	Titan RTX RX5700 XT	GeForce® RTX™ 3090Ti GeForce® RTX™ 3080Ti
Variety of Placements	12	12	12
Gb/s/pin	11.4	14-16	19-24
GB/s/placement	45	56-64	76-96
GB/s/system	547	672-768	912-1152
Configuration (Instance)	384 IO (12pcs x 32 IO package deal)	384 IO (12pcs x 32 IO package deal)	384 IO (12pcs x 32 IO package deal)
Body Buffer of Typical System	12GB	12GB	24GB
Common Machine Energy (pJ/bit)	8.0	7.5	7.25
Typical IO Channel	PCB (P2P SM)	PCB (P2P SM)	PCB (P2P SM)

Information Supply: Harukaze5719