Well, it looks like Micron has officially started mass production of its next-generation GDDR6X memory chips that deliver speeds of up to 24 Gbps.
Micron’s 24Gbps GDDR6X memory enters mass production, coming to NVIDIA’s next generation GPUs later this year
Micron announced earlier this year that they were preparing new 24Gbps GDDR6X memory modules for future graphics cards. The announcement came at a time when NVIDIA’s GeForce RTX 3090 Ti was entering a market where Micron produced 16Gb DRAM modules with speeds up to 21Gbps.
Now, Micron will offer faster delivery speeds of up to 24Gbps at 16Gbps which means that again, we’ll be looking at 2GB for video and up to 24GB over the 384-bit bus interface. Just like the GeForce RTX 3090 Ti which houses all of its memory modules on the front of the PCB, the next generation GeForce RTX 4090 will have 12 GDDR6X modules on the front, allowing for more cooling compared to solutions featuring memory modules on the back of the PCB Like the RTX 3090 (Non-Ti).
While full 24Gbps speeds are not expected to be used by the NVIDIA GeForce RTX 4090 (rumored to offer die speeds of up to 21Gbps), 24Gbps GDDR6X memory modules can provide up to 1,152 TB/s of bandwidth which is a 14% increase versus the 1,008 TB/s available on current 21GDDR6 memory modules.
Here are the bandwidth numbers you can expect with a 24Gbps DRAM solution:
- Solve 512 bits – 1.5 TB/sec
- Solve 384 Bits – 1.1 TB/sec
- Solve 320 bits – 960 GB/sec
- Solve 256 bits – 768 GB/sec
- Solve 192 Bits – 576 GB/sec
- Solving 128 bits – 384 GB/sec
- 92-bit solution 276 GB/sec
- 64-bit solution 192 GB/sec
The NVIDIA GeForce RTX 4090 will be among the first next-generation graphics cards to use the latest GDDR6X memory modules when it launches later this year. Given that these chips will likely be slashed to hit power targets, there will be a huge overclocking potential for enthusiasts to take advantage of.
On the other hand, AMD will likely rely on its partner, Samsung, for its next-generation RDNA 3 lineup. Samsung is also working on a 24Gbps GDDR6 memory module which is expected to hit mass production soon.
| Graphics memory | GDDR5X | GDDR6 | GDDR6X |
|---|---|---|---|
| Workload | Graphics | AI inference accelerator | AI inference accelerator |
| Articles of Association (Example) | titanic | Titan RTX RX5700 XT |
GeForce® RTX ™ 3090Ti GeForce® RTX ™ 3080Ti |
| number of positions | 12 | 12 | 12 |
| gigabyte/sec/pin | 11.4 | 14-16 | 19-24 |
| GB/s/ mode | 45 | 56-64 | 76-96 |
| GB/s/system | 547 | 672-768 | 912-1152 |
| configuration (example) | 384 IO (12pcs x 32 IO pack) |
384 IO (12pcs x 32 IO pack) |
384 IO (12pcs x 32 IO pack) |
| Frame buffer for model system | 12 GB | 12 GB | 24 GB |
| Average device power (pJ/bit) | 8.0 | 7.5 | 7.25 |
| Typical I/O Channel | PCB (P2P SM) |
PCB (P2P SM) |
PCB (P2P SM) |
news source: Harukaze 5719
Products mentioned in this post
“Excited about social media. obsessed with beer. Evil constant. Pop culture lover. prone to bouts of apathy.”