Specifications
Product
Micron GDDR7 graphics memory, 24 Gb density
Density
24 Gb per device (3 GB)
Process
Micron 1β (1-beta) DRAM
Data Rate — Production
28 GT/s
Data Rate — Sampling
32 GT/s
Targeted Peak
Up to 36 Gb/s per pin
IO Width
384-bit (example configuration, 12 placements)
Bandwidth per Placement
112–128 GB/s
System Bandwidth
Greater than 1.5 TB/s at 384-bit
Frame Buffer Example
24 GB typical system
Comparison
GDDR6X 19–24 Gb/s; GDDR6 14–20 Gb/s
ECC
Supported at the GPU level (inline ECC on professional parts)
Workloads
AI inference accelerators, graphics, HPC
Device Packaging
GDDR7 BGA, 3 GB modules
Platform Examples
NVIDIA RTX PRO 6000 Blackwell, RTX PRO 5000 72 GB, RTX 5090 Laptop
Public Catalog Status
Listed by Micron as production and sampling entries
Supply Note
3 GB GDDR7 pricing has been a constraint on RTX 50 SUPER-class launches
Availability
Production / sampling — quote for allocation
Overview
Micron's 24 Gb GDDR7 device is the 3 GB-per-chip entry that moved the graphics memory market past the 2 GB ceiling. Built on Micron's 1β DRAM process, the catalog lists a production part at 28 GT/s and a sampling part at 32 GT/s, with the company stating the design can reach 36 Gb/s.
The density step matters more than the speed step for AI workstations. With 24 Gb devices, a 384-bit interface reaches 24 GB with twelve placements, and the same interface can carry 48 GB or 72 GB without adding chips on the back of the board. That is exactly the configuration NVIDIA uses on RTX PRO 6000 and on the 72 GB RTX PRO 5000 — professional cards whose value is the size of the model they can hold, not the frame rate they produce.
Micron is one of three suppliers shipping 24 Gb GDDR7 alongside Samsung and SK hynix. Because 3 GB GDDR7 supply and pricing have already delayed consumer RTX 50 SUPER-class launches, sourcing GDDR7 through a distributor with multiple qualified suppliers is the practical risk control. QS Compute supplies GDDR7-based boards and modules from all three memory vendors.
Key Benefits
3 GB per device reaches 24 GB on a 384-bit bus or 48–72 GB with back-side placements. 32 Gb/s sampling with 36 Gb/s targeted gives >1.5 TB/s of system bandwidth. 1β process improves power efficiency at the same speed, and three qualified suppliers (Micron, Samsung, SK hynix) reduce allocation risk on a part that has already constrained GPU launches.
Applications
AI inference accelerators and workstation GPUs holding 48–72 GB models; professional visualisation and simulation; high-refresh gaming GPUs; and any accelerator where memory capacity per board is the binding constraint.
Request a Quote — MICRON 24GB GDDR7 (3 GB) GRAPHICS MEMORY — 28 & 32 GB/S
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →