The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Micron’s catalog lists 24Gb GDDR7 memory chips in 28GT/s and 32GT/s versions. Each chip holds 3GB—not 24GB. Used in groups, the higher-density parts could let GPU makers build cards such as a 24GB model on a 256-bit bus, but Micron’s listing does not confirm any particular graphics card.
What Micron has listed
The two parts are GDDR7 DRAM components for integration into graphics-card boards, not user-installable memory modules. Micron’s catalog specifies a 768Mb ×32 configuration, 1.2V I/O voltage and a 266-ball TFBGA package measuring 12 × 14 × 1.10mm for both.
| Part number | Capacity per chip | Rated data rate | Catalog status | Configuration |
|---|---|---|---|---|
| MT68A768M32DF-28:A | 24Gb (3GB) | 28GT/s | Production | x32, 1.2V, 266-ball TFBGA |
| MT68A768M32DF-32:A | 24Gb (3GB) | 32GT/s | Sampling | x32, 1.2V, 266-ball TFBGA |
Those part numbers, specifications and statuses are from Micron’s GDDR7 part catalog. “Production” is a component catalog status, not proof that a finished consumer graphics card using the part is on sale. “Sampling” indicates a different availability stage; it should not be read as volume shipment.
Free tools Windows power users keep installed
One-click scans. No signup required.
Why 24Gb means 3GB, not 24GB
The lowercase b denotes bits; uppercase B denotes bytes. Divide 24 gigabits by eight bits per byte and the result is 3 gigabytes per chip. The “24GB” figure applies only to a card or system using enough chips to add up to that capacity. This is a 50% increase per package over a 16Gb chip, which stores 2GB.
#1 Best Overall
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
With one x32 chip for each 32 bits of a conventional bus, these are possible capacity configurations. They are arithmetic examples, not announced products; the GPU’s controller, board design, validation and product decisions determine which configurations are practical.
| Memory bus | Chips | With 16Gb (2GB) chips | With 24Gb (3GB) chips |
|---|---|---|---|
| 128-bit | 4 | 8GB | 12GB |
| 192-bit | 6 | 12GB | 18GB |
| 256-bit | 8 | 16GB | 24GB |
| 320-bit | 10 | 20GB | 30GB |
| 384-bit | 12 | 24GB | 36GB |
| 512-bit | 16 | 32GB | 48GB |
Micron also uses a 24GB framebuffer in a typical 384-bit GDDR7 system example, but that is a system-level capacity, not the amount stored on one chip. See Micron’s GDDR7 overview.
What 28GT/s and 32GT/s mean for bandwidth
GT/s describes billions of transfers per second per data pin. Micron’s part catalog uses GTPS; its product material also uses Gb/s for the effective data rate. These figures describe transfer rate, not the chip’s storage capacity. At the same bus width, 32GT/s offers about 14.3% more theoretical raw bandwidth than 28GT/s (32 ÷ 28 − 1). It does not imply a 14.3% increase in game performance.
Rank #2
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
Theoretical bandwidth is calculated as: transfer rate in GT/s × bus width in bits ÷ 8. The table shows decimal GB/s and assumes the full stated bus width and rate; it is not a measurement of a retail card.
| Bus width | At 28GT/s | At 32GT/s |
|---|---|---|
| 128-bit | 448GB/s | 512GB/s |
| 192-bit | 672GB/s | 768GB/s |
| 256-bit | 896GB/s | 1,024GB/s |
| 320-bit | 1,120GB/s | 1,280GB/s |
| 384-bit | 1,344GB/s | 1,536GB/s |
| 512-bit | 1,792GB/s | 2,048GB/s |
Actual GPU performance also depends on compute resources, cache, memory-controller efficiency, power limits and workload. Faster memory matters most when the application is constrained by memory bandwidth; greater chip density, by itself, increases capacity rather than bandwidth.
How GDDR7 differs from earlier graphics memory
Micron says its GDDR7 is built on its 1β DRAM process and uses PAM3 signaling. PAM3 encodes data differently from the signaling used by earlier GDDR generations; supporting GDDR7 therefore requires a compatible memory controller and board design, not a simple swap into a GDDR6 or GDDR6X graphics card.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Micron lists features including four independent channels per device, on-die ECC, command/address parity, 9-bit CRC and low-power modes. Its product material says GDDR7 can reach up to 32Gb/s and more than 1.5TB/s of system bandwidth in a 384-bit, 12-placement configuration. Micron also claims up to 60% higher bandwidth and more than 50% better power efficiency than its GDDR6 comparison, and up to 70% lower standby power with its new sleep mode. These are Micron’s stated comparisons and conditions, not a guarantee of improvement in every GPU or workload. Details are in Micron’s GDDR7 product information and its GDDR7 product brief.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →What this could mean for graphics cards
The main design opportunity is more framebuffer capacity without increasing the number of chips or widening the bus. For example, eight 3GB chips could provide 24GB on a 256-bit bus; twelve could provide 36GB on a 384-bit bus. Such capacities can benefit workloads that need a large working set, including high-resolution rendering, large texture collections, 3D creation, professional visualization and some local AI inference. The parts make those configurations technically possible; they do not establish that a card using them has been announced.
More VRAM can prevent capacity limits from forcing an application to move data out of the framebuffer, but it does not make a GPU’s processor, ray-tracing hardware or memory bandwidth faster. A card with less VRAM can outperform one with more when its GPU is substantially more capable or the workload does not need the extra capacity.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
What the catalog does—and does not—confirm
Micron’s catalog currently lists 16Gb and 24Gb GDDR7 devices with 28GT/s and 32GT/s speed options. It marks the 16Gb and 24Gb 28GT/s parts as Production, the 24Gb 32GT/s part as Sampling, and the 16Gb 32GT/s part as Contact Sales. Status labels indicate component availability through Micron, not broad retail availability of graphics cards.
The listing does not name a GPU customer or card model, set a retail launch date or price, or confirm that the 32GT/s part is shipping in volume. There are separate steps between a component appearing in a catalog and a finished card reaching buyers: GPU and board makers must qualify the memory, build compatible designs and bring products to market.
NVIDIA identifies its RTX 50-series family as using GDDR7, but that does not show that any specific RTX 50-series card uses Micron’s new 24Gb parts. Its official RTX 50-series page does not turn Micron’s component listing into confirmation of a particular capacity or future model. The same caution applies to predictions about other GPU makers.
What buyers should watch for
- Check the published specifications of the finished card—especially its VRAM capacity, bus width and memory data rate—rather than inferring them from a memory supplier’s catalog.
- Match capacity to your workload: large scenes, textures or models may need more VRAM, while many games and applications will be limited by other parts of the GPU first.
- Compare whole-card performance and power requirements. A higher memory rate or capacity alone is not a reliable proxy for frame rates or application speed.
- Do not treat these packages as DIY upgrade parts. GDDR7 needs compatible controllers, board routing, power design and validation, and is soldered to the graphics board.
Micron’s announcement is a capacity step for GPU designers: each 24Gb component holds 3GB, making larger framebuffer configurations possible on familiar bus widths. The 28GT/s part is listed as Production and the 32GT/s part as Sampling, but neither status confirms a specific retail card or launch. The practical news is greater flexibility for future GPU designs—not a confirmed 24GB graphics card.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
