Instinct MI300X specs
AMD · Shipping now · first available Dec 2023. Dense compute, memory, bandwidth and power from the vendor's documents; each precision stays separate.
All accelerators · Compare with Instinct MI325X
Specifications
| Dense BF16 · FP32 accumulate | 1,307 TFLOPS |
|---|---|
| Dense FP8 · FP32 accumulate | 2,614 TFLOPS |
| Dense FP4 · FP32 accumulate | Not published |
| Memory | 192 GB · HBM |
| Memory bandwidth | 5.3 TB/s |
| Maximum chip power | 750 W |
| Interconnect | Infinity Fabric |
| Process node | TSMC 5/6nm |
| First availability | Dec 2023 |
| Availability | Shipping |
CDNA 3 chiplet design; led on memory capacity at launch.
Dense tensor rates with FP32 accumulation, never sparse or FP16-accumulate headline figures. Memory is in decimal GB. The est marker flags a value its vendor-document review did not confirm.
Rent it
$2.99/GPU-hrCheapest on-demand · Hot Aisle · observed
2 providers: Hot Aisle, Oracle OCI. Quotes checked through .
MI300X rental prices from every provider →Value ratios
| $ per dense BF16 PFLOP-hour | $2.29 |
|---|---|
| Dense BF16 TFLOPS per watt | 1.74 |
| Bandwidth bytes per 1,000 BF16 FLOP | 4.06 |
| $ per dense FP8 PFLOP-hour | $1.14 |
| Dense FP8 TFLOPS per watt | 3.49 |
| Bandwidth bytes per 1,000 FP8 FLOP | 2.03 |
| $ per GB of memory per hour | $0.016 |
Prices divide the cheapest on-demand list price; theoretical ratios do not measure application speed.
Sources
Specifications reviewed . Dense rates confirmed; the datasheet's more precise 1,307.4/2,614.9 TFLOPS round to the table values.
First availability: AMD Advancing AI 2023 ↗ (2023-12-06): “announced the availability of the AMD Instinct MI300X”
Rental price: Hot Aisle published pricing ↗ (1x AMD MI300X 192GB VM (Small plan); 8 or 13 CPU cores and 224GB system RAM per VM; published per GPU-hour; billed by the minute).