Loading the catalog…
MI300X (AMD)
Blurred values are subscriber-only. Sign in to start a free trial — 2 full pages a day — or subscribe for unlimited access.
| Process Node | 5nm FinFET |
|---|---|
| Host Interface | 16-lane PCIe® Gen 5 host interface |
| Tdp W | 750 |
| Peak Engine Clock Mhz | 2100 |
| Gpu Compute Units | 304 |
| Architecture | CDNA 3 |
| Gpu Memory Gb | 192 |
| Memory Bandwidth Tb S | 5.3 |
| Interconnect | 7x 128 GB/s |
| Memory Bus Width Bits | 8192 |
| Accelerator Form Factor | OAM |
| Memory Type | HBM3 |
| Fp16 Bf16 Tflops Dense | 1307.4 |
| Fp8 Tflops Dense | 2614.9 |
| Fp16 Bf16 Tflops Sparse | 2614.9 |
| Fp32 Tflops | 163.4 |
| Int8 Tops Dense | 2614.9 |
| Amd Infinity Cache Last Level | |
| Document Type | |
| Fp64 Tflops | |
| Fp8 Tflops Sparse | |
| Int8 Tops Sparse |
| Matrix Cores |
|---|
| Media Engines |
|---|
| Product Name |
|---|
| Ras Features |
|---|
| Software Platform |
|---|
| Sr Iov Support |
|---|
| Stream Processors |
|---|
| Tf32 Tflops Dense |
|---|
| Tf32 Tflops Sparse |
|---|
| SKU | MI300X |
|---|
| Product description | |
|---|---|
| Compute units (high-throughput) | |
| Memory (HBM3 on GPU accelerator) | |
| Peak AI/ML workload performance advantage vs MI250X (FP8 with sparsity vs FP16) | |
| Peak HPC workload advantage vs MI250X (FP32) | |
| AI Peak Theoretical Performance - TF32 | |
| AI Peak Theoretical Performance - FP16 | |
| AI Peak Theoretical Performance - BFLOAT16 | |
| AI Peak Theoretical Performance - INT8 | |
| AI Peak Theoretical Performance - FP8 | |
| HPC Peak Theoretical Performance - FP64 vector | |
| HPC Peak Theoretical Performance - FP32 vector | |
| HPC Peak Theoretical Performance - FP64 matrix | |
| HPC Peak Theoretical Performance - FP32 matrix | |
| I/O Dies (IODs) | |
| Memory clock | |
| Scale-out network bandwidth | |
| Footnote - MI200 Series support | |
| Architecture generation | |
| CDNA 3 architecture | |
| Infinity Fabric links | |
| Platform / base board | |
| OAM module - Accelerated compute dies (XCDs) | |
| OAM module - L1 cache per CU | |
| OAM module - Shared L2 cache | |
| OAM module - AMD Infinity Cache shared across XCDs | |
| OAM module - Decoders | |
| OAM module - HBM3 memory | |
| OAM module - Virtualization | |
| Local memory bandwidth | |
| Inter-GPU bidirectional bandwidth | |
| Supported AI & ML frameworks | |
| Copyright |
