Loading the catalog…
MI350X PLATFORM - PID#253461432A (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 | TSMC 3nm/6nm FinFET |
|---|---|
| Host Interface | 8 PCIe Gen 5 x16 (128 GB/s) |
| Multi Instance Support | Up to 8 @ 36 GB |
| Tdp W | 1000 |
| Peak Engine Clock Mhz | 2200 |
| Gpu Compute Units | 2048 |
| Architecture | CDNA |
| Gpu Memory Gb | 288 |
| Memory Bandwidth Tb S | 8 |
| Interconnect | AMD Infinity Fabric mesh, direct connectivity between each GPU over 153.6 GB/s bidirectional links |
| Memory Bus Width Bits | 8192 |
| Accelerator Form Factor | UBB |
| Memory Type | HBM3E |
| Amd Infinity Cache Last Level | |
| Matrix Cores | |
| Media Engines | |
| Product Name | |
| Ras Features | |
| Stream Processors |
| System Total Memory Gb |
|---|
| SKU | MI350X PLATFORM - PID#253461432A |
|---|
| Product subtitle | |
|---|---|
| Number of GPUs | |
| Baseboard standard | |
| Accelerator module type | |
| Inter-GPU links per MI350X | |
| Host/IO connectivity per OAM | |
| HBM memory per accelerator | |
| Memory type per accelerator | |
| Memory bandwidth per accelerator | |
| Total coherent shared memory | |
| Datatype support (next-gen) | |
| GPU upgrade compatibility | |
| FP16 VECTOR | |
| FP16 MATRIX | |
| BFLOAT16 MATRIX | |
| INT8 MATRIX | |
| MXFP8 | |
| OCP-FP8 | |
| MXFP6 | |
| MXFP4 | |
| FP64 VECTOR | |
| FP32 VECTOR | |
| FP64 MATRIX | |
| FP32 MATRIX | |
| SCALE-UP AMD INFINITY FABRIC LINKS | |
| Decoders footnote | |
| Software stack | |
| AI framework Day-0 support | |
| ROCm release | |
| Inference quantization support | |
| Mixed-precision training datatypes | |
| Product ID (PID) | |
| Architecture diagram - Infinity Fabric connectivity | |
| Architecture diagram - I/O connectivity | |
| Architecture diagram - Memory per OAM | |
| Copyright |
