Bring Enterprise AI to Your Existing Data Center
A Simpler Path to Enterprise AI
Generative AI and agentic AI are rapidly becoming essential across industries. As enterprises scale AI adoption, AMD Instinct™ MI350P PCIe® Card enables high-performance AI deployment on existing data center infrastructure.
AMD CDNA™4 Multi-Chip Architecture
Built on AMD's advanced 3nm multi-chip architecture, AMD CDNA™ 4 integrates 4 accelerated compute dies, 185 billion transistors, and 128 Compute Units to deliver higher compute density and scalable AI acceleration.
Support for BF16, FP8, MXFP8, MXFP6, and MXFP4 data types further improves AI inference efficiency and computational throughput fo1r enterprise workloads.
Technology Highlights
3nm Process Technology | 185B Transistors | 128 Compute Units | 4 Compute Dies (XCDs) |
AMD ROCm™ Software Stack
Open. Scalable. Ready for AI
Powered by AMD ROCm™ and the AMD Enterprise AI Stack, MI350P provides optimized software libraries, including rocAL™ for accelerated image and video decoding, preprocessing, and augmentation, enabling efficient enterprise AI inference pipelines.
New in ROCm™ 7
- Support for AMD Instinct™ MI350 GPUs
- New FP6/FP4 data types for large-scale AI
- Enhanced HIP 7.0 portability
Ideal for Enterprise AI Inference
Engineered for low-latency and memory-intensive compute, delivering high performance and flexible deployment.
LLM | SLM | RAG Use Cases | ~200-250B Model Size |
Up to 8* PCIe® Cards per node | 144 GB HBM3E Memory |
*Varies by model
Deploy AMD Instinct™ MI350P PCIe® Cards with MiTAC
MiTAC TN85-B8261
- Support (4) AMD Instinct™ MI350P PCIe® Cards*
- Dual AMD EPYC™ 9005 Processors
- (24) DDR5-6400 DIMM Slots
- (2) HH/HL PCIe Gen 5 x16 Expansion Slots
- (8) NVMe U.2 Storage Bays
- (1+1) 2700W PSU, 80 plus Titanium*
*PSU redundancy is supported for GPUs under 450W.
Stay Connected
Get industry insights, news, and event highlights.