AMD Instinct MI300X-192GB-HBM3 AI GPU Accelerator

• Brand: AMD
• Product family: AMD Instinct MI300 Series
• Accelerator: MI300X
• SKU: MI300X-192GB-HBM3
• Product type: Data-center GPU accelerator
• Architecture: AMD CDNA 3
• Form factor: OAM module
• GPU compute units: 304
• Matrix cores: 1,216
• Stream processors: 19,456
• Peak engine clock: Up to 2,100MHz
• Memory capacity: 192GB
• Memory type: HBM3
• Memory interface: 8192-bit
• Memory clock/data rate: Up to 5.2 GT/s
• Peak theoretical memory bandwidth: 5.3TB/s
• AMD Infinity Cache: 256MB
• FP64 vector: 81.7 TFLOPS
• FP64 matrix: 163.4 TFLOPS
• FP32: 163.4 TFLOPS
• TF32: 653.7 TFLOPS
• FP16: 1.307 PFLOPS
• BF16: 1.307 PFLOPS
• FP8: 2.615 PFLOPS
• INT8: 2.615 POPS
• FP16/BF16 with structured sparsity: 2.615 PFLOPS
• FP8/INT8 with structured sparsity: 5.23 PFLOPS
• Host interface: PCIe Gen 5 x16
• Host I/O bandwidth: 128GB/s
• Infinity Fabric links: Up to 7
• Infinity Fabric bandwidth: 128GB/s per link
• GPU peer-to-peer transport: Up to 896GB/s aggregate theoretical in supported configurations
• ECC: Full-chip ECC
• RAS: Yes
• Page retirement: Yes
• Page avoidance: Yes
• SR-IOV: Yes, up to 8 partitions

Why Choose This Product?

• 192GB HBM3 per accelerator for memory-intensive AI and HPC workloads
• 5.3TB/s memory bandwidth for high-throughput data processing
• CDNA 3 architecture purpose-built for data-center compute acceleration
• 304 compute units and 1,216 Matrix Cores for highly parallel workloads
• Strong FP64 capability for scientific and engineering computing
• PCIe Gen 5 x16 connectivity for high-speed host communication
• Infinity Fabric connectivity for scalable multi-GPU deployments
• ECC and RAS features for enterprise and HPC reliability requirements