CPU
| Number of Cores | 128 |
| Vector Processor - FP64/FP32/Integer | 4 x 512-bit per core |
| Frequency (GHz) | 4.5 |
| Multiprocessor Scalability | 1S |
| Pipeline | Out-of-Order, 8 instructions per clock |
| L1 Cache Size | 128 KB ICache, 64 KB Dcache, both with ECC |
| L2 Cache Size | 1 MB with DECTED ECC |
| L2/L3 Cache Size | 128 MB with DECTED ECC |
| Matrix Processor - TAI/FP4 | 4 x 2048-bit per core |
| Matrix Processor - FP8 | 3 x 2048-bit per core |
| Matrix Processor - BF16/TF32 | 2 x 2048-bit per core |
Memory
| Number of Memory Controllers | 8 |
| Maximum Memory Speed | DDR5-9600 |
| DIMMs per Channel | 2 |
| Maxiumum Capacity (256 GB DIMMs) | 4 TB |
| Maximum Capacity (2 GB DIMMs) | 32 TB |
I/O
| PCIe 7.0 Lanes | 96 |
| PCIe Controllers | 48 |
| C2C Lanes (Hardware Coherency) | N/A |
| C2C Bandwidth per Lane (Gb/sec) | N/A |
AI/ HPC
| Data Types | FP64, FP32, TF32, BF16, Int8, FP8, FP4, TAI |
| Sparse Matrix Support | Sparsity (2:4) |
AI Performance
| TAI (PF) | 38 |
| FP4 (PF) | 19 |
HPC Performance
| FP64 (TF) | 38 |
Package, Power, Thermal
| Package | 120 mm x 63 mm FCLGA |
| Maximum TDP (W) | 150 |
| Tj | 85C |
| Cooling | Air |
Markets
| Target Markets and Applications | Cloud, Databases, Storage |