Skip to content
NVNEXUS
info@nvnexus.com
Talk to an Expert →
GPU Comparison

AMD Instinct MI325X vs NVIDIA H100 NVL

How the AMD Instinct MI325X and NVIDIA H100 NVL compare across memory, compute, power and form factor, with the stronger figure highlighted on each row. Vendor-specific figures (CUDA vs stream processors, Tensor vs matrix cores) are listed on their own rows.

Specification CDNA 3 AMD Instinct MI325X NVIDIA alternative: NVIDIA H200 SXM → NVIDIA H100 NVL HOPPER NVIDIA H100 NVL View Product →
Pricing
Price Not sold by NVNexus $39,600.00
Overview
Architecture AMD CDNA 3 Hopper
Memory
Memory 256 GB HBM3E 94 GB HBM3
Memory Bandwidth 6 TB/s 3.9 TB/s
Memory Interface 8192-bit –
Compute
Stream Processors (AMD) 19,456 (304 CUs) –
Matrix Cores / AI Accelerators (AMD) 1,216 –
FP32 163.4 TFLOPS (matrix) 60 TFLOPS
FP64 Tensor 163.4 TFLOPS (matrix) 30 TFLOPS / 60 TFLOPS
TF32 Tensor 653.7 TFLOPS / 1,307 (sparse) 835 TFLOPS (with sparsity)
FP16/BF16 Tensor 1,307 TFLOPS / 2,615 (sparse) 1,671 TFLOPS (with sparsity)
FP8 Tensor 2,615 TFLOPS / 5,230 (sparse) 3,341 TFLOPS (with sparsity)
INT8 Tensor 2,615 TOPS / 5,230 (sparse) 3,341 TOPS
Platform
Interface PCIe Gen5 x16 –
NVLink / Infinity Fabric Infinity Fabric, 896 GB/s aggregate NVLink bridges, PCIe Gen5
TDP 1000W 350–400W (configurable)
Form Factor OAM module (8-GPU platform) Dual PCIe dual-slot air cooled
Cooling Passive (chassis airflow) –
Transistors 153 billion –
Not sure which fits?

Talk it through with a specialist

Tell us your workload and we will recommend the right GPU, configuration, and sourcing options.

Talk to an Expert →

More Comparisons