AMD Instinct MI300A 128GB vs AMD Instinct MI325X OAM

The Instinct MI325X is the newer part: CDNA 3, launched in 2025, against the Instinct MI300A's CDNA 3 from 2023. Newer architectures typically add lower-precision formats and better throughput per watt, so check the precision rows your workload actually uses.

Dense throughput for the Instinct MI300A against the Instinct MI325X: FP64 61 vs 82 TFLOPS, FP32 123 vs 163 TFLOPS, FP16 981 vs 1,307 TFLOPS, FP8 1,961 vs 2,615 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 128 GB against 256 GB, fed at 5.3 TB/s versus 6.0 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

Power budgets are 760 W for the Instinct MI300A and 1.0 kW for the Instinct MI325X. At FP32 that works out to 0.16 against 0.16 TFLOPS per watt.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.