NVIDIA GH200 144GB HBM3e vs NVIDIA GH200 96GB HBM3

The GH200 144GB HBM3e is the newer part: Hopper, launched in 2024, against the GH200 96GB HBM3's Hopper 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 GH200 144GB HBM3e against the GH200 96GB HBM3: FP64 34 vs 34 TFLOPS, FP32 67 vs 67 TFLOPS, FP16 990 vs 990 TFLOPS, FP8 1,979 vs 1,979 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 144 GB against 96 GB, fed at 4.9 TB/s versus 4.0 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 1.0 kW for the GH200 144GB HBM3e and 1.0 kW for the GH200 96GB HBM3. At FP32 that works out to 0.07 against 0.07 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.