Glossary3 sections1 min read
What is HBM3e?
HBM3e is an extended generation of HBM3 stacked memory with higher speed per pin and larger stacks. It is the memory used on the NVIDIA H200 and on Blackwell data center GPUs such as the B200.
Why it matters when renting a GPU
HBM3e raises both capacity and bandwidth per GPU, so larger models fit on fewer GPUs and memory-bound inference runs faster.
Two GPUs with the same architecture can differ mainly by memory: the H200 keeps the Hopper compute of the H100 but moves to HBM3e.
Worked example
The H200 has 141 GB of HBM3e at 4,800 GB/s, against 80 GB of HBM3 at 3,350 GB/s on the H100 SXM: 1.8 times the capacity and 1.4 times the bandwidth, with the same Tensor Core throughput (989 BF16 TFLOPS on both).
Frequently asked questions
- Does HBM3e make a GPU faster?
- On memory-bound work such as LLM decoding, yes, because more bytes reach the compute units per second. Compute-bound work depends more on Tensor Core throughput.