H100
The reference is H100 SXM 80 GB. H100 NVL uses a different memory configuration; an offer must identify the exact variant.
Explore H100 →GLOBALATLAS AI · Compare GPU specifications and pricing
Compare GPU specifications and pricing
Compare GPU memory size, memory technology and architecture across the GlobalAtlas AI hardware catalog.
| Parameter | H100 | H200 | A100 |
|---|---|---|---|
| GPU memory | 80 GB | 141 GB | 80 GB |
| Memory technology | HBM3 | HBM3e | HBM2e |
| Architecture | Hopper | Hopper | Ampere |
| Workload profile | AI training, inference and accelerated computing | Memory-intensive LLM inference and scientific computing | Established CUDA workloads and model training |
| Reference variant | The reference is H100 SXM 80 GB. H100 NVL uses a different memory configuration; an offer must identify the exact variant. | 141 GB is GPU memory, not system RAM. Confirm the number of GPUs and required interconnect before choosing a training topology. | The reference is A100 80 GB. A100 also exists with 40 GB, so the model name alone is not enough to select capacity. |
| Price | Price on requestConfiguration and region to be confirmed | Price on requestConfiguration and region to be confirmed | Price on requestConfiguration and region to be confirmed |
Which configuration fits your workload?
H200
H200 is a candidate when an H100 memory budget is restrictive. Its larger VRAM does not automatically establish more requests per second.
Explore H200 →A100
This page uses the 80 GB reference. Compare total server cost and tested runtime behavior with H100 before moving an existing workload.
Explore A100 →Compare the same workload
Use the same model or scene, precision, batch size and software version. Record throughput, latency, peak memory and total server cost. This table contains hardware specifications, not measured application performance.