H200 vs H100 SXM
Manufacturer specifications of the H200 and the H100 SXM, side by side, per GPU. A value the manufacturer does not publish for the exact model is marked as such, never estimated.
NVIDIA H200 141GB
Hopper
- GPU memory
- 141 GB
- Memory bandwidth
- 4,800 GB/s
Live price on EverXCloud · from $4.19 per GPU-hour
Suited workloads
NVIDIA H100 SXM 80GB
Hopper
- GPU memory
- 80 GB
- Memory bandwidth
- 3,350 GB/s
Live price on EverXCloud · from $2.88 per GPU-hour
Suited workloads
Key differences
- The H200 has 1.8× the GPU memory of the H100 SXM (141 GB vs 80 GB).
- The H200 has 1.4× the memory bandwidth of the H100 SXM (4,800 vs 3,350 GB/s), which usually sets the pace of LLM token generation.
- Both have about the same dense BF16 tensor throughput (989 vs 989 TFLOPS).
- Both have about the same dense FP8 tensor throughput (1,979 vs 1,979 TFLOPS).
Specifications
| Specification | H200 | H100 SXM |
|---|---|---|
| GPU memory | 141 GB (higher) | 80 GB |
| Memory type | HBM3e | HBM3 |
| Memory bandwidth | 4,800 GB/s (higher) | 3,350 GB/s |
| FP16 tensor | 989 TFLOPS | 989 TFLOPS |
| BF16 tensor | 989 TFLOPS | 989 TFLOPS |
| FP8 tensor | 1,979 TFLOPS | 1,979 TFLOPS |
| Max power | 700 W | 700 W |
| Architecture | Hopper | Hopper |
| Interconnect | NVLink 4 (900 GB/s) · PCIe 5.0 | NVLink 4 (900 GB/s) · PCIe 5.0 |
The higher value of each measured row is shown in bold. Manufacturer specifications, per GPU. Tensor throughput is shown without sparsity. Values the manufacturer does not publish for this exact model are left out.
What fits in memory, on one GPU
| Workload | H200 | H100 SXM |
|---|---|---|
| LLM inference, 16-bit | ≈ 55B parameters | ≈ 30B parameters |
| LLM inference, 4-bit | ≈ 180B parameters | ≈ 100B parameters |
| Full training, mixed precision | ≈ 7B parameters | ≈ 4B parameters |
Cautious estimates counting 80% of memory for model weights; context length, batch size and framework change the real limit. Several GPUs together hold larger models.
H200 vs H100 SXM: frequently asked questions
Which has more memory, the H200 or the H100 SXM?
The H200 has 141 GB of GPU memory, against 80 GB for the H100 SXM.
Which is faster for LLM inference, the H200 or the H100 SXM?
Generating tokens is usually bound by memory bandwidth, and the H200 has 1.4× the bandwidth of the H100 SXM. Memory size then decides which models fit. Measure your own model before a large run.
Can I rent the H200 and the H100 SXM by the hour?
Yes, when sellers list them on the EverXCloud spot exchange: you buy GPU hours at the live price from a prepaid balance in euros, US dollars or pounds. Prices are shown excluding tax and EverXCloud is the merchant of record for every rental.
Be ready on launch day
Create your account now: you can top up your balance and rent the GPU that fits your workload as soon as the market opens.
Have GPU servers?
Businesses can join the EverXCloud Network and sell hours of their GPU servers on the spot exchange.