[HOPPER]
NVIDIA H100
The workhorse — the most-deployed AI training GPU.
The NVIDIA H100 (Hopper SXM5) is the industry standard for AI training in 2024–2026. 8×H100 NVLink nodes are the canonical pod for training 7B–70B models, and H100 PCIe is the most common production inference GPU.
VRAM
80 GB HBM3
FP8 TFLOPS
1,979
Mem BW
3.35 TB/s
TDP
700 W
Best for
- 7B–70B model training
- Production inference (FP16/FP8)
- Fine-tuning and RLHF
Benchmarks
| WORKLOAD | METRIC | VALUE |
|---|---|---|
| Llama-2 70B pretraining | tokens/sec/GPU | ~1,250 |
| Llama-3 8B inference | tokens/sec | ~3,800 |
| FP8 dense matmul | TFLOPS | 1,979 |
NVIDIA H100 availability from $12.90/hr
NVIDIA Hopper
8x NVIDIA H100 SXM5
Memory640GB HBM
RegionUS-East-2
Availability88%
ProviderLambda Labs
$24.50/hr
Deploy →NVIDIA Hopper
8x NVIDIA H100 PCIe
Memory640GB HBM
RegionEU-West-1
Availability96%
ProviderRunPod
$18.40/hr
Deploy →NVIDIA Hopper
4x NVIDIA H100 NVL
Memory376GB HBM
RegionEU-Central-1
Availability84%
ProviderCoreWeave
$12.90/hr
Deploy →Definition
What is Servers.Computer?
Servers.Computer is an AI compute routing and procurement layer that benchmarks, compares, and deploys GPU clusters (NVIDIA H100, H200, B200 and AMD MI300) across global cloud providers in real time.