Data updated 2026-11-04 15:09 UTC · refreshed every 5 min
Current Price
$4.78
-3.4% (24h)
Available Now
48 GPU
Good availability
Total in Network
198 GPU
Live data
Utilization
75.8%
Stable
Market history
USD/hour (minimum · median)
Ready to rent
Infrastructure Growth
Estimated with our VRAM calculator for a single card: 8K context, 4 concurrent users, vLLM-style serving.
117B MoE · 5.1B active
~150 tok/s per stream
Where the memory goes
35B MoE · 3B active
~130 tok/s per stream
Where the memory goes
14B
~72 tok/s per stream
Where the memory goes
Estimates, not benchmarks: real usage depends on the runtime, context and batch settings.
Memory
80GB HBM3
FP32 Performance
134 TFLOPS
Tensor Performance
3958 TFLOPS
Memory Bandwidth
3.35 TB/s
Architecture
Hopper
TDP
700W
FP64 Performance
67 TFLOPS
CUDA Cores
16896
Tensor Cores
528
The NVIDIA H100 is the flagship Hopper AI accelerator with 80GB HBM3 memory, available in SXM, PCIe, and NVL form factors.
The NVIDIA H100 features 80GB of HBM3 memory with 3.35 TB/s bandwidth, making it excellent for large language models, ai training, hpc. With 3958 TFLOPS tensor performance and 134 TFLOPS FP32 compute, this NVIDIA Hopper GPU delivers exceptional performance for demanding workloads.
Currently, 198 GPU H100 GPUs are available in the GPUniq network with 75.8% utilization. The average rental price is $4.78 per hour, with 48 GPU ready for immediate deployment.
Prices are live minimum / median USD per hour on the GPUniq marketplace, as of 2026-11-04 15:09 UTC.
| GPU | VRAM | Architecture | FP32 | Tensor | Bandwidth | TDP | Min $/hr | Median $/hr | Available |
|---|---|---|---|---|---|---|---|---|---|
| H100 | 80GB HBM3 | Hopper | 134 TFLOPS | 3958 TFLOPS | 3.35 TB/s | 700W | $1.35 | $4.29 | 33 |
| H200 | 141GB HBM3e | Hopper | 134 TFLOPS | 3958 TFLOPS | 4.8 TB/s | 700W | $5.19 | $6.03 | 91 |
The current rental price for NVIDIA H100 is $4.78 per hour on GPUniq. Prices may vary based on demand and availability.
NVIDIA H100 is ideal for Large Language Models, AI Training, HPC, Scientific Computing, Deep Learning. With 80GB of HBM3 memory and 3958 TFLOPS tensor performance, it excels at demanding AI and compute workloads.
Currently, there are 48 GPU available for immediate rental out of 198 GPU total in the GPUniq network. The current utilization is 75.8%.
On a single NVIDIA H100, our VRAM calculator fits GPT-OSS 120B (INT4 / AWQ, 63.1 of 80 GB); Qwen3.6 35B A3B (FP8, 43.3 of 80 GB); Qwen3-14B (FP16, 35.6 of 80 GB) — estimated at 8K context with 4 concurrent users.
NVIDIA H100 features 80GB HBM3 memory, 134 TFLOPS FP32 performance, 3958 TFLOPS tensor performance, 3.35 TB/s memory bandwidth, and 700W TDP. It uses the Hopper architecture.
NVIDIA official specificationsAPI docsAbout GPUniqContact supportPrivacy policyTerms of use
$4.78/hr
H100 · 48 GPU