Data updated 2026-11-04 15:09 UTC · refreshed every 5 min
Current Price
$6.23
+6.7% (24h)
Available Now
85 GPU
High availability
Total in Network
395 GPU
Live data
Utilization
78.5%
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.
229B MoE · 10B active
~119 tok/s per stream
Where the memory goes
122B MoE · 10B active
~119 tok/s per stream
Where the memory goes
35B MoE · 3B active
~109 tok/s per stream
Where the memory goes
Estimates, not benchmarks: real usage depends on the runtime, context and batch settings.
Memory
141GB HBM3e
FP32 Performance
134 TFLOPS
Tensor Performance
3958 TFLOPS
Memory Bandwidth
4.8 TB/s
Architecture
Hopper
TDP
700W
FP64 Performance
67 TFLOPS
CUDA Cores
16896
Tensor Cores
528
The NVIDIA H200 is the enhanced Hopper accelerator featuring 141GB of HBM3e memory with 4.8 TB/s bandwidth, delivering up to 1.9x faster inference on large language models compared to the H100.
The NVIDIA H200 features 141GB of HBM3e memory with 4.8 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, 395 GPU H200 GPUs are available in the GPUniq network with 78.5% utilization. The average rental price is $6.23 per hour, with 85 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 |
|---|---|---|---|---|---|---|---|---|---|
| H200 | 141GB HBM3e | Hopper | 134 TFLOPS | 3958 TFLOPS | 4.8 TB/s | 700W | $5.19 | $6.03 | 91 |
| H100 | 80GB HBM3 | Hopper | 134 TFLOPS | 3958 TFLOPS | 3.35 TB/s | 700W | $1.35 | $4.29 | 33 |
The current rental price for NVIDIA H200 is $6.23 per hour on GPUniq. Prices may vary based on demand and availability.
NVIDIA H200 is ideal for Large Language Models, AI Training, HPC, Scientific Computing, Deep Learning. With 141GB of HBM3e memory and 3958 TFLOPS tensor performance, it excels at demanding AI and compute workloads.
Currently, there are 85 GPU available for immediate rental out of 395 GPU total in the GPUniq network. The current utilization is 78.5%.
On a single NVIDIA H200, our VRAM calculator fits MiniMax M2 (INT4 / AWQ, 123.5 of 141 GB); Qwen3.5-122B-A10B (INT4 / AWQ, 75.3 of 141 GB); Qwen3.6 35B A3B (FP16, 82.4 of 141 GB) — estimated at 8K context with 4 concurrent users.
NVIDIA H200 features 141GB HBM3e memory, 134 TFLOPS FP32 performance, 3958 TFLOPS tensor performance, 4.8 TB/s memory bandwidth, and 700W TDP. It uses the Hopper architecture.
NVIDIA official specificationsAPI docsAbout GPUniqContact supportPrivacy policyTerms of use
$6.23/hr
H200 · 85 GPU