Rent NVIDIA CMP 170HX GPU by the hour
The NVIDIA CMP 170HX is an Ampere-architecture datacenter GPU equipped with 64 GB of VRAM and a 250 W TDP. It delivers about 1516 GB/s of memory bandwidth alongside about 12.1 TFLOPS of single-precision compute at compute capability 8.0. On ONIE, you can deploy this card with per-hour billing, full SSH access, and ready templates for PyTorch/CUDA, vLLM, Ollama, and ComfyUI.
Specifications
VRAM per GPU
Architecture
Compute capability
FP32 compute
Memory bandwidth
Power limit
Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.
Available configurations
Live stock and prices, updated continuously. The final price is confirmed before you order.
Loading live prices…
What the CMP 170HX is good for
Memory-Bound LLM Inference
Serve large language models via vLLM or Ollama by leveraging the 64 GB VRAM capacity and about 1516 GB/s of memory bandwidth.
High-Memory Deep Learning
Run PyTorch/CUDA workloads that require substantial memory capacity and compute capability 8.0 without splitting models across devices.
Complex Image Generation Workflows
Load multiple large diffusion checkpoints and control nets simultaneously in ComfyUI within the 64 GB memory pool.
Data Processing and Simulation
Execute memory-intensive datacenter tasks that benefit from high memory throughput within a 250 W power envelope.
Is it the right GPU for you?
Choose the CMP 170HX when your workload is bottlenecked by memory capacity and bandwidth rather than raw single-precision compute. Opt for a GPU with higher compute throughput if you require faster training iterations, or select a smaller consumer card if your models do not need 64 GB of VRAM.
Compare with similar GPUs
Frequently asked questions
Is the CMP 170HX suitable for LLM inference?
Yes, its 64 GB of VRAM and about 1516 GB/s of memory bandwidth make it well suited for loading and serving large language models using templates like vLLM and Ollama. However, because it provides about 12.1 TFLOPS of single-precision compute, prompt processing will be slower than on datacenter GPUs with higher compute density.
How does this GPU compare to standard datacenter accelerators?
The CMP 170HX offers a large 64 GB memory buffer and high memory bandwidth similar to standard Ampere datacenter cards, while operating at a 250 W TDP. Its primary trade-off is a lower compute throughput of about 12.1 TFLOPS, making it more appropriate for memory-bound tasks than heavy compute workloads.
What software environments are available for this GPU?
The card supports CUDA compute capability 8.0 and can be launched on ONIE using pre-configured templates including PyTorch/CUDA, vLLM, Ollama, and ComfyUI. You also get direct SSH access to install custom libraries and manage your environment.
How does billing work when renting the CMP 170HX?
Instances on ONIE use per-hour billing, so you only pay for the duration your GPU remains active.
Rent the CMP 170HX by the hour
Create an account, choose a configuration and connect over SSH. Billing stops when you delete the GPU.
Rent CMP 170HX