Instance Types
NeuroHash offers a range of NVIDIA GPU instance types optimized for different AI workloads.
Available Instance Types
NVIDIA B200 (Blackwell)
The latest generation. 3x faster training and 15x faster inference compared to prior architectures.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x HGX B200 | 180 GB HBM3e | 30 | 200 GiB | 1.5 TB | $3.79/hr |
| 8x HGX B200 | 1,440 GB HBM3e | 240 | 1,800 GiB | 25 TB | $30.32/hr |
Best for: Large language model training, next-gen AI research, high-throughput inference
NVIDIA H200
Near 2x the memory capacity of H100 with significantly higher bandwidth.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x H200 | 141 GB HBM3e | 26 | 200 GiB | 1 TB | $3.29/hr |
| 8x H200 | 1,128 GB HBM3e | 208 | 1,800 GiB | 25 TB | $26.32/hr |
Best for: Memory-intensive models, large batch training, long-context LLMs
NVIDIA H100 SXM
The industry standard for LLM training and large-scale AI workloads.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x H100 | 80 GB HBM3 | 26 | 200 GiB | 1 TB | $2.69/hr |
| 8x H100 | 640 GB HBM3 | 208 | 1,800 GiB | 25 TB | $21.52/hr |
Best for: LLM training, diffusion models, large-scale distributed training
NVIDIA A100 SXM
Proven, reliable GPU for training and inference at scale.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x A100 | 80 GB HBM2e | 30 | 200 GiB | 1.4 TB | $1.29/hr |
| 2x A100 | 160 GB HBM2e | 60 | 400 GiB | 2.8 TB | $2.58/hr |
| 4x A100 | 320 GB HBM2e | 120 | 800 GiB | 5.6 TB | $5.16/hr |
| 8x A100 | 640 GB HBM2e | 240 | 1,800 GiB | 21 TB | $10.32/hr |
Best for: General training, fine-tuning, production inference
NVIDIA A10
Cost-effective option for inference and lighter workloads.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x A10 | 24 GB GDDR6X | 30 | 200 GiB | 1.4 TB | $0.75/hr |
Best for: Inference, model serving, development, prototyping
NVIDIA GH200
Grace Hopper Superchip — CPU and GPU on a single module.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x GH200 | 96 GB HBM3 | 72 | 480 GiB | 2 TB | $2.49/hr |
Best for: CPU-GPU tightly coupled workloads, memory-bound models
NVIDIA RTX A6000
Professional-grade GPU for research and visualization.
| Config | GPU Memory | vCPUs | RAM | Storage | Price |
|---|---|---|---|---|---|
| 1x RTX A6000 | 48 GB GDDR6 | 14 | 100 GiB | 512 GB | $0.80/hr |
Best for: Research, visualization, smaller training runs
Choosing the Right Instance
| Workload | Recommended GPU | Why |
|---|---|---|
| LLM training (7B-70B) | 8x H100 or 8x H200 | High memory, fast interconnect |
| LLM fine-tuning | 1x H100 or 1x A100 | Good balance of cost and performance |
| Inference serving | 1x A10 or 1x A100 | Cost-effective for serving |
| Diffusion models | 1x-4x A100 | Good memory/compute ratio |
| Development & prototyping | 1x A10 | Lowest cost |
| Cutting-edge research | 8x B200 | Maximum performance |
Availability
Instance availability varies by region. When selecting an instance type in the launch modal, only regions with available capacity are shown. If your preferred GPU type is unavailable, try:
- A different region
- A different instance type
- Checking back later (capacity changes frequently)