Budget GPU Cloud
About TensorDock
TensorDock is a budget-focused GPU cloud provider offering on-demand access to H100, A100, L40S, and B300 GPUs at some of the lowest rates in GridStackHub's network. Founded with a focus on cost efficiency, TensorDock targets ML researchers and startups who need reliable GPU compute without hyperscaler pricing. No minimum commitments, transparent per-GPU billing.
- Lowest H100 pricing in network at $2.09/hr on-demand
- EU and US regions for latency-sensitive workloads
- No long-term commitments required
- Transparent per-GPU billing, no egress surprises
- B300 Blackwell access at $4.99/hr (early availability)
Ideal Use Cases
TensorDock excels at workloads where cost efficiency matters more than enterprise features. The H100 at $2.09/hr is particularly competitive for fine-tuning and medium-scale training jobs.
- Fine-tuning open-source LLMs (Llama, Mistral, Qwen)
- Batch inference at scale with cost constraints
- Computer vision training (ResNet, YOLO, SAM)
- RLHF and preference learning experiments
- Early-access B300 Blackwell workloads
💰TensorDock GPU Pricing — Live Rates
| GPU Model | VRAM | Instance Type | Hourly Rate | vs Network Avg | Availability | Status |
|---|---|---|---|---|---|---|
| H100 SXM 80GB | 80 GB | 1x H100 SXM | $2.09/hr | −30% vs avg | On-demand | Verified |
| A100 SXM 80GB | 80 GB | 1x A100 SXM | $1.15/hr | −47% vs avg | On-demand | Verified |
| L40S 48GB | 48 GB | 1x L40S | $0.59/hr | −63% vs avg | On-demand | Verified |
| B300 192GB | 192 GB | 1x B300 SXM | $4.99/hr | −18% vs avg | On-demand | Blackwell |
GridStackHub network average: H100 $2.99/hr on-demand · A100-80GB $2.19/hr · L40S $1.60/hr · B300 $6.12/hr. TensorDock beats network average on all 4 tracked GPU models.
⚡TensorDock GPU Specifications
H100 SXM5
80GB HBM3 · PCIe
A100 SXM
80GB HBM3e
L40S
48GB GDDR6
B300 SXM
192GB HBM3e · NVLink 5
vCPUs
8–24 cores per GPU
RAM (system)
64–384 GB
Regions
US + EU
Billing
Per-hour, on-demand
📊TensorDock vs Network — H100 Price Comparison
H100 on-demand hourly rate across key providers in GridStackHub's 34-provider network (sorted by price).
TensorDock
$2.09
H100 SXM on-demand
✓ BestRunPod
$2.49
H100 on-demand
Shadeform
$2.79
H100 best price
FluidStack
$2.40
H100 SXM5
Lambda
$4.29
H100 SXM on-demand
CoreWeave
$3.92
H100 SXM on-demand
Calculate Your TensorDock GPU Costs
Enter your workload — GPU hours, GPU count, training vs inference — and get a full cost breakdown across TensorDock and the full provider network.
❓Frequently Asked Questions
Is TensorDock reliable for production workloads?
TensorDock targets cost-sensitive workloads. For production with strict SLAs, CoreWeave or Lambda may offer better uptime guarantees. For development, fine-tuning, and batch inference where cost matters more than enterprise support, TensorDock is a strong choice at these pricing levels. GridStackHub recommends comparing uptime data across providers before committing to production workloads.
How does TensorDock H100 pricing compare to AWS and GCP?
TensorDock H100 at $2.09/hr is 70–85% cheaper than AWS ($32.77/hr for 8x H100 p5.48xlarge, ~$4.10/GPU) and GCP ($29.39/hr for 8x A100 a2-ultragpu, ~$3.67/GPU). The gap reflects TensorDock's focused GPU-cloud model vs hyperscaler overhead. For workloads that don't require hyperscaler integrations (SageMaker, GKE, Azure ML), TensorDock delivers equivalent GPU performance at a fraction of the cost.
Can I get TensorDock spot pricing?
TensorDock's current GridStackHub data shows on-demand rates only. For spot/interruptible pricing, Vast.ai and RunPod Community tend to offer lower rates. Check the live comparison table at /gpu-cloud-pricing-comparison for the latest spot prices across all providers.
Does TensorDock support multi-GPU clusters?
Current data shows single-GPU instances. For multi-GPU training (NVLink/NVSwitch), CoreWeave and Lambda offer better configurations. TensorDock's strength is single-GPU cost efficiency — evaluate multi-node needs separately.
Is B300 at $4.99/hr on TensorDock worth it vs H100?
The B300 carries 192GB HBM3e (vs H100's 80GB) at 2.4x the price. For inference on 70B+ parameter models where H100 requires quantization, B300 at $4.99/hr may deliver better cost-per-token. For smaller models that fit on H100, the H100 at $2.09/hr is the better value. GridStackHub's cost-per-token calculator at /cost-per-token can model your specific workload.