GPU Inferencing & Cloud

Dedicated GPU Infrastructure for AI at Scale

Bare-metal NVIDIA GPUs — from T4 to full DGX H100 nodes — with Triton, JupyterLab, MIG, InfiniBand, and automated fine-tuning jobs built in.

Browse GPU Plans Estimate Cost

Sub-60s Provisioning

10 Global Regions

DDoS Protected

Hourly or Monthly

GPU Metrics Dash

Auto Fine-Tuning

GPU Cloud Instances

Every instance includes CUDA 12, cuDNN 9, and a full ML framework stack. Commit monthly to save up to 13% vs hourly.

NVIDIA T4

16 GB GDDR6

65 TFLOPS FP16

PCIe 3.0 ×16

GPU-T4

$320/mo

$0.504/hr if hourly

vCPUs

System RAM

32 GB

NVMe Storage

200 GB

Bandwidth

10 TB

Best for

Inference servingSmall model fine-tuningVideo analyticsNLP pipelines

CUDA 12 + cuDNN 9
TensorRT 10 pre-installed
NVMe SSD storage
Snapshots included
IPv4 + IPv6

Deploy GPU-T4

Everything You Need to Ship AI in Production

Not just raw compute — every GPU instance ships with a production-grade ML platform built in.

Included on A100 / H100

Model Serving Stack

NVIDIA Triton Inference Server pre-configured with dynamic batching, model ensembles, and gRPC/REST endpoints out of the box.

All plans

JupyterLab Environment

One-click JupyterLab with GPU passthrough, pre-installed PyTorch 2, TensorFlow 2, JAX, and Hugging Face Transformers.

All plans

Persistent Checkpoints

Attach up to 50 TB of NVMe block storage for training checkpoints. Snapshots are crash-consistent and taken automatically every 6 hours.

H100×8 only

InfiniBand HDR (H100×8)

100 Gbps InfiniBand fabric with RDMA for sub-microsecond GPU-to-GPU communication across the DGX node. NCCL fully tuned.

A100 + H100

Automated Fine-Tuning Jobs

Submit LoRA / QLoRA / full fine-tuning jobs via a REST API. Jobs are queued, executed on your GPU, and results pushed to your S3-compatible bucket.

All plans

GPU Metrics Dashboard

Real-time GPU utilisation, VRAM consumption, power draw, temperature, PCIe / NVLink throughput, and SM occupancy via a hosted Grafana dashboard.

H100 + H100×8

Secure Enclaves & NVDEC

Confidential computing with NVIDIA Hopper TEE (H100 only). Hardware-accelerated video decode/encode via NVDEC/NVENC for media pipelines.

A100 / H100

Multi-Instance GPU (MIG)

Slice an A100 or H100 into up to 7 isolated MIG instances — each with dedicated VRAM, bandwidth, and fault isolation — via the Hostoya control panel.

Supported Models

Deploy open-source checkpoints in seconds or bring your own fine-tuned model.

Model	Category	Parameter sizes	Min. GPU
LLaMA 3.1 / 3.3	Text Generation	8B · 70B · 405B	T4
Mistral / Mixtral	Text Generation	7B · 8×7B · 8×22B	T4
Stable Diffusion XL	Image Generation	6.6B	T4
FLUX.1	Image Generation	12B	A100-40
Whisper v3 Large	Speech-to-Text	1.5B	T4
CLIP / SigLIP	Vision-Language	400M · 4B	T4
Gemma 2	Text Generation	2B · 9B · 27B	T4
DeepSeek-R1	Reasoning	7B · 32B · 671B	A100-80
Qwen 2.5	Text Generation	7B · 72B	A100-40
Emu3 / CogVideoX	Video Generation	8B	A100-80
AlphaFold 3	Protein Folding	—	H100
Custom BYOM	Any	Any checkpoint	T4

GPU Pricing Calculator

Estimate your monthly cost based on GPU type, daily usage hours, and inference throughput.

Configure Workload

GPU TypeNVIDIA T4 (16 GB GDDR6)

T4A100-40A100-80H100H100×8

Billing Commitment

Active GPU Hours / Day8 hr

1 hr24 hr (always-on)

Inference Requests / Hour5,000

100 req/hr100 K req/hr

Monthly Cost Estimate

NVIDIA T4 compute (monthly)$320.00

Inference API (1200K req/mo @ $0.08/K)$96.00

Network egressIncluded

JupyterLab + Triton servingIncluded

GPU metrics dashboardIncluded

Total / month$416.00

Effective rate: $1.733/hr GPU-hour

Deploy NVIDIA T4

No setup fees

Sub-60s provisioning

99.9% uptime SLA

Pause & resume any time