DZ Labs runs NVIDIA Blackwell and Hopper servers in British Columbia, Canada. Bare-metal nodes for training, fine-tuning and inference, with no hardware to buy and no long-term lock-in.
Every node is yours alone: no virtualization, no shared GPUs. Rent a single node or a multi-node cluster, delivered as plain SSH, Slurm or Kubernetes.
GPUs, NVLink and local NVMe are all yours. You get the performance on the spec sheet, not a slice of it.
Run a short experiment by the hour, or lock in a monthly or reserved rate for a long training job. Pay for what you use.
Stable, mostly hydroelectric power and Canadian data residency, close to the US West Coast with low latency.
One clear invoice in US dollars. No egress fees on the quote you didn't see coming.
Multi-node NVLink clusters with fast networking for FSDP, DeepSpeed and Megatron-style distributed training.
Enough memory per GPU to serve 70B-class models on a single card, with vLLM or TensorRT-LLM.
Skip the six-figure hardware purchase. Start on one invoice, stop when the experiment ends.
GPU type, node count, duration, and whether you need cluster networking. We reply with a quote within one business day.
Sign the service agreement. We image the nodes with your drivers, CUDA, container runtime or scheduler.
Receive node addresses and keys and start running. Engineers stay on call; billing follows actual usage.
Tell us about your workload and we'll come back within one business day with current pricing and availability.
account@dzlabs.caVancouver, British Columbia, Canada