Six US regions online
H100, H200 and RTX 4090 GPUs by the hour, KVM servers from $6 a month and bare metal from $108. NVMe storage, 10 Gbps uplinks and root access in under a minute.
Every region sits in a carrier-neutral facility with redundant power and at least three upstream transit providers. Move workloads between cities over our own links without paying for transfer.
No proprietary layers between you and the machine. You get a KVM guest or a whole box, a real console and a network that stays up.
Every GPU is passed through whole, never sliced between tenants. Images ship with CUDA 12.6 and drivers ready, and every server gets a browser console, ISO mounts and rescue mode.
$ ssh root@203.0.113.42 root@dfw-gpu-07:~# nvidia-smi --query-gpu=name,memory.total --format=csv name, memory.total [MiB] NVIDIA H100 80GB HBM3, 81559 MiB NVIDIA H100 80GB HBM3, 81559 MiB root@dfw-gpu-07:~# python -c "import torch; print(torch.cuda.device_count())" 2 root@dfw-gpu-07:~#
Layer 3 and 4 attacks are scrubbed at the edge before they reach your server. No null-routes for small floods, no extra charge.
One dedicated IPv4 and a routed /64 on every server, with reverse DNS you set yourself.
Two free snapshots per server. Add nightly off-site backups kept for seven days.
Support is handled by the people who rack the hardware, around the clock.
Every plan includes NVMe storage, DDoS filtering and the 99.9% uptime SLA. GPUs bill by the hour, servers by the month, with no setup fees.
Side projects, bots, VPN endpoints
Web apps, small databases, CI runners
Production apps and game servers
Busy databases and heavy traffic
Single-tenant web and mail hosting
Game hosting, build farms, high clock work
Virtualization hosts and large databases
Anything else, send us a note and an engineer will reply.
Contact SupportVirtual servers are usually reachable within a minute of payment. Dedicated servers in stock are handed over within 4 hours, custom builds within 2 business days.
By the minute from prepaid account credit, with no minimum. Stop the server and billing stops. A 1-year reserved term costs the equivalent of ten months.
No. Each GPU is passed through to one server in full, with its own VRAM. Multi-GPU servers on A100, H100 and H200 are connected over NVLink inside the same chassis.
Ubuntu, Debian, Rocky, AlmaLinux, Fedora, FreeBSD and Windows Server images are ready to install. You can also upload any ISO and install it through the console.
Yes. VPS plans resize up in place with a reboot. Moving to another city is a snapshot copy over our backbone and is free.
Nothing is shut off. Your port is limited to 100 Mbps until the next cycle, or you can add 1 TB blocks for $2 each.
Virtual servers can be cancelled for a full refund within 7 days of the first order. Dedicated servers are billed monthly and can be cancelled before the next cycle.
Spam, open proxies, attacks, malware and cryptocurrency mining on VPS or GPU plans. The full list is in our Acceptable Use Policy.
Start on Thread for $6 a month, or an RTX 4090 for $0.79 an hour.