Pages List
List view
Pricing
Base starting price: $4.50 / GPU / Hour - 1 Node of 8 GPUs, 1 week minimum
There are percentage discounts based on quantity, percentage prepayment and longer term reservations, which can bring the rate down significantly. Reach out with your use case and we can absolutely work together on a pricing model that works for you. Don’t worry, we are really nice and won’t spam you.
Pricing includes
- Virtual machines on our separate management servers (K8S, Slurm, Ray, ...) and we can help you set it all up
- Your choice of Linux operating system. Ubuntu is our preference. We preinstall latest ROCm and anything else you need
- 100G Internet and unlimited bandwidth
- Public IPv4/6 addresses
- Full access to BMC / iDRAC / BIOS for the ultimate configuration control
- White glove service and support. We are here to ensure your success and that you make the most with your time with us
- Realtime communication with our team (Slack or Discord)
Here is the math: $4.50 * 8 GPUs * 24 hours * 7 days = $6,048 (plus tax, if applicable)
Integration with ShadeForm
We offer per-GPU, per-hour pricing through our excellent partnership with ShadeForm, enabling flexible access to our hardware for any duration.
Currently, you can choose between two options: a single GPU in a pre-configured Docker container with ROCm, or full bare-metal access with 8 GPUs.
Both options are priced the same as listed above and can be negotiated down as well. With ShadeForm's managed solution, you can get started immediately via credit card, without needing to contact us directly. They also offer an easy to use API so that you can integrate us into your CI/CD workflow.
Please visit their site for more information and choose their MI300x option:
Why are you more expensive than other providers?
Great question. We recognize that GPU compute is a very competitive market. However, we’re taking a different approach to GPU/hr pricing, focusing on long-term sustainability rather than a “race to zero” that forces providers to cut corners.
Much like how Uber and Airbnb started cheap thanks to VC funding, but eventually adjusted for reality. We’re skipping that unsustainable phase and building a robust business model from the start that our customers can rely on. We’d like to think of ourselves as providing a premium niche based service that stands above the rest with additional value.
Here’s a few reasons why our pricing reflects the value we deliver
- More Memory for Bigger Models: Our GPUs offer 2.4x the memory of an H100, making them ideal for cutting-edge models like the 405B LLama3 that can’t even fit into a system full of H100s. With H100s, you’d need double the compute, doubling your costs. Models aren’t getting smaller, so our infrastructure is ready for today’s needs and the future.
- Top-Tier Data Center: We’re hosted in a hyperscale-class data center powered by 100% green energy (pictured above). This is the same secure, high-quality infrastructure where CoreWeave operates hundreds of megawatts of compute. As such, powering and cooling our equipment properly, which is critical to reliability and performance, isn’t an issue. Cheap GPU providers don’t offer this level of reliability and sustainability.
- Enterprise-Grade Hardware: We use Dell servers with 3-year next business day ProSupport warranties, ensuring our uptime and reliability. This reduces risks for your business while leveraging our strong partnership with Dell. Cheap GPU providers don’t have warranties on their equipment.
- Redundant, High-Performance Equipment: Our racks feature 6 Vertiv PDUs, meaning a single PDU failure doesn’t impact the rest of the rack. Cheap GPU providers have one, maybe two, PDU’s in a rack.
- White-Glove Support: You’re not just another nameless customer. You get multiple layers of expert support: first from us (Hot Aisle), followed by AMD, Dell, Broadcom, and a robust community of specialists.
All these factors ensure reliability, scalability, and performance, more than justifying the price difference.