Enquire nowLog in

Pricing - GPU Clusters, Storage & Networking

Breakthrough energy efficiency provides uniform cost effectiveness for GPU clusters across all our cloud regions.

SMC infrastructure uses up to 50% less power, alongside a denser, more efficient hosted environment. Ultimate energy-efficiency drives down the cost to train and inference AI workloads.

No tricks, just good engineering.

Managed metal GPU cloud pricing

Pricing for cloud based managed metal GPU instances. Pricing shown is for month-by-month consumption, in USD. SMC provides further discounts based on committed use, volume and level of prepayment. Reach out to our sales team to learn more.

GPU Model
GPUs PER INSTANCE
GPU VRAM
vCPUS PER INSTANCE
RAM PER INSTANCE
LOCAL STORAGE PER INSTANCE
GPU COMPONENT COST PER HOUR
NVIDIA HGX H100
8
80GB
128
2,048GB
30.72TB
$4.36
NVIDIA HGX A100
8
80GB
256
2,048GB
30.72TB
$1.24
NVIDIA L40S
8
48GB
64
1,024GB
30.40TB
$1.86
  • Storage

    Fast NVMe onboard storage is complimented by ultra-performant parallel file storage from WEKA and an S3 compatible object file store. Each node or instance in an SMC cluster is connected via a dedicated storage fabric at up to 200gbps non-blocking

    • Offering

      Rate

    • WEKA NVMe PFS

      $0.065 per GB / month

    • S3 Object Storage

      $0.025 per GB / month

  • Networking

    RDMA InfiniBand east-west compute fabric comes at up to 3,200 gbps comes standard with SMC clusters, along with a dedicated non-blocking in-band and storage network. For public connectivity, SMC offers flexible ways to access deliver clusters to end users.

    • Offering

      Rate

    • Public IP Address

      $3.50 per IPv4 per month

    • High-capacity direct connect

      Based on scope

    • Cloud cross connect

      Based on scope

    • High-capacity IP Transit

      Based on scope

GPU Cloud Pricing FAQ

  • Do you offer on-demand consumption?

    On-demand or per hour usage is on our roadmap for future release, but is not something that we offer today. Customers can reserve managed metal clusters in monthly increments, over multi-year commitments.

  • How does billing work?

    Approved and on-boarded accounts receive bills for consumption based on contracted capacity.

  • Do you offer custom pricing?

    Yes. Pricing shown on this page are our reference prices for monthly consumption without commitment. Variable discounts are available for committed use over longer terms, with additional incentives provided for accounts that choose to prepay for a portion of their contracted capacity.

  • Is connectivity included?

    A minimum 1gbps/1gbps IP transit is provided with each cluster. For higher capacity ingress and egress, customised connectivity solutions are available, which may incur additional costs, depending on the engineered solution.

    In addition to IP transit, SMC provides a range of options to cross-connect into the public cloud.

  • Tell me more about the 'sustainability discount'

    We use this phrase to describe, in simple terms, that sustainability really does cost less.

    Energy costs, power consumption and hardware efficiency has a material impact on the economics of operating AI infrastructure. SMC's technology-based approach to efficiency means that we have the ability to use substantially less power, and data centre space, that other clouds.

    As a consequence, our operating costs are lower, and we pass these savings onto you.

  • How can I sign up?

    The first step to coming on board is filling out our enquire now form with information about you and your needs. Within a day or two a team member will be in contact with you to arrange a scoping call.

    Once we have an idea of your requirements and fit for our infrastructure, we can arrange a proof-of-concept trial phase of an SMC cluster or nodes as required.

    With commercials and a solution signed off, onboarding commences, subject to the availability of capacity.

    The end to end process - from contact to onboarding - can conclude within 2-3 weeks.

  • Do you have H100 capacity?

    SMC has cloud regions built or under build in Singapore, Australia, India, Thailand, with Europe coming in the second half of 2024.

    All AZs in these regions feature our HyperCube high-efficiency hosting infrastructure.

    Each region is capable of hosting multi-thousand single node clusters within the same network fabric, based on customer requirements.

    H100 & L40S infrastructure is being installed in these regions, however with current market dynamics, demand is high, and so we work closely with customers to capacity plan for current and current and future infrastructure.