↓Skip to main content

A100 GPU Cloud Availability and Pricing

Table of Contents

Availability #

  • FluidStack - 1 instance, max up to 25 GPUs on our account - instant access
    • Max A100s avail: 2,500 GPUs (min 1 GPU)
    • Pre-approval requirements: fill out a web form
    • Pricing: $1.10 per/GPU per/Hour
  • Runpod - 1 instance - instant access
    • Max A100s avail instantly: 8 GPUs
    • Pre-approval requirements for getting more than 8x A100s: open a web chat and request a spending limit increase
      • Some information requested:
        • Which model are you training?
        • GPU SKU
        • Number of servers (instances) required.
        • Pricing: $1.79 per/GPU per/Hour
  • Vast.ai - 1 instance - instant access
    • Max A100s avail: depends on the day, I could see a total of 3x 80GB PCIe cards available with 98%+ host reliability and less than $2/hr when I looked (min 1 GPU)
    • Pre-approval requirements: none.
    • Pricing: varies, I see 1x (80GB PCIe) for $1.4/hr, but host reliability varies
  • Lambda Labs - up to 8x A100s instant access
    • Max A100s avail: Unlimited (min 1 GPU)
      • As of June 16 Lambda has 1x A100 40 GBs available, no 1x A100 80 GBs available, some 8x A100 80 GBs available.
    • Pre-approval requirements: Unknown, didn’t do the pre-approval.
    • Pricing: $1.10 per/GPU per/Hour
  • Google Cloud - unlimited access after pre-approval.
    • Max A100s avail: Unlimited (min 1 GPU) (thanks garciasn)
    • Pre-approval requirements: Work with a sales rep to create a custom solution.
      • Some Information requested:
        • How many total instances (VM’s)?
          • What OS will you run?
          • How many cores/processors/vCPUs?
          • How much persistent disk space? (HDD/SSD)
          • Boot Disk Space:
          • Use case of the VM’s for the project
          • Company Domain
          • How soon are you going to have these VM’s?
          • How are you managing it currently (Current infrastructure)?
    • Pricing: $3.93 per/GPU per/Hour. (Edit: Added pricing.)
  • Oracle Cloud - unlimited access after pre-approval.
    • Max A100s avail: Unlimited
    • Pre-approval requirements: Unknown, didn’t do the pre-approval.
    • Pricing: Unknown.
  • AWS - 20 instances after pre-approval.
    • Max A100s avail: 160 GPUs (min 8 GPUs)
    • Pre-approval requirements: Submit request via Service Quota Console
      • For adjustable quotas, you can request a quota increase.
      • Smaller increases are automatically approved.
      • Larger increase requests will take time to review, process, approve, and deploy.
      • You can track your request case in the AWS Support console.
      • AWS Support can approve, deny, or partially approve your requests.
    • Pricing: $4.10 per/GPU per/Hour
  • Azure - limited access after pre-approval.
    • Max A100s avail: Unknown, didn’t do the pre-approval. (min 1 GPU)
    • Pre-approval requirements: Unknown.
    • Pricing: $3.67 per/GPU per/Hour for 1x GPU, pricing varies by region
  • CoreWeave - limited access after pre-approval
    • Max A100s avail: Unknown (min 1 GPU)
    • Pre-approval requirements: contact sales department
      • Some information requested:
        • Which model are you training?
        • GPU SKU
        • Cluster Size
        • Timeline
    • Pricing: $2.06-$2.21 per/GPU per/Hour depending on the SKU
  • Crusoe Cloud - Unknown - site access available by waitlist.
    • Max A100s avail: Unknown (min 8 GPUs)
    • Pre-approval requirements: Unknown
    • Pricing: $1.65 per/GPU per/Hour

Pricing #

  1. Lambda Labs - $1.10 per A100 per hour
    • Max A100s available with Lambda Labs = “Unlimited”
    • Hoops needed to get access to A100s on Lambda Labs:
      • Any A100s = no hoops, just need a free account
      • Max A100s = Work with an Account Executive
  2. Crusoe Cloud - $1.65 per A100 per hour
    • Max A100s available with Crusoe Cloud = Unknown
    • Hoops needed to get access to A100s on Crusoe Cloud:
      • Any A100s = join waitlist
      • Max A100s = join waitlist
  3. Vast.ai - approx $1.40 per A100 per hour
    • Max A100s available with Vast.ai = varies on the day, I saw 3x
    • Hoops needed to get access to A100s on Vast.ai:
      • Any A100s = no hoops, just need a free account
      • Max A100s = no hoops, just need a free account
  4. Runpod - $1.79 per A100 per hour
    • Max A100s available with Runpod instantly = 8 GPUs
    • Hoops needed to get access to A100s on Runpod:
      • Any A100s = no hoops, just need a free account
      • Max A100s = contact sales department
  5. FluidStack - $1.49 per A100 per hour
    • Max A100s available with FluidStack = 2,500 GPUs
    • Hoops needed to get access to A100s on FluidStack:
      • Any A100s = no hoops, just need a free account
      • Max A100s = fill out request form on website
  6. CoreWeave - $2.33 per A100 per hour
    • Max A100s available with CoreWeave = Seems like 8
    • Hoops needed to get access to A100s on CoreWeave:
      • Any A100s = get account approved, can take 3 business days.
      • Max A100s = contact sales department
  7. Azure - $3.40 per A100 per hour
    • Max A100s available with Azure = Unknown
    • Hoops needed to get access to A100s on Azure:
      • Any A100s = contact sales department
      • Max A100s = contact sales department
  8. AWS - $4.10 per A100 per hour
    • Max A100s available with AWS = 160 GPUs
    • Hoops needed to get access to A100s on AWS:
      • Any A100s = fill out request form on website
      • Max A100s = fill out request form on website

The best provider if you need 100+ A100s and want minimal costs is likely: Lambda Labs or FluidStack.

↑