Introducing Linode GPU Instances

June 19, 2019 5:00 pm

We are excited to announce the pilot launch of Linode GPU instances. These instances are tailored towards workloads such as machine learning, artificial intelligence, high quality graphics rendering, video transcoding and scientific computing.

Linode GPU instances are built on NVIDIA Quadro RTX 6000 GPU cards. This is the first time these cards are being offered by a cloud provider, and includes all three major types of processing cores (CUDA, Tensor, and Ray Tracing). It’s the next step in cloud computing for Linode, and advances our mission of making cloud computing simple and accessible to everyone.

All Linode GPU instances come with dedicated CPU cores and your choice of one to four cards per instance.

We’re rolling this out initially as a pilot. There’s limited availability (these things are expensive!) and it’s only available in Newark for now. But most importantly, we want you to help shape our next steps. So, please take a look at what we have, maybe give it shot, and let us know what you think or what your needs are by sending us an email or by opening a ticket.

Ready to start crunching numbers? Check out our GPU Getting Started guide.

9 Responses

  1. Only costs $760/month for a GPU… AKA 1/5th of the GPU’s purchase price.

  2. I’m intrigued by all this GPU stuff, and this is a genuine question.

    For $1000 a month I could get 2x 32 dedicated CPU cores and 64GB RAM Linodes – I.e. 64 cores and 128GB RAM for $960.

    That seems like an insane amount of power. What makes a GPU so special that a normal (high spec) CPU can’t do?

  3. Andy: Great question. GPU processors excel at handling tasks that require parallel processing, especially when compared to CPUs, which are optimized for serial processing. This means that GPU instances are a good fit for use cases such as machine learning, video processing, and data science workloads.

  4. Looks like y’alls marketing department got to your spec sheet before you could publish it.

    40Gbps network in
    10000Mbps network out

    Why not just say 10Gbps network out so that it has the same format as the line directly above it? A bigger number looks better right? 😉

  5. Russell: As recently as just a few years ago, we actually had outbound bandwidth rates of 125 and 250 Mbps on some of our plans, and the notation difference is a remnant of those times. Since then all of our limits have routinely been in multiples of 1000 Mbps, so we’ll look into getting our pricing page updated. Let us know if you have any other suggestions.

  6. This is very expensive. Any plans for less powerful GPUs that can be used by poor ML and AI practitioners?

  7. Thanks for sharing your thoughts with us, Mohamed, it’s very much appreciated. I’ve passed your request for more affordable GPU plan options on to our team for consideration. Let us know if you have any other suggestions here or at

  8. are these being charged even when turned off like regular linodes, or will there be an option to pay for only for the hours the resources are being used, similar to paperspace’s model?

  9. Matt: Yes, billing works the same for GPU instances in that you’re charged even when they’re turned off. To pay only for the hours the resources are used you could remove the Linode when you’re done, and you’d only be billed for the time it was on your account (rounded up to the nearest hour).

Leave a Reply