Nvidia H100s are Available On Demand 🚀

Nvidia H100's VMs are now available to start right away on Paperspace by DigitalOcean. Try it out by going to start an H100 VM. ⚡️

What you’ll have access to:

  • Nvidia H100: Access H100 machines in either an H100x1 (single chip) or H100x8 (entire H100 host) configuration with everything working straight out of the box on the update ML in a Box VM template
  • Faster Networking: By the end of January, all users on H100x8 configurations will have access to 3.2TB interconnect speeds for intense multi-node training runs

On-Demand Access & Guaranteed Access

Start H100 on-demand

Start Now

Guarantee access to H100s
(term discounts & multi-node training available)

Schedule a Call

Other Improvements

Simplified Container Registry Experience

  • Redesign of the container registry experience to provide field validation/checks and enhanced management of existing private container registries
  • New experience in adding containers into Gradient deployments to ensure successful deployment starts

Endpoint Security on Gradient Deployments

  • When creating a deployment, you can choose whether to set the endpoint to public or protected. A protected endpoint is secured with basic access authentication encoded tokens
  • When a user tries to access a protected endpoint they are required to supply a username/password

Improved Cloud Networking

  • Internet access now supports sustained 2Gbps connectivity inbound and outbound per VM


Faster storage speeds on VMs and Shared Drives

Paperspace has been working to provide users with faster storage speeds, and we are excited to announce that NVMe storage is now available in NY2! VM and shared drive creates are now backed by NVMe when capacity is available, and you can expect read/write speeds to improve by as much as 2.5x compared to resources created before March 2023.

Over the last several months, we have also been rapidly expanding the cloud, specifically with Nvidia Ampere Series GPUs (A4000, A6000, A100). You can also join the waitlist to get early access to Nvidia H100 machines. Waitlist

Fixes & Improvements

  • Gradient Deployments now have liveness and readiness health checks, so you can monitor the health of your containers. Learn more here.
  • Fixed several architecture bugs related to shutting down notebooks, which will bring capacity back online faster for users.
  • Deployment optimizations were made to more intelligently add and remove replicas when spec edits are submitted.
  • The Core AMS region now has 10Gb internet uplinks.
  • We updated our payment modal to better accommodate international users.

Gradient launches new Graphcore Machines

In partnership with Graphcore, we are excited to share that we are launching additional types of IPUs in Gradient Notebooks. This means that you can now start training on POD4, POD16 and BOW-POD16 IPUs within the notebook experience for longer than 6 hours. Check out these machine specs!

Graphcore’s Intelligence Processing Unit (IPU) are chips designed from the ground up specifically for machine learning user cases and can deliver training speed improvements.

You can try IPU-POD4’s for free by going to your private workspace to test out this new hardware. Check out this blog on how to get started using IPUs on Gradient. If you want to dive into an example of how to get stable diffusion working on IPUs, give this blog a try.

Fixes & Improvements

  1. Fixed a bug in the deployments list view which prevented users from seeing all deployments
  2. Bug fixes to Core Shared Drive creation
  3. Improved error messaging on payment failures
  4. Improved the loading speeds for the notebook create and billing pages

Compute Limits are now easier to use

We improved Compute Limit management for team administrators.  Compute Limits assist teams in monitoring and restricting monthly compute spend by providing fine-grained control over compute usage.  You can now create and edit email alerts and absolute maxes for both teams and team members from the billing page.  Administrators will receive an email when an alert or max amount is reached, and users will be notified in-app if they are blocked from accruing additional compute.  We hope these changes will give administrators more insight and control over their monthly spend.

For more information about Compute Limits, please visit our documentation.

Fixes & Improvements

  • Added the ability to edit team names
  • Improved invoice clarity by adding a section to represent Gradient storage costs
  • Improved visibility into Core machine status by more accurately rendering real-time machine state
  • Fixed a bug where some users could not scroll on streaming Core machines
  • Fixed a bug where non-US countries were not selectable during signup onboarding

Stable Diffusion on Notebooks

We just published a two part blog on getting stable diffusion up and running with Gradient notebooks using Dreambooth. In part 1, we walk through each of the steps for creating a Dreambooth concept from scratch within a Gradient Notebook, generated novel images from inputted prompts, and showed how to export the concept as a model checkpoint. In part 2, we show how to train textual inversion for Stable Diffusion, and use it to generate samples that accurately represent the features of the training images using control over the prompt.

To make it easier for our free users to take advantage of the platform, we have also released the Stable Diffusion models as Public Datasets. These can be mounted in any Gradient Notebook, and removes the need to download the files from HuggingFace each time you restart the notebook. Furthermore, these files will not count toward the storage limits for Free GPU users, so they will no longer be limited by storage space. Be sure to try out the new process!

Deployments Autoscaling

We’ve added the ability to autoscale your Gradient deployments by adding scaling criteria to the spec document. You can autoscale the deployment based on specific metrics including CPU utilization, memory utilization and # of requests. Documentation on how to get started can be found here.

Additionally, the spec has been updated with an enabled flag both for the deployment as a whole and the autoscaling feature. This can be used to turn the deployment and feature on and off. Previously, you had to go into the spec and change the number of replicas to 0 to turn a deployment off.

Activity Log

To track autoscaling events, deployment updates, and deployment start/stops we also added an activity log which can be viewed from the Activity Log tab in the project view.

Fixes and Improvements

  • Added email notifications for when Gradient Deployment do not properly get provisioned
  • Updated the Nvidia RAPIDS container to RAPIDS 22.10
  • Fixed a few documentation links in the console that were taking users to stale urls
  • Upgraded our database to support enhanced metrics to track network and database performance
  • Upgraded the Nvidia templates to support Nvidia 510 drivers
  • Fixed a bug where deployments would sometimes get deleted when team compute limits were hit
  • Fixed a bug that was preventing private S3 buckets from mounting in Gradient notebooks
  • Improved the container caching process allowing more frequent updates to notebook runtimes

Fixes and Improvements

Over the past few weeks, we've shipped a number of improvements to improve your product experience.

  • Improved private network performance and stability in AMS1: A major upgrade has been completed for the network infrastructure in AM1 resulting in increased performance and stability of the network.
  • Improved reliability of machine starts: Starting machines will now encounter fewer issues due to improved automatic remediation of event errors.
  • Improved availability of GPUs: Resolved an issue where specific machine types GPU availability were lower than expected.
  • Prevent users from accidentally creating machines from deleted custom templates: There was an issue that allowed users to create machines from deleted templates resulting in inaccessible machines. Users will now see an error message if they try to create a machine from a deleted template.
  • Improved management of inaccessible machines: Users can now shut down and deactivate machines that are inaccessible from the list views and details view.
  • All new Linux machines have a dynamic public IP by default: A public IP is currently required to access a Linux machine except on a private network. By default, a newly created machine will have a dynamic public IP assigned.
  • Improved summary of billing charges on machine create: We have improved the machine create billing summary to better reflect the total cost for creating and running a machine. Prices are broken down into monthly and active charges to differentiate the costs of running machines vs having machines active.
  • Improved network performance and reliability: We’ve made a change to our QOS policy that now results in greater network availability and reliability for all machines.

Gradient Models are easier to upload

We’ve enhanced the Gradient model upload process through the console and CLI. You can now track progress of the model upload, bring in nested directories and more gracefully handle upload errors and aborts.

Previously, the model upload dialogue would not show you progress and allowed you to close it before the model would complete. This could cause upload failures mid-process. This now lets you know exactly what is being uploaded!

We have also parallelized large uploads by splitting them into chunks, making better use of bandwidth and drastically reducing the amount of time it takes for an upload to finish.

Lastly, the CLI prevents users from uploading with an old version of the CLI which was previously breaking the ability to get data into Gradient.

Fixes & Improvements

  • Fixed a bug causing the workflows status pagination to limit itself to 2 pages
  • Fixed a bug that did not allow you to rename files in the Gradient Notebook file manager
  • Fixed a bug causing readiness errors in notebooks
  • Updated the Gradient Notebooks Rapids runtime to RAPIDS 20.08
  • Fixed a bug that caused the edit button in Gradient Deployment Specs to be stuck in a perpetual load
  • Added the ability to update certain Notebooks to the newest version of our Gradient container
  • Fixed a bug that allowed users to choose an IPU on the Fast.ai runtime
  • Added two shortcuts to Gradient Notebooks: Markdown (m) and code(y)
  • Improvements to the project page to more easily create notebooks and deployments

Autosave, revamped logs, file manager improvements, and more! 💾

We've added a number of new features to Gradient Notebooks, including autosave, improved access to logs, and a host of file management features.

Let's dive in!

Enable autosave in Gradient Notebooks! improvement 

Autosave is now available in Gradient Notebooks. Enable autosave with the button in the bottom right of the IDE.

When autosave is enabled, your notebook will save automatically every thirty seconds.

New panel in the IDE for logs improvement 

Logs have received a refreshed look and feel! We've moved the logs pane from the sidebar to the main window in the IDE. We think this provides a much more comfortable amount of real estate to scroll through system logs. 

Toggle the log pane using the button in the bottom left corner of the IDE above the IDE host metrics.

New file manager CRUD capabilities are now available offline! improvement 

We've added a number of new CRUD capabilities to the IDE -- and not just that but these capabilities are available when your machine is offline! 

The list of upgrades includes uploading and deleting multiple files and/or folders as well as drag-and-drop capabilities for moving multiple files and/or folders.

In addition, we've added the ability to duplicate files and/or folders.

Bugfixes fix 

  • We fixed a bug that sometimes caused the terminal window pane to resize incorrectly
  • We fixed a bug that prevented access to secrets from within a notebook
  • We fixed an issue with font rendering in the terminal


Paperspace partners with Graphcore to provide IPU-powered notebooks 🔋

We're excited to launch a partnership to bring new machine learning hardware to Paperspace!

Graphcore IPUs now available in Gradient! announcement 

As of today, Gradient Notebooks users can launch IPUs from Graphcore on Paperspace -- for free up to 6 hours!

Graphcore IPUs are specialty accelerated computing chips designed to maximize machine learning workloads. 

We're pleased to offer Graphcore's IPU-POD16 machine with 10GB of free storage. 

We've made it extremely easy to get started. Just head over to the Gradient console in Paperspace, create a new notebook, and select one of the new Graphcore runtimes.

Once in the notebook it's easy to start running code.

We've created three different runtimes to start -- including Hugging Face, PyTorch, and TensorFlow 2.

TRY NOW


For more information be sure to read the announcement

DALL-E Mini and all-new free Ampere GPUs for Growth plan subscribers! 🧑‍🎨

The hype around the internet for DALL-E 2 and DALL-E Mini has been building for weeks -- and now you can train your own DALL-E Mini model on a Gradient Notebook! 

Let's get into the updates.

DALL-E Mini runtime now live on Gradient! announcement 

We're excited to release a new runtime tile for DALL-E Mini. The runtime is based on JAX and makes it easy to create generative art on high-powered Paperspace GPUs.

To get started head over to the console, create a new notebook, select the DALL-E Mini tile and get going!

New ultra-powerful A4000 and A5000 GPUs now FREE on Gradient Growth plan! announcement 

As we continue to offer the best selection of cloud GPUs on the market we also continue to extend our lead in the number of unlimited instances we offer to Gradient subscribers.

We've just added A4000 and A5000 machines to the Gradient Growth plan, which means the list of free GPUs available on Growth is longer than ever.

Check out all the free GPUs available to Gradient subscribers below!


GPU
Price
Architecture
Launch Year
GPU RAM
CPUs
System RAM
Current Street Price (2022)
M4000
Free (Gradient Free-tier)
Maxwell
2015
8 GB
8 vCPU
30 GB
$433
P4000
$8/mo (Gradient Pro)
Pascal
2017
8 GB
8 vCPU
30 GB
$859
P5000
$8/mo (Gradient Pro)
Pascal
2016
16 GB
8 vCPU
30 GB
$1,795
RTX4000
$8/mo (Gradient Pro)
Turing
2018
8 GB
8 vCPU
30 GB
$1,247
RTX5000
$8/mo (Gradient Pro)
Turing
2018
16 GB
8 vCPU
30 GB
$2,649
A4000
$8/mo (Gradient Pro)
Ampere
2021
16 GB
8 vCPU
45 GB
$1,099
A5000
$39/mo (Gradient Growth)
Ampere
2021
24 GB
8 vCPU
45 GB
$2,516
A6000
$39/mo (Gradient Growth)
Ampere
2020
48 GB
8 vCPU
45 GB
$4,599


For more information, be sure to read the docs.


Updated PyTorch, TensorFlow, and RAPIDS runtimes announcement 

We also wanted to let you know that we've rolled out updated Notebook runtimes for PyTorch, TensorFlow, and RAPIDS. 

In the notebook console you'll now find 1-click runtime tiles for PyTorch 1.12, TensorFlow 2.9.1, and RAPIDS 20.6.


Is there another runtime you wish we'd support out of the box? Let us know!

Show Previous EntriesShow Previous Entries