Hi everyone -
Very excited to be here and evangelize Lambda’s GPU cloud service for deep learning.
Lambda GPU Cloud provides instant access to cloud NVIDIA GPUs.
@Justinpinkney , @Jordan_Lambda and myself (@mitesh) - we will be happy to answer any questions/comments/queries that anyone here has as best and as fast as we can (you can reply under this topic for any questions). We look forward to learning from your feedback as we work on making Lambda a robust cloud offering.
With Lambda GPU Cloud, you have:
- TensorFlow, JupyterLab, PyTorch, and other popular ML software pre-installed
- persistent storage to save your datasets and other files (in beta - for certain instance types only)
- root access to your instances via SSH
Some of our current limitations are:
1 - No API offering
2 - No instance suspension
3 - Persistent storage in Beta
4 - No fast.ai image yet
Here are some resources that can help with FAQs:
Sign up and FAQ website - GPU Cloud - VMs for Deep Learning | Lambda
Our (just started) documentation portal - Lambda GPU Cloud | Lambda Docs
Other interesting GPU benchmarks and blogs from Lambda - The Lambda Deep Learning Blog
@Justinpinkney is the creator of the viral text-to-Pokemon model (based on Stable Diffusion) and his how to is here:
How to fine tune stable diffusion: how we made the text-to-pokemon model at Lambda
Finally - what can you look forward to on Lambda Cloud in the next few months(part of what we are working on to improve our cloud offering):
- APIs (for programmatic spin up and down of instances)
- Region based persistent storage also rolling out (alpha)
- Instance suspension
- Security group rules (user based control)
- Team based accounts
Currently - Lambda Cloud has NVIDIA A100 40GB, A6000s and V100 GPUs.
Lambda will be launching:
- A100 80GB in November
- H100 80GB in Q1 2023 (sorry for the vagueness - at mercy of GPU supply chain)
We are very very excited to provide the GPUs at competitive rates for on demand cloud services - anywhere from 10-70% cheaper (depending on whether you are using spot or on demand instances on any other public cloud providers). Here is a third party summary of all the cloud providers and their pricing that we have found to be accurate:
SIGN UP CREDIT - For availing $150 sign up credit for Lambda - please email email@example.com from the email that you use to sign up for Lambda cloud with the email Subject:
Please note -
- You will have to sign up on Lambda cloud here first for us to be able to apply the credits: GPU Cloud Sign Up | Lambda
- for internationals - our firewall and authentication checks might prevent you from creating an account - due to credit card authentication not going through. Please still email as per above (email from your account email to firstname.lastname@example.org with the Subject: “fast.ai credit” and we will make sure to authenticate your account and apply the credits).
- If you already use Lambda cloud - please still email as above and we will apply the credit to your account.
Looking forward to a great part 2 from Jeremy and learning from everyone’s feedback.
Lambda Cloud Team