Saturn Cloud Reviews

I'm leveraging a cloud-based platform for competitive machine learning. Tight deadlines and resource-intensive models demand powerful hardware. The cloud provides scalable GPUs and RAM, letting me experiment with cutting-edge architectures without limitations. 

Its collaborative features are perfect for distributed teams, enabling seamless code sharing and analysis. I stay focused on model development, not infrastructure, thanks to the platform's streamlined setup. 

My toolkit – Python, Jupyter Notebooks, and standard data science libraries – works seamlessly in the cloud environment. This ensures a smooth transition from local prototyping to large-scale competition training.

Saturn Cloud has become an indispensable part of my data science and machine learning toolkit. Their Dask cluster support is fantastic, allowing me to set up distributed computing with just a few clicks. 

The ability to scale resources up and down effortlessly ensures my work isn't constrained by hardware and helps me control costs. 

Overall, the platform provides everything I need: performance, ease of use, and a focus on the data science workflow itself. This allows me to stay focused on building solutions, not managing infrastructure.

Their Dask cluster support is a standout feature. Setting up a cluster takes mere clicks, making it incredibly simple to harness distributed computing power. I love the flexibility to scale environments up or down. This lets me optimize for performance during intensive tasks and save costs during less demanding phases. Shared folders are a game-changer for collaboration. 

They provide a centralized space for data, code, and results. Also, I can set up pre-boot scripts, which would allow me to configure the instances with all the libraries I need for the particular task.

My main suggestion for improvement centers on pricing. Introducing a tier modelled after AWS spot instances would be a game-changer. Users could bid on unused compute capacity, potentially leading to significant cost savings during off-peak hours or for less time-critical tasks. 

Spot instances empower users with tighter budgets or fluctuating workloads to strategically leverage lower-cost resources for development, experimentation, and background tasks. This frees up on-demand instances for truly time-sensitive work.

I've used the solution for one year.

Saturn Cloud provides a hosted environment where it's possible to work with various software programming tools (e.g., Jupyter Python notebooks, Julia, R and more). 

The system is containerized and accessible both via Jupyter Notebook web pages and SSH—a feature that Google Colab restricts to PRO subscriptions only. I’m currently working on porting a machine learning project to CPU, which provides image Segmentation via Large Language Models. This project handles both image description, image analysis and image object segmentation. Since this project currently relies on CUDA and my local PC has no Nvidia GPUs, I’ve found the computational resources and ease of use provided by Saturn Cloud to be invaluable.

The project I’m currently working on relies on CUDA, but my local PC does not have any Nvidia GPUs. I’ve found the computational resources and ease of use provided by Saturn Cloud invaluable.

Also, there are many ready-to-use Docker images and a rich documentation portal with useful examples. 

The dashboard for creating a new virtual environment contains almost all the features I needed: environment variable definitions, git repositories cloning directly from the new resources page, and an edit field to define a custom script during the boot process. For this reason, Saturn Cloud.io is a very good solution for creating POCs, training machine learning models, and generally experimenting a bit without worrying about local resources.

The solution is valuable thanks to:

- plenty of computational resources (both GPU, CPU and disk space) 

- a big amount of Docker image recipes 

- SSH connection on free subscriptions On Google Colab, the biggest competitor in this field, this feature works only for PRO subscriptions 

- possibility to personalize the characteristics of the new virtual environment directly from the dashboard page, adding new environment variables 

- installing Python pip or CONDA packages and also system packages 

- definition of a custom script that will be executed during the system boot process

I would like more documentation about edge and advanced use cases. 

The official Docker images are only based on Debian: I would like to find official Docker images also based on other systems like Fedora or SUSE operative systems.

It would be nice to have more hardware category options, like TPU coprocessors or ARM64 CPUs. 

I would like a pricing plan associated with a dedicated serverless platform specifically tailored to machine learning inference. 

It would be nice to create a custom serverless API system using my own custom machine-learning model.

I've used the solution for three months. 

The service is stable, I've never experienced problems.

Right now, I'm using it more for creating a POC and experimenting; I didn't try to scale up the service.

I've never requested customer support.

Positive

I also tried Google Colab. I switched since Colab is a little limited for normal use cases based on LLM (at the moment, disk space is only 10GB), and it restricts SSH access on PRO subscriptions.

The initial setup is easy. It only needed to pay attention to the hardware features (e.g., I need CUDA capabilities in my example, so I chose a T4-XLarge instance with a Nvidia T4 GPU) and install Python or system dependencies. Also, pay attention to the Docker image version: an older project will need an older Docker version

There are a good amount of official Docker images (both from StaturnCloud and third-party providers like Nvidia) but also custom Docker image by other users. I'm also satisfied with the code quality and the stability of their deployed virtual systems.

Right now, I'm using only the free plan. However, I'm evaluating an upgrade to a bigger instance (T4-4XLarge with 16 vCPU and 64GB of RAM).

The free plan makes it a good alternative to more famous products like Google Colab, and the pricing plan is reasonable.

I've used and evaluated Google Colab.

Saturn Cloud provides a good Jupyter system based on Python, Julia, or R. 

We use the solution for development, testing, and experiments. Here are some examples of how we use it:

1) There are some workloads that would be really uncomfortable to test in Visual Studio and require more than Google Collab and Azure Notebooks.

2) Testing that's pretty extensive and that's great to have a sensible offering ready in just a few minutes.

3) Saturn's R environment offering is by far the best cloud offering. Lots of time was saved on building a stable environment just by working with Saturn's offering.

4) Some experiments involving code that would crash other environments were performed in a stable, isolated, and secure way. 

1) It offered an excellent development environment while not touching our production cloud resources.

2) Support costs were saved with an evidently tunned and stable PaaS offering.

3) VM performance is truly the one you select for your environment, so you can use it to improve your code performing benchmarking.

4) Tutorials had concrete and pretty useful ideas on how to use the platform. I strongly suggest taking a look at them!

5) Also integration with SSH is a great feature. It allows pretty interesting developments.

1) Mainly Jupyter Labs Server and Dask Clustering. 

2) A stable Jupyter Lab was great for testing some of its features in a completely isolated way. Some testing on our own resources would have an impact on production or have an impact in case of a failure, which is not the case. In a few minutes, we were able to clean up everything and start again from scratch.

3) GitHub integration was a highly useful feature, excellent for QA analysis, and testing the approved code on a standard environment, without any IaaS development.

I'm anxiously asking (and waiting for) Scala Spark Kernel, also the GPU offering can be improved with test offerings, like a spot instance for heavy testing.

Kubernetes Clustering would be really interesting. We'd like a Spark Cluster as well.

Public Clouds integration and sandbox environments would also be a true game changer. I would use it for cybersecurity forensics, paying one invoice for a test environment and cybersecurity testing/validation.

Crontab jobs would also make this service much more interesting for the corporate market.

I used this service for almost a year and I'm really surprised. It offers high performance, is full of interesting features, and has excellent support and high stability. 

The solution is pretty stable; more than Google Colab or Azure Notebooks.

The solution is scalable. It has a combination of nice CPU, GPU, and Dask offerings to cover almost all scenarios.

They offered, again, what exactly we were looking for: Chat with human analysts offering solutions, and replying on a reasonable time.

Positive

We used Google Colab, Azure Notebooks, R Studio Cloud, and home-made platforms over VMs that were evaluated and found more expensive or unstable.

The setup took just a few minutes and by then everything was ready.

We managed the setup in-house with expertise in Jupyter, Dask, and Python.

I believe, not having exact values to answer, that the cost was reasonably expensive and exceeded expectations.

I'd advise users to start with the cheapest offering and grow with their workload. Saturn is great for benchmarking/testing resource consumption.

We looked at Google Colab, Azure Notebooks, R Studio Cloud, and home-made platforms over VMs were evaluated and found more expensive or unstable.

They are frequently adding new features and this review probably will become obsolete soon (what is great).

I am primarily using Saturn Cloud as a student, primarily for training deep reinforcement learning (RL) agents. My projects usually involve Python 3.10, CUDA, and PyTorch and they typically require heavy computation power like GPUs and multi-core CPUs. 

Saturn Cloud's environment has been able to sufficiently cater to these needs, providing a platform where I can run and manage these resource-intensive tasks easily. The support has been very helpful in clearing up questions, such as creating a custom image.

Before using Saturn Cloud, I was relying on my laptop for running my deep RL models. The training process was time-consuming and it was difficult to multitask. However, with Saturn Cloud, I can offload this heavy computing to the cloud. This not only accelerated my research process but also enabled me to multitask efficiently. 

It didn't take long to see that Saturn Cloud could scale with my needs, providing more resources when required. This scalability is invaluable for a student researcher like myself where tasks can vary widely in computational requirements.

One of the features I appreciate the most about Saturn Cloud is its seamless integration with Jupyter notebooks. It provides an interface that I am familiar with and use extensively. It makes it easy for me to run, track, and debug my RL models. 

The second feature I like is the ability to easily scale up and down the resources (like GPUs and CPUs), providing a flexible and cost-effective solution. 

The third feature that is useful is the availability of pre-configured environments. It saves a lot of time and hassle, especially when working with complex setups involving packages like CUDA and PyTorch.

While Saturn Cloud offers a rich set of features, there are areas where I feel there could be improvements. For instance, the process of setting up custom environments could be more user-friendly. Currently, it is a bit technical and might be daunting for beginners. 

Additionally, providing more detailed and beginner-friendly documentation, especially for advanced features, could greatly enhance the user experience. So far, I was relying on the support for setting up custom images and understanding why things don't work properly.

I've been using the solution for one month.

My uptime has been 100%.

The solution is very scalable; machines can go up to extreme configurations.

Support is very friendly, proficient, and helpful. They typically reply within hours.

Positive

Before using Saturn Cloud, I was primarily running models on my laptop and occasionally using Colab. However, Colab was not feasible with the captchas and deleting the runtimes after a few minutes of being away from the keyboard. 

As my models became more complex and required more computational power, I found that this was not scaling well with my needs. I switched to Saturn Cloud primarily for its superior computational power, scalability, and ease of use. In the beginning, the free 30 hours of computing convinced me to try it.

The custom image was a bit more tricky and I required support. That said, it was solved within the same day.

The initial setup was handled in-house.

I don't have an ROI; this is an academic project I am working on.

In terms of pricing, they should make it as transparent as it is here.

I did not evaluate other options. 

We use Saturn Cloud to perform data analysis on large volumes of data. Saturn fetches and updates data, We can use it for machine learning training and prediction, and perform experimental work on various data using multiple machine learning techniques. In some cases, parallel computation is also required to perform the analysis as quickly as possible.

The environment has eight CPU cores and 64 GB RAM. In some cases, we are using GPU. The development environment includes Python, Scikit Learn, XGBoost, Jupyter Lab, and Jupyter Notebook.

Saturn Cloud provides prebuilt images that allow us to quickly spin up a readymade Python environment, speeding up the setup of the environment required for data science projects. 

Saturn Cloud also provides interactive development environments, such as Jupyter Lab and Jupyter Notebook, enabling faster code editing. Saturn's Cloud environment offers high availability and reliability, improving the efficiency of our development work. 

Saturn Cloud supports GPU as part of the environment, which is essential for many computational tasks in machine learning projects. It also allows us to edit the environment, including the image, before we start the cloud resources. This feature lets us quickly set up the environment without the hassle of moving the data and code to another cloud device. 

The solution supports Jupyter Lab and Jupyter Notebook environments as direct options. We can simultaneously work with Saturn Cloud in both modes, enabling faster interaction and sophisticated development support. 

Saturn Cloud should include prebuilt images for advanced data science packages like LightGBM in the next release. If possible, they should also provide a Kaggle image, which contains the most common Python packages used in machine learning.

I would like the option to check or uncheck the data science subpackages we need in the environment. Usage reporting should be more precise and quantify use in the number of minutes instead of just hours. 

I have used Saturn Cloud for more than a year.

Saturn Cloud provides reliable and available resources most of the time.

The computation and memory can be easily scaled from the current requirement.

Our previous solution lacked fast interactive environments like Jupyter Lab and Jupyter Notebook. 

We also considered Paperspace Gradient and Kaggle.

I rate Saturn Cloud 10 out of 10. 

I use Saturn Cloud to run my machine learning models. 

Since it is super fast and has lots of choices of language support and libraries. 

It also gives storage for my large datasets (images, text corpus..). For example, one project was OCR (computer vision). It can scan an answer sheet and grade it automatically. 

When I need to fine-tune and tweak the models, it was impressive. Saturn Cloud makes it so easy and fast to do so. This is an advantage in the industry when you can do prototype and software iteration quickly.

It allowed me to finish projects in time since I can run computer vision or other models repeatedly in a short time span. 

It would be unable to be as fast as if I use Google. It improves my confidence as a software developer and as a machine learning engineer. Since I got nice results and deployment, my professional network increased and business opportunities come quickly. I get new leads for new projects and eventually, I switched to be a full-time machine learning engineer.

The feature I like the most about Saturn Cloud is that it has lightning-fast CPUs. It even provides you with lots of free computing resources up front. 

If you are able to run models multiple times with different tweaks, you can come up easily with an optimal hyperparameter set. Since fine-tuning is an important part of deployment, prototyping, and iterating quickly would get you a head start against your competitors - no matter whether you are in academia or in industry work. 

We'd like to have the capability for installing more libraries. I can rewrite my code to process images in Sklearn and Marplotlib. However, it would be much nicer if you provide a wider range of libraries. 

The need to give more language support (C++). Some models give different results in different implementations which can be a headache since you don't see the same results for the same hyperparameters.

They should provide a way to host API/web apps that is easy and free. It is very common for data scientists to deploy a web app for their machine learning models these days. Saturn Cloud should have this feature.

I've used the solution for one year.

With the support so far, it would scale well to whatever you need.

Customer service and support were perfect. I just needed to email the issues and they got resolved quite fast and with details. The first time the blocker was removed soon. The second time, I was asked something that was a bit out of range and I still received great help. I was a bit surprised support team could get so technical in general.

Positive

The setup was easy. We just needed to choose the templates or start a new server, with (Python) libraries. You need extra software if you want a normal Linux computer.