Q Blocks Case Study
About Startup
This unicorn startup is a giant online e-tailer. Their health tech platform is using emerging technologies like machine learning and AI to edge out competition and scale up the operations beyond top metros in India.
Use Case
Their team leverages a lot of Machine Learning and Big Data solutions for use cases such as developing recommendation engines to add more value to customer experience.
Computing requirements
Machine learning model training process is computationally intensive and thus requires GPU infrastructure to train model algorithms faster on large datasets. GPUs with large memory also helps fit large datasets and thus reduce training time.
More GPU compute also helps perform more experimentation & tuning under the available project timelines. Thus, helping you achieve higher accuracy for your ML models.
Challenge
Setting up a GPU cluster on local premises is time-consuming, unscalable and requires maintenance overhead. Also, it can lead to wastage if the resources are under-utilized.
Renting a GPU cluster on a cloud platform reduces maintenance overhead and wastage. It is also highly scalable. But GPU servers on cloud turn out to be very costly and can surpass the organisation's research budget very quickly. Thus reduces the experimentation capabilities of the entire ML team.
Q Blocks Solution
Q Blocks offered a unique proposition to this startup, enabling on-demand scalable and secure access to GPU servers but at upto 80% lower cost than renting a GPU server on a traditional cloud. Not only that, the GPU servers were pre-installed with compatible GPU drivers, AI frameworks like Pytorch / TensorFlow / Scikit-learn and Jupyter Lab.
Result
- Scalability: On-demand upscale / downscale
- Flexibility: Able to choose from a spectrum of low to high end GPUs.
- Cost: $3/hr on Q Blocks v/s $30,000 for setting up, operating and maintaining a similar GPU infrastructure on-premise
- Total Savings: $200K over a period of 12 months of using Q Blocks v/s a traditional cloud platform.
- Launch Time: Access in 2 minutes. Ready to use & pre-configured Jupyterlab, AI frameworks and GPU drivers.