neurothink™ brings the new era of
discovery in machine learning.
A low friction process focused on making robust models
Our own dedicated industry leading GPU servers providing efficient and low-cost compute
End-to-end API-based security delivered on our own secure hardware
Partners that have unique IP to deliver ML models to the edge
Creating a community of ML enthusiasts with a platform for networking and knowledge sharing
Serious Machine Learning compute power
CY2021 statistics for neurothink™
- Available GPU peak capability 489 teraflops FP32 single precision, and 3.9 petaflops (10^15) mixed-precision and deep learning
- Greater than 170K CUDA cores
- Available RAM 12.512 TB
- All flash on-site storage array greater than 250TB
- GPUs available A100, V100 and T4
- Co-location facility, Energy Star rated, Tier III uptime certified, redundant power, and active cooled cages
How do we do it?
- Dedicated GPU compute resource (Tesla, Volta and Amper generation) in a modern tier III certified colocation facility.
- Latest server virtualization, container monitoring and GPU management software from NVIDIA and VMWare.
- Machine Learning instances are customized at the CUDA level for maximum GPU efficiency.
- A user interface that does not require switching between confusing environments.
- Work in a notebook and have access to a command line interface.
- Integrated storage environment to upload, attach, save and push the
objects you need to do your work.
- Resume work on your development instance with auto-save.
Steps in the model development process
- Work starts in a secure and containerized instance. We will guide you as much as you need or give you editing power during any of the steps.
- Analyze the results, rerun the instances, test the results, or compare many instances at once to create a model.
- Complete a risk review, audit your models, and be able to explain and create documentation.
- Prepare your model for a variety of edge applications for various compute devices.
Installed Libraries & Frameworks
+ additional libraries supported thru PIP install in your container