QuantConnect Cloud GPU nodes are now out of beta and supported across our entire research stack! Using our new GPU nodes, you can spin up research notebooks attached to our library of financial data, run point-in-time GPU-enabled backtests for accurate simulations and optimizations, and then perform high-demand inference and training in a live environment on low-latency data.

The new nodes are available in the following configurations:

R4-16-GPU$400Jupyter Research Notebook
B4-16-GPU$400Backtesting
L8-16-GPU$400Co-located Live Environment

Depending on your strategy setup, machine learning strategies can experience acceleration of more than 100x with the GPU configurations. In the coming months, we'll explore ways to make these hosts available hourly to make them more accessible to the community.

We recommend combining these GPU hosts with our improved object store to store trained models and access them across the QuantConnect ecosystem with lower latency and higher redundancy.

Low Latency Tickerplants
In addition to the new GPU servers, we installed new top-of-the-line AMD EPYC 5GHz ticker plant servers and network routing hardware to stream data to your strategies with the lowest latency possible. Every day, we process and distribute hundreds of terabytes of data within milliseconds for users in our live environment. We now process every tick of data across 12 markets, build bars your strategies need, and deliver it to your Python strategies, with the vast majority of ticks being available within 10ms. By self-hosting or using competing public clouds you will likely experience delays of 100-500ms.

Expanded Live History
We expanded the capacity of our history requests to support the growing alternative data collection on the QC platform and improved its redundancy. We tested to ensure robust failover of the live historical data requests in the event of failover.

You can migrate to this new environment anytime by stopping and restarting your live trading strategy! With QuantConnect you can spend your time searching and quickly deploying alpha, leaving the infrastructure to our team. 

Happy Coding!