Autonomous, a US company, has unveiled the first of its kind, the Brainy workstation, a very high-performance workstation designed to accelerate artificial intelligence (AI) development. The device is designed for researchers, developers, and startups who want to go beyond the cloud and run the most complex AI models locally, which means they can do it quickly and securely.
Image source: Autonomous
Brainy is built for the most complex AI tasks and is equipped with Nvidia RTX 4090 graphics cards. Its performance exceeds one petaflop in the maximum configuration, which can include two to eight GPUs. This allows you to train models with 70 billion parameters, achieve supercomputer-class results right from the desktop, and optimize the training of AI models. As TechPowerUp notes, the station is especially effective for fine-tuning large language models (LLM), computer vision tasks, and other areas of deep learning.
Autonomous developed Brainy to solve the problems of expensive cloud GPU solutions. While cloud services are known to be highly flexible, their pay-as-you-go nature can become prohibitively expensive for long-term projects. Autonomous offers an alternative in the form of on-premises power, increased data security, and significant savings. The company estimates that Brainy owners can save thousands of dollars in the first year compared to renting cloud power.
Another advantage of Brainy is its stability. The system eliminates typical cloud problems such as queues, sudden shutdowns, and Internet delays. In addition, users can start with a configuration of two video cards and expand it to eight if necessary. Finished projects can be easily transferred to the cloud or data center without additional settings.
The manufacturer notes that Brainy supports popular frameworks, including PyTorch and TensorFlow, as well as Nvidia technologies – CUDA, cuDNN and TensorRT. All these capabilities accelerate the development of AI models and scientific computing. In addition, Autonomous is part of the Nvidia Inception program, which gives it access to resources and training to optimize Brainy, and also allows it to offer customers competitive prices. According to the company, with the help of the station, startups and organizations will be able to more quickly develop AI solutions in such areas as medicine, education, logistics and finance.
In addition to Nvidia graphics cards, Autonomous Brainy systems offer AMD Ryzen Threadripper processors – in the basic model, this is a 24-core Ryzen Threadripper 3960X, while more advanced configurations offer the latest 16-core Ryzen Threadripper 5955WX. And the most powerful version is equipped with a 16-core AMD Epyc 9124 chip. The amount of RAM is 96, 128 and 192 GB, respectively.
Note that in one of the configurations, the video cards will not be able to work at full capacity. In particular, the Ryzen Threadripper 5955WX has only 80 PCIe lanes, and six video cards require 96 lanes. And you also need to allocate four lanes for SSD. So even the flagship model with 128 PCIe 4.0 lanes will have a shortage of them. Cooling is also a concern: will the video cards be cramped in the case?
In terms of cost, Autonomous Brainy systems are priced between $9,000 and $40,000.
