show index hide index
- has chosen to take on established players such as NVIDIA’s Hopper chips. As the need for real-time computing power continues to grow, this company distinguishes itself by its ability to reduce both acquisition costs and energy consumption. In fact, the first generation of the Atlas server
- FPGAs (Field Programmable Gamuts), Positron was able to quickly validate its architecture. Each Archer chip offers 32 GB of HBM2e memory, which is essential for processing large volumes of data. This architecture enables bandwidth utilization of over 93%, far surpassing the performance of GPU-based systems, which often stagnate between 10% and 30%.
- A cost-effective and high-performance alternative
In a world where the demand for AI computing is exploding, a new player is entering the market: the Positron AI Atlas. Promising unprecedented energy efficiency, this server stands out for its ability to dramatically reduce power consumption compared to industry giants like NVIDIA. While proprietary GPUs dominate the landscape, Atlas’s bold proposition challenges the established order and paves the way for an era where resource optimization and performance converge. Get ready to explore this technological revolution that could well redefine the standards of AI and computing infrastructure. The Positron AI Atlas server is poised to be a true disruptor in the artificial intelligence landscape. Promising unparalleled energy efficiency, it challenges giants like NVIDIA while meeting the growing demand for intensive computing. With an architecture specifically designed for inference tasks, this emerging player could well redefine industry standards. A New Era of AI The promise of Positron AI Atlas lies in its energy efficiency, which is 4.5 times greater per watt compared to traditional systems. In a market largely dominated by proprietaryGPUs, this announcement is making waves. As energy consumption becomes crucial due to the explosion in demand for AI applications, Positron is positioning itself as a viable alternative for companies looking to reduce their environmental footprint while increasing their processing power. An Architecture Optimized for Language Models Tea Positron AI Atlas server is designed for specific workloads, particularly those related to language models. Far from the general-purpose approach of GPUs, it emphasizes proprietary acceleration units called Archer.
Each unit features ample HBM memory, enabling fast and efficient data processing. With a total capacity of 256 GB of HBM and 384 GB of DDR5, this system addresses bandwidth requirements that are often problematic in traditional AI architectures. A challenge to established giants Rezidant in the market since 2022, Positron
has chosen to take on established players such as NVIDIA’s Hopper chips. As the need for real-time computing power continues to grow, this company distinguishes itself by its ability to reduce both acquisition costs and energy consumption. In fact, the first generation of the Atlas server
shipped in record time, just 15 months after its initial announcement, thus challenging the much longer development cycles of other industry players. The Foundations of a Promising Technology Archer Accelerators At the heart of Atlas’s technology, the Archer accelerators pave the way for a new approach to AI model execution. Thanks to
FPGAs (Field Programmable Gamuts), Positron was able to quickly validate its architecture. Each Archer chip offers 32 GB of HBM2e memory, which is essential for processing large volumes of data. This architecture enables bandwidth utilization of over 93%, far surpassing the performance of GPU-based systems, which often stagnate between 10% and 30%.
Memory and Storage The sophistication of the Atlas system also relies on a redefined memory hierarchy. HBM memory is specifically dedicated to model weights, ensuring fast access. Meanwhile, DDR5 manages the user context and the KV cache, thus optimizing data flow during execution. The server supports up to 2TB of DDR5 memory, allowing for smooth management of various models. Power ConsumptionWith a total power consumption of 2000 W, the Positron AI Atlas stands out as a strategic choice. In comparison, NVIDIA’s DGX H200 consumes 5900 W, allowing the Atlas to offer significantly higher computing density. This performance is further enhanced by the integration of titanium-grade power supplies, ensuring a power conversion efficiency of 96%. Configurations tailored to market needs The Positron AI Atlas is available in several configurations, offering a flexible solution to meet the diverse needs of businesses. The standard configuration, designed for small and medium-sized enterprises, provides access to advanced infrastructure at a reduced cost, ranging from €80,000 to €100,000. Conversely, the Enterprise version, for large data centers, includes advanced management features at a cost of €120,000 to €150,000. Finally, the Atlas Custom offering allows for complete modularity, adjusting system resources according to specific customer requirements. This model is ideal for demanding sectors such as fintech and scientific research, with prices ranging from €100,000 to €200,000.
A cost-effective and high-performance alternative
Compared to a DGX H200, priced between €300,000 and €400,000, the Atlas Positron AI positions itself not only as an innovative challenger but also as a cost-effective solution. With a performance factor of 4.54 times per watt and 3.08 times per dollar, Atlas is redefining the standards of the AI computing economy, shifting the financial burden towards more sustainable operating expenses. A strategy that disrupts the status quo While NVIDIA and AMD dominate the market, Positron intends to reshape the competitive landscape with its energy efficiency and flexibility. By focusing on the needs of inference rather than resource-intensive training, the company is establishing a disruptive business model. Operating in a market where investments in AI infrastructure are crucial, Positron is undoubtedly attracting the attention of cloud service providers and large enterprises. Future Prospects and Ambitions With its first production racks deployed, Positron appears determined to expand further. Future collaborations with regional hyperscalers could strengthen its market presence, particularly to ensure digital sovereignty. The company’s roadmap also anticipates the arrival of even more optimized systems, meaning this fintech could play a crucial role in the future of AI computing.
