PCIe accelerator card cuts power consumption in AI servers
At CES 2024, DEEPX is introducing the DX-H1 cutting-edge PCIe card designed to accelerate AI server performance while reducing energy consumption.
Honored with the CES 2024 Innovation Award in the Computer Hardware and Components category, DX-H1 prioritizes performance, power efficiency, and cost-effectiveness. The PCIe accelerator card offers exceptional power-to-performance efficiency that surpasses current GPU-based offerings by more than 10 times, enabling it to play a significant role in reducing carbon emissions for AI servers and data centers.
The AI server market relies on power-intensive GPUs, whose high energy consumption leads to substantial carbon emissions. However, the imperative to minimize environmental impact and boost efficiency, specifically in data centers, is driving demand for more sustainable, low-power AI accelerators — a market poised to expand rapidly. As per TrendForce projections, the AI server market is expected to grow at a staggering CAGR of over 36% through 2027. Fueled by both environmental considerations and immense growth potential, cutting-edge low-power designs like the DX-H1 will see surging demand. For data centers seeking to simultaneously curb carbon emissions and gain a competitive edge through top-level AI computing performance, products such as the DX-H1 represent the way forward.
DEEPX leads the market for low-power AI, achieving the world’s highest power-to-performance ratio through proprietary advancements in model efficiency. This success is attributed to DEEPX’s pioneering work in two core technologies:
- IQ8™: an INT8 model compression technology; and
- Smart Memory Access: minimizes D-RAM usage to less than one-tenth of typical GPU levels, particularly beneficial for inference workloads.
By combining IQ8™’s model compression capabilities with memory optimizations, DEEPX can drive excellent performance even with LPDDR memory, instead of using expensive HBM solutions. As a result, the company delivers world-class capabilities — over 10 times higher performance-per-watt than GPUs.
The DX-H1 also seamlessly integrates with existing GPU products, offering a versatile interface to support GPU-based trained AI models. This ensures a hassle-free transition for customers looking to upgrade performance without the need for any systems modifications.
“AI technology’s edge deployment will soon be as ubiquitous as the internet today. As this rapid growth unfolds, DEEPX is set to lead the way, harnessing the surging demand for low-power, high-performance AI semiconductor solutions while already capable of delivering efficiency gains of over ten times,” said Lokwon Kim, CEO of DEEPX.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
