UnifabriX Blasts Out of Stealth Mode
The company’s CXL-based Smart Memory Node™ is the first to live demonstrate in SC22 HPC performance acceleration with memory pooling, sharing and CXL™ 3.0 fabric, using industry standard benchmarking. The Smart Memory Node powered by UnifabriX RPU™ promises to unlock significant performance gains in both CPUs and memory that will upend HPC standards.
Haifa – November 7th, 2022 – UnifabriX, maker of the first performance-focused CXL™ system, is announcing today that it is exiting stealth mode to reveal the full scope of its unique CXL based product. UnifabriX Smart Memory Node offers data centers a lifeline for overcoming the processor and memory innovation gap, to unlock significant performance gains and utilize far more CPU cores. With high-performance memory pooling and sharing coupled with extra memory bandwidth, UnifabriX is bringing the HPC industry an entirely novel way to structure data centers and solve performance challenges and bottlenecks.
As the first company in the CXL arena to focus on performance, UnifabriX is introducing its Resource Processing Unit™ (RPU), which evolves the Data Processing Unit (DPU) to unleash the full power of CXL and PCIe, to improve the utilization of the CPU and increase system-wide bandwidth. UnifabriX fully-developed prototype is set to be the first to report on performance with standardized industry-used benchmarking tools. With its RPU, UnifabriX enables HPC and data centers players to move beyond the limits of traditional physical memory, bringing about the next generation of HPC performance. Further enabling this shift is the company’s founding team, which were among the first to work on CXL technology and have the experience necessary to help achieve its full potential.
At SC22, the international conference for HPC, UnifabriX will showcase its Smart Memory Node based on UnifabriX RPU, in the CXL Consortium booth. The demonstration will involve multiple CXL-capable servers connected to two Smart Memory Nodes that are connected through the first ever CXL 3.0 fabric. The CPUs will be able to access local memory, as well as remote memory across. It will also mark the first live demonstration to show enhanced performance related to core utilization, memory capacity, and bandwidth according to a recognized HPC framework. In addition, UnifabriX CTO and Co-founder, Danny Volkind, will also be presenting daily Tech Talks at Intel’s booth.
“Setting out to achieve and document CXL enabled performance in a real environment according to an industry standard HPC benchmarking is a difficult task that we are excited to demonstrate live at SC22. We are setting out to showcase the significant improvements and immediate potential that CXL solutions provide, to upend HPC performance and close the gap between the evolution of processors and memory”, says Ronen Hyatt, CEO and Co-Founder of UnifabriX. “UnifabriX solution goes above and beyond HPC performance demands. UnifabriX is excited to exit stealth mode with a holistic solution that is at the forefront of improving HPC performance.”
“The CXL Consortium is pleased to have several practical technology demonstrations in our booth at SC’22, including the new Smart Memory Node system from member company UnifabriX,” said Siamak Tavallaei, president of CXL™ Consortium. “The direction UnifabriX is taking to significantly increase CPU and memory performance is exciting, to say the least, and we look forward to the impact this will have on the industry.”
UnifabriX develops solutions that address the inefficiencies of large-scale system deployments, enabling the data center operators to fully unlock their infrastructure’s speed, density, and scale. Our CXL-based products achieve exceptional performance and elasticity in bare-metal and virtualized environments over a wide range of applications, including the most demanding HPC, AI, ML and big-data workloads. Founded in 2020, UnifabriX are pioneers in CXL™ and a major contributor to the standard.
For more information on UnifabriX, please visit – https://www.UnifabriX.com