MemryX, the semiconductor startup makes MX3 an AI Accelerator that rivals established offerings. MemryX designs AI processing solutions for edge devices, and in August started its customer sampling of the MX3 AI Accelerator.
The intent here is to produce an ease-of-use and efficiency accelerator. Able to achieve optimized model performance within minutes of installation. MemryX is hoping MX3 shaves off months of painstaking software development from companies seeking to accelerate their edge device compute times.
The MX3 is built for older and newer edge computing systems and can accelerate AI processing in most systems as it connects directly to any device’s processor.
MemryX CEO Keith Kressin said: “Reaching the milestone of customer sampling is gratifying, but what’s even better is the feedback from initial customer experience with the MX3. Customers are seeing first-hand how our solution is different from other AI accelerators.”
Co-founded in 2019 by Kressin and CTO Wei Lu, an electrical engineer and computer science professor at the University of Michigan.
The accelerator’s at-memory compute and dataflow architecture offer low latency, high performance and low power consumption. The architecture was designed to give customers the ability to seamlessly scale performance, power, and latency according to requirements set by developers in their applications.
The MX3 AI Accelerator is deployed through a simple M.2 slot, or PCIe slot. Scalable to any size computer system the modules come in three sizes for M.2 and two sizes for PCIe. It is compatible with x86, ARM, or RISC-V instruction set architectures and can be mounted in any series the customer likes.
Each MX3 chip offers
- >5 TFLOPs
- 4/8/16 bit weights
- 1 = Batch
- Activations: Bfloat16
- Parameters on-die ~10M
- ~1.0W average power
- USB 2.0/3.x
- PCIe Gen 3 I/O
- MX-SDK with a 1-click compilation
More information on MemryX products can be found on the website.
MemryX | memryx.comSponsor this Article
For the past 8 years, I have been writing about embedded technologies, added to my technical, academic, and medical editorial experience, with companies like Elsevier and Cambridge University Press. I tell people to read what I write, not try to pronounce my last name. I am always available for comments and suggestions you can reach me at firstname.lastname@example.org and I promise I will take the time to reach back out to you. I live in the North East with my wonderful family.