InferX offers 1/10th the performance of the fastest FPGA or GPU in an SoC.th Reduce power consumption and cost while maintaining full reconfigurability
mountain view, california, April 24, 2023 /PRNewswire/ — Flex Logix® Technologies, Inc., a leading innovator of DSP and AI inference IP and a leading supplier of eFPGA IP, today announced the availability of InferX.TMs IP and software for DSP and AI inference. InferX joins EFLX® eFPGA as his second IP offering for Flex Logix. Device makers and system companies looking for DSP-FPGA or AI-GPU performance in their SoCs at a fraction of the cost and power. The company’s EFLX eFPGA product line has designs from 180nm to 7nm and he has already proven himself in a number of chips under development at 5nm.
Flex Logix InferX IP for AI Inference provides super-resolution in a single chip at a fraction of the cost and power consumption.
“By integrating InferX into the SoC, our customers not only retain the performance and programmability of expensive, power-hungry FPGAs or GPUs, but also benefit from significant power and cost savings. increase.” Jeff Tate, Founder and CEO of Flex Logix. “This is for system customers who are designing their own ASICs, or chip companies that traditionally have DSP-FPGAs or AI-GPUs next to their chips and can integrate them to generate revenue and save costs. A big advantage for us: InferX is 80% hardwired, but 100% reconfigurable.”
The end-user benefit is a smaller system with more powerful DSP and AI, reducing power consumption and cost. With InferX AI, users can process megapixel images using much more accurate models such as Yolov5s6 and Yolov5L6 and detect images at smaller sizes/longer distances than they are currently affordable.
Advantages of InferX
InferX DSP is InferX hardware combined with Softlogic for DSP operations and Flex Logix provides operations such as FFTs that can be switched between sizes on-the-fly (e.g. 1K To 4K To 2K); FIR filter with any number of taps. Complex matrix inversion 16×16 or 32×32 or other size. etc. The InferX DSP streams gigasamples per second, can perform multiple DSP operations, and can chain DSP operations. The DSP runs on Real/Complex INT16 with 32-bit accumulation for very high precision. With InferX DSP, you can integrate 1/10th the DSP performance of leading FPGAs or better.th Reduce cost and power while retaining all the flexibility of near-instantaneous reconfiguration.As an example, the InferX DSP with less than 50 square millimeters of silicon for N5 performs complex INT16 FFTs at 68 Gigasamples per second, scaling FFT sizes from 256 to 8K point. This is faster than the best his FPGA currently available at a fraction of the cost, power and size.
InferX AI is InferX hardware combined with an inference compiler for AI inference. Inference Compiler ingests customer neural network models in Pytorch, Onnx, or TFLite formats, quantizes models with high precision, compiles graphs for increased utilization, and generates runtime code that runs on InferX hardware. increase. A simple, easy-to-use API is provided to control the InferX IP. With InferX AI, you can integrate AI inference performance equal to or better than state-of-the-art AI modules at 1/10.th Reduce cost and power while retaining all the flexibility and the ability to run multiple models or change models on the fly. InferX AI is optimized for megapixel batch=1 operations and an inference compiler is available for evaluation. For example, with about 15 square millimeters of silicon for the N7, InferX AI can run his Yolov5 at 175 inferences/second. This is 40% faster than the fastest edge AI module, the Orin AGX 60W.
The InferX technology is 16nm proven, production qualified and will be available on the most popular FinFet nodes.
InferX hardware is also scalable. Its building blocks are compute tiles that can be arranged to increase throughput. For example, a 4-tile array is 4x the performance of a 1-tile array. With the performance you demand, InferX arrays come with an AXI bus interface for easy integration into your SoC.
For general information on the InferX and EFLX product lines, visit our website at this link. For more information under the NDA, eligible customers can contact us at this link.
About Flex Logix
Flex Logix is a reconfigurable computing company that provides cutting-edge eFPGA and AI inference technology to semiconductor and systems companies. The Flex Logix eFPGA allows a volume FPGA user to integrate her FPGA into a companion SoC. The result is a 5- to 10-fold reduction in FPGA cost and power, and increased compute density critical for communications, networking, data centers, microcontrollers, and more. Its scalable AI inference is the most efficient, with much higher inference throughput per square millimeter and watt. Flex Logix supports process nodes from 180nm to 7nm, with 5nm under development. It can support other nodes out of the box. Flex Logix is headquartered in mountain view, california has an office in Austin, TexasFor more information, please visit https://flex-logix.com..
media contacts
Kelly Carr
tanis communication
[email protected]
+408-718-9350
Copyright 2023. All rights reserved. Flex Logix and EFLX are registered trademarks and INFERX is a trademark of Flex Logix, Inc.
Source Flex Logix Technologies, Inc.
