One of the
great pleasures in working with technology is the opportunity to meet emerging,
leading edge, highly innovative companies that we can introduce to our clients
and social media followers. Meet Gyrfalcon
Technology, Inc. (GTI)[1]
a company expanding the market for AI-driven solutions. In just two years, they
have carved out a significant market position by delivering extremely dense,
high performance AI inference accelerator chips at low cost.
Positioning GTI and Chips
GTI has a
solid history of successful performance with chips embedded in a large number
and variety of commercial products. These include smart home & office
products, robot vacuums, monitoring devices (baby and pet), defect detection
equipment and AI Data Center solutions.
The typical
chip for AI applications is vendor-customized to provide optimal performance at
a narrowly defined task or application. Such customized chips require
specialized knowledge and close collaboration between the manufacturer and
buyer. The results are long development
cycles, low production runs, large devices and high cost which slows market
growth and frustrates buyers.
Unlike most, GTI’s architecture is a
general-purpose design using software for optimization and development by GTI
or purchaser staff. The result is a high-performance, flexible-use chip at much
lower cost. A proprietary (Matrix Processing Engine™) architecture yields a
chip with extremely high processing speeds, very low power consumption and very
small footprint.
In today’s
market, vendors designing AI-driven products want easier, faster, even transparent
access to AI capabilities. Chips/systems must be more accessible (easier to
use), more powerful (on-chip memory and processing), adaptable
(general-purpose) as well as small enough and cheap enough for large volume
implementation and distribution.
To serve and
expand this market, GTI is licensing[2] its GAINBOARD™ System of proprietary technology and support
items. These include the technology itself (chip, architecture, system),
development and software optimization tools (documentation and product), and (optional)
technical support services. GAINBOARD™ is packaged as 16 chips on a single PCIe
card.
Two GTI chips
are available. The Lightspeeur® 2801 delivering 2.8 TOPS at 300 mW for Edge AI
applications. The Lightspeeur® 2803 delivers 16.8 TOPS at less than a watt for
Cloud AI applications. GTI continues to develop more powerful and efficient
chips.
Licensee can
access a variety of GTI proprietary software, hardware development kits[3], e.g. Raspberry Pi and USB
3.0 dongles that work with Windows and Linux PCs. These allow for model
creation, chip evaluations and Proof-of-Concept designs. Also available is domain
specific software for the customer to optimize and customize the system for
their own use or do so with GTI’s help.
The Final Word
Since launch, GTI products have established a record of successful performance embedded in a wide variety of commercial AI products. They have proven the value and versatility of general-purpose technology. Chips are used in smart home &office products, robot vacuums, monitoring devices (baby and pet), defect detection equipment, situation assessment monitoring and AI Data Center solutions. The hi-volume, general use chip design for use in optimized local devices has proven attractive to hi-volume, distributed AI-driven applications.GTI’s products are well-thought through. Their strategy is makes sense as they add significant value by making easier and cheaper to use AI commercially. This appeals to innovators; it expands the universe of conceivable applications. For those considering an AI-driven solution, product or service, they are well worth a look!
No comments:
Post a Comment