Q: What does Adapteva do?
Adapteva designs and builds energy efficient manycore accelerator chips and computer modules for applications with extreme energy efficiency requirements such as real time image classification, machine learning, autonomous navigation, and software defined radio.
Q: Is the technology for real?
Adapteva has shipped silicon products since May 2011. We are on our firth generation of silicon and have validated the architecture on a wide range of applications. The technology has been validated by over 100 publications to date and has shipped to over 10,000 customers.
Q: How much better is Adapteva than the competition?
Independent studies have shown that Epiphany demonstrates a 25X advantage in energy efficiency compared to general purpose CPUs, while supporting standard familiar programming models like ANSI-C, C++, OpenMP, and MPI.
Q: If Adapteva has a 25X advantage, why isn’t the technology everywhere?
Introducing a new general purpose programmable architecture to the market takes 10-20 years. Building the hardware is the easy part compared to the task of convincing the world to rewrite software that has been built with billions (or trillions!) of dollars of total investment. For most applications, CPU efficiency is “good enough” and software compatibility is a hard requirement. In applications where the capability of the system is completely constrained by size, weight, and power of the system, Adapteva’s 25X efficiency advantage unlocks disruptive capabilities.
Q: What is Epiphany?
The Epiphany is a scalable distributed memory architecture, featuring thousands of processors on a single chip connected through a sophisticated high-bandwidth Network-On-Chip.
Q: What is Parallella?
Parallella is a $99 credit card sized computer launched on Kickstarter in 2012 and made generally available in 2014. The board has shipped to over 10,000 customers to date. You can now buy it at Amazon, Digikey, and RS Components.
Q: How do you program the Epiphany?
Thanks to the Parallella community, you can now program Epiphany using ANSI-C/C++, OpenMP, MPI, OpenCL, Python, Erlang, and BSP.
Q; What are the killer applications?
The Epiphany is a great architecture for most compute intensive applications, but two in particular have stood out: Software Defined Radio, Real Time Image Processing, and Machine Learning
Q: What is the big deal, GPU’s have 1000’s of “cores”?
There is no agreed upon definition of “core” so hardware vendors have taken the liberty of calling anything from a floating point unit to a full fledged out of order CPU a “core”. The Epiphany CPU “core” is an openly documented dual issue RISC floating point core. Comparisons between different processor platforms should NOT be made by comparing vendor marketing material.
Q: Why does energy efficiency matter?
In any portable device battery capacity is fixed and limited by the acceptable size and weight of the attached battery. The more power you need for processing, the faster the battery runs out. Mobile devices need to have a certain minimal on time to be practical, so in effect the energy efficiency of the processing limits the types of applications that are practical to implement on a given device.
Q: Hasn’t this been done before?
Multicore and multiprocessor architectures have been around almost as long as there have been computers. However, Adapteva is unique in being the first company to offer a general purpose C programming model, native floating point capability, and great energy efficiency in an architecture that scales up to thousands of cores on a single die.
Q: What is Adapteva’s “secret sauce”?
Adapteva has taken a clean-slate approach to computer design and designed each sub-component specifically for massively parallel processing and for low power embedded computing. Another key to our success has been the very tightly coupled optimization flow that considered all aspects of the chip, including performance, ease of use, power, ease of implementation, engineering development costs, and EDA tool restrictions, etc. We went through over 50 iterations of this architecture in less than 6 months before arriving at the final version.
Q: What does Adapteva’s technology mean to the end user markets?
Adapteva’s Epiphany IP will be a give an immediate 25x performance boost to mobile devices without imposing radical changes to the SOC development process or system programming model. The industry has been putting accelerators on SOCs for two decades, the major difference is that the Epiphany accelerator is programmable in standard languages and supports floating point programs out of the box. The Epiphany IP efficiency edge is equivalent to a 3-5 node process technology leap. As device scaling will soon come to an end, this means standard CPU’s will NEVER be able to catch up to Adapteva’s Epiphany architecture.
Q: Will Adapteva technology displace the incumbent microprocessors?
No, Adapteva Epiphany is not a replacement of existing microprocessor technology. Traditional CPUs are better suited for running traditional operating systems like Windows and Linux. Epiphany is best leveraged as a massively parallel autonomous application specific processor.
Q: How are programmers going to be impacted by this technology?
Adapteva’s architecture is truly ANSI-C programmable and has a powerful optimizing compiler, so an engineer’s existing legacy code or favorite open source library will execute correctly on Adapteva architecture from day one without modification. Once the application is up and running correctly the programmer can start optimizing the application for performance. Using multiple cores to create a fast implementation is done through a robust and well understood task-channel programming model.
Q: What are the current standards for this type of multicore processing?
Epiphany enables placing 1,000 general purpose floating point RISC processors on a single chip. This is an order of magnitude (10x) greater than any previously announced general purpose multicore device.
Q: Where does the name Adapteva come from?
It is a combination of the Hebrew word “Teva”, meaning nature and adaptable. It signifies that our technology and our company is by nature able and willing to change based on the environment.