KNUPATH Hermosa-based Commercial Boards Expected in Q1 2017

By John Russell

December 15, 2016

Last June tech start-up KnuEdge emerged from stealth mode to begin spreading the word about its new processor and fabric technology that’s been roughly a decade in the making. It’s nice to have patient capital, a rare commodity for startups these days. The company contends its KNUPATH Hermosa processor with 256 DSP cores and its Lambda fabric will bring performance, scalability, energy, and programmability advantages over CPUs, GPUS, and FPGAs to a wide swath of machine learning applications. The first commercial boards – code named Mavericks – are expected around March this year.

Founded in the 2005 timeframe by Daniel Goldin, the long time NASA administrator, KnuEdge has raised roughly $100M no doubt stemming from investor confidence in Goldin’s extensive technology creation and delivery history. Goldin and company believe their investors’ patience is about to start paying off. KnuEdge has two business units, KNUPATH focused on hardware accelerators based on Hermosa and Lambda technology, and KnuVerse, focused on voice and face recognition systems. The latter, said Steve Cumings, CMO, KnuEdge, has customers in the government sector. Company revenues are somewhat north of $20 million so far.

Broadly, KnuEdge’s view is that a highly scalable processor in a single socket is handicapped in addressing growing machine learning and large-scale computing challenges. In contrast, the company’s Lambda Fabric enables a large number of “KNUPATH Hermosa processors to be interconnected in low latency, high throughput mesh for massively parallel processing which is well suited for application needs that will drive the compute engines of the future.”

This isn’t exactly a new idea. The Hermosa chip and Lambda technology will enter the market amid a gush of machine learning technologies all striving to advance data-driven science and enterprise data analytics. Indeed the emergence of heterogeneous computing architectures relying on a variety of accelerator engines is a key feature of today’s computing landscape. Given Goldin’s remarkable achievements at NASA it should be interesting to watch KnuEdge’s progress.

Early developer boards with two Hermosa chips have been available for some time. Volume sales of individual chips are planned to begin in January followed by the Mavericks offering, a PCIe board with four Hermosa chips, towards the end of the quarter.

Presented as a “neural computing” approach, the KNUPATH architecture actually attempts to mimic nervous system communication more than brain-inspired spiky neuron ‘inference logic’ (discussed further below).

Patrick Patla, senior vice president and general manager of KNUPATH and a former AMD executive, said, “What’s unique about Hermosa’s 256 DSP cores is that they are hooked together at a central part of the processor with a router that has 16 ports. Using the Lambda fabric, it’s possible, at least theoretically, to scale to 500,000 Hermosa processors.

“We are a data flow machine. So you push data through the system and can have the calculation and different algorithms change on the fly. We are different than a GPU accelerator in that they use a SIMD architecture. We use multiple programs, multiple data, so on our 256 cores we could have 256 separate algorithms running. You would push data through those algorithms and then you have hits on the data at different hit rates based on the algorithms and you can tune and resend algorithms to those DSPs through packets,” explained Patla.

“Basically the packets that we send through the Lambda network are what allows the programming of the DSP, so packets deliver the program, the algorithm, and then bring the payload, and push the data through it. Not only are you getting all the data and the operating instructions with each packet, but each core also knows the next destination for that information so it’s extremely efficient.” One result is very low latency at various systems levels (see diagram below).

Patla also contrasted Hermosa’s ease of use with emerging brain-inspired neuromorphic chips such as IBM’s TrueNorth, which uses “spiking neuron” architecture.

“Spiky algorithms are notoriously difficult to program. Commonly they are trained on other networks first and then moved onto the neuromorphic chip so the actual software side of that is different,” he said.

As noted earlier the Hermosa-Lambda architecture emulates neuronal connectivity more than brain processing. “If you look at the different neuron-based approaches, our inspiration really gives you lots of little engines – that’s the background of the DSP cores, what we affectionately sometimes call tDSPs or tiny DSPs,” said Patla. Reliance on familiar DSP architecture eases programming.

“Our tools sit on a C/C++ library set on top of LLVM (compiler). And everybody is familiar with OpenCL as well as OpenMPI which is very comfortable in our architecture,” said Patla. The Hermosa/Lambda architecture also supports NUMA (non uniform memory access) and each processor has memory directly (72MB) on it. “Much of the advantage is the dataflow but also all the advantages of common programming techniques for anybody that has worked on OpenMPI. Many of the other [neuromorphic] architecture require a different set of tools.”

Hermosa Development Board

KnuEdge has had a software developer kit out for “quite some time” and it is already in the hands of many developers, according to Patla.

It all sounds great. In April KnuEdge will hold a Hermosa developers’ conference at UCSD as well as a “heterogeneous neural network conference” in partnership with UCSD for the development of next generation algorithms that can take advantage of new architectures such as Hermosa. Patla said performance benchmarks for chip will be forthcoming with the release of the commercial product; it seems like the developer conference would be a good place to do so, but he wouldn’t specify when beyond the first half of the year.

“Right now, as you would imagine, we are in the labs with our SDKs and final verification of those commercial systems as we are tuning and bringing all of our code to the processors. In the future we’ll show configurations of 4, 8, 12, 16, Hermosas together to show the scalability of the Lambda fabric. When Steve talked about mimicking the nervous systems it really is about our connectivity and the fact that when you add more Hermosas to the network, we continue to scale because with every socket you are adding more memory as well. Each processor has 72MB of onchip memory that is sufficient for the programming of our kinds of algorithms and the workload we are trying to tackle.”

Currently the chip is being fabbed by GLOBALFOUNDARIES on the 32nm process. “It’s a well behaved chip where these 256 cores and fabric and everything lives in a 35-watt part,” said Patla.

The KNUPATH folks believe Hermosa has the potential to meet a wide variety of machine-learning kinds of applications performed in heterogeneous computing environments as well as an opportunity to replace existing approaches to those applications.

‘We have a demo on the website that compares us to the most current NVIDIA card and we have a 2.5x performance. It is very interesting that a video card isn’t very good at video compression that we are good at because of the parallelism of communication we handle across the memory. So that’s one of the spaces we’ll be aiming at. And of course it will also find its way into many of the single board computer spaces because at 35 watts and the ability to do signal processing and such fine grained computing we actually expect it to replace many FPGAs in a lot of environments.”

Patla argues Hermosa/Lambda’s flexibility is a major benefit and door opener – one could divvy the chips up and have a multipurpose SOC instead of dedicating it to just one task. He used a video analysis application as an example of flexibility and reprogrammability.

“You can reprogram a core by just delivering a new packet. For example, if you were doing video analysis and were searching within videos, you could be looking for ball caps. You could have all the different algorithms looking at ball caps and you could just all of a sudden reprogram and divide the chip and have 25 percent of the chip looking for red ball caps and 25 percent looking for blue caps. You could flip to four different algorithms in nanoseconds. Then when you have high hit rates and you realize the one you are really looking for, and you could say OK now all care about our green ball caps and that algorithm would propagate against all the cores and you’d be able to take your throughput up. It’s very fast, very flexible,” he said.

At SC16, the KNUPATH team was busily evangelizing. Patla said they talked to a number of cloud providers as well as national labs that expressed interest to the point that he is expecting some new workloads to emerge.

There’s still much to do. Patla ticked off desirable milestones for 2017 – getting out of the lab, showcasing a couple of commercial customers and workloads, integrating the many machine learning frameworks, making sure Hermosa-based systems get into the cloud somewhere for development and production purposes, to name but a few.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

Data Vortex Users Contemplate the Future of Supercomputing

October 19, 2017

Last month (Sept. 11-12), HPC networking company Data Vortex held its inaugural users group at Pacific Northwest National Laboratory (PNNL) bringing together about 30 participants from industry, government and academia t Read more…

By Tiffany Trader

AI Self-Training Goes Forward at Google DeepMind

October 19, 2017

DeepMind, Google’s AI research organization, announced today in a blog that AlphaGo Zero, the latest evolution of AlphaGo (the first computer program to defeat a Go world champion) trained itself within three days to play Go at a superhuman level (i.e., better than any human) – and to beat the old version of AlphaGo – without leveraging human expertise, data or training. Read more…

By Doug Black

Researchers Scale COSMO Climate Code to 4888 GPUs on Piz Daint

October 17, 2017

Effective global climate simulation, sorely needed to anticipate and cope with global warming, has long been computationally challenging. Two of the major obstacles are the needed resolution and prolonged time to compute Read more…

By John Russell

HPE Extreme Performance Solutions

Transforming Genomic Analytics with HPC-Accelerated Insights

Advancements in the field of genomics are revolutionizing our understanding of human biology, rapidly accelerating the discovery and treatment of genetic diseases, and dramatically improving human health. Read more…

Student Cluster Competition Coverage New Home

October 16, 2017

Hello computer sports fans! This is the first of many (many!) articles covering the world-wide phenomenon of Student Cluster Competitions. Finally, the Student Cluster Competition coverage has come to its natural home: H Read more…

By Dan Olds

Data Vortex Users Contemplate the Future of Supercomputing

October 19, 2017

Last month (Sept. 11-12), HPC networking company Data Vortex held its inaugural users group at Pacific Northwest National Laboratory (PNNL) bringing together ab Read more…

By Tiffany Trader

AI Self-Training Goes Forward at Google DeepMind

October 19, 2017

DeepMind, Google’s AI research organization, announced today in a blog that AlphaGo Zero, the latest evolution of AlphaGo (the first computer program to defeat a Go world champion) trained itself within three days to play Go at a superhuman level (i.e., better than any human) – and to beat the old version of AlphaGo – without leveraging human expertise, data or training. Read more…

By Doug Black

Student Cluster Competition Coverage New Home

October 16, 2017

Hello computer sports fans! This is the first of many (many!) articles covering the world-wide phenomenon of Student Cluster Competitions. Finally, the Student Read more…

By Dan Olds

Intel Delivers 17-Qubit Quantum Chip to European Research Partner

October 10, 2017

On Tuesday, Intel delivered a 17-qubit superconducting test chip to research partner QuTech, the quantum research institute of Delft University of Technology (TU Delft) in the Netherlands. The announcement marks a major milestone in the 10-year, $50-million collaborative relationship with TU Delft and TNO, the Dutch Organization for Applied Research, to accelerate advancements in quantum computing. Read more…

By Tiffany Trader

Fujitsu Tapped to Build 37-Petaflops ABCI System for AIST

October 10, 2017

Fujitsu announced today it will build the long-planned AI Bridging Cloud Infrastructure (ABCI) which is set to become the fastest supercomputer system in Japan Read more…

By John Russell

HPC Chips – A Veritable Smorgasbord?

October 10, 2017

For the first time since AMD's ill-fated launch of Bulldozer the answer to the question, 'Which CPU will be in my next HPC system?' doesn't have to be 'Whichever variety of Intel Xeon E5 they are selling when we procure'. Read more…

By Dairsie Latimer

Delays, Smoke, Records & Markets – A Candid Conversation with Cray CEO Peter Ungaro

October 5, 2017

Earlier this month, Tom Tabor, publisher of HPCwire and I had a very personal conversation with Cray CEO Peter Ungaro. Cray has been on something of a Cinderell Read more…

By Tiffany Trader & Tom Tabor

Intel Debuts Programmable Acceleration Card

October 5, 2017

With a view toward supporting complex, data-intensive applications, such as AI inference, video streaming analytics, database acceleration and genomics, Intel i Read more…

By Doug Black

Reinders: “AVX-512 May Be a Hidden Gem” in Intel Xeon Scalable Processors

June 29, 2017

Imagine if we could use vector processing on something other than just floating point problems.  Today, GPUs and CPUs work tirelessly to accelerate algorithms Read more…

By James Reinders

NERSC Scales Scientific Deep Learning to 15 Petaflops

August 28, 2017

A collaborative effort between Intel, NERSC and Stanford has delivered the first 15-petaflops deep learning software running on HPC platforms and is, according Read more…

By Rob Farber

Oracle Layoffs Reportedly Hit SPARC and Solaris Hard

September 7, 2017

Oracle’s latest layoffs have many wondering if this is the end of the line for the SPARC processor and Solaris OS development. As reported by multiple sources Read more…

By John Russell

US Coalesces Plans for First Exascale Supercomputer: Aurora in 2021

September 27, 2017

At the Advanced Scientific Computing Advisory Committee (ASCAC) meeting, in Arlington, Va., yesterday (Sept. 26), it was revealed that the "Aurora" supercompute Read more…

By Tiffany Trader

How ‘Knights Mill’ Gets Its Deep Learning Flops

June 22, 2017

Intel, the subject of much speculation regarding the delayed, rewritten or potentially canceled “Aurora” contract (the Argonne Lab part of the CORAL “ Read more…

By Tiffany Trader

Google Releases Deeplearn.js to Further Democratize Machine Learning

August 17, 2017

Spreading the use of machine learning tools is one of the goals of Google’s PAIR (People + AI Research) initiative, which was introduced in early July. Last w Read more…

By John Russell

Nvidia Responds to Google TPU Benchmarking

April 10, 2017

Nvidia highlights strengths of its newest GPU silicon in response to Google's report on the performance and energy advantages of its custom tensor processor. Read more…

By Tiffany Trader

GlobalFoundries Puts Wind in AMD’s Sails with 12nm FinFET

September 24, 2017

From its annual tech conference last week (Sept. 20), where GlobalFoundries welcomed more than 600 semiconductor professionals (reaching the Santa Clara venue Read more…

By Tiffany Trader

Leading Solution Providers

Graphcore Readies Launch of 16nm Colossus-IPU Chip

July 20, 2017

A second $30 million funding round for U.K. AI chip developer Graphcore sets up the company to go to market with its “intelligent processing unit” (IPU) in Read more…

By Tiffany Trader

Amazon Debuts New AMD-based GPU Instances for Graphics Acceleration

September 12, 2017

Last week Amazon Web Services (AWS) streaming service, AppStream 2.0, introduced a new GPU instance called Graphics Design intended to accelerate graphics. The Read more…

By John Russell

EU Funds 20 Million Euro ARM+FPGA Exascale Project

September 7, 2017

At the Barcelona Supercomputer Centre on Wednesday (Sept. 6), 16 partners gathered to launch the EuroEXA project, which invests €20 million over three-and-a-half years into exascale-focused research and development. Led by the Horizon 2020 program, EuroEXA picks up the banner of a triad of partner projects — ExaNeSt, EcoScale and ExaNoDe — building on their work... Read more…

By Tiffany Trader

Delays, Smoke, Records & Markets – A Candid Conversation with Cray CEO Peter Ungaro

October 5, 2017

Earlier this month, Tom Tabor, publisher of HPCwire and I had a very personal conversation with Cray CEO Peter Ungaro. Cray has been on something of a Cinderell Read more…

By Tiffany Trader & Tom Tabor

Cray Moves to Acquire the Seagate ClusterStor Line

July 28, 2017

This week Cray announced that it is picking up Seagate's ClusterStor HPC storage array business for an undisclosed sum. "In short we're effectively transitioning the bulk of the ClusterStor product line to Cray," said CEO Peter Ungaro. Read more…

By Tiffany Trader

Intel Launches Software Tools to Ease FPGA Programming

September 5, 2017

Field Programmable Gate Arrays (FPGAs) have a reputation for being difficult to program, requiring expertise in specialty languages, like Verilog or VHDL. Easin Read more…

By Tiffany Trader

IBM Advances Web-based Quantum Programming

September 5, 2017

IBM Research is pairing its Jupyter-based Data Science Experience notebook environment with its cloud-based quantum computer, IBM Q, in hopes of encouraging a new class of entrepreneurial user to solve intractable problems that even exceed the capabilities of the best AI systems. Read more…

By Alex Woodie

HPC Chips – A Veritable Smorgasbord?

October 10, 2017

For the first time since AMD's ill-fated launch of Bulldozer the answer to the question, 'Which CPU will be in my next HPC system?' doesn't have to be 'Whichever variety of Intel Xeon E5 they are selling when we procure'. Read more…

By Dairsie Latimer

  • arrow
  • Click Here for More Headlines
  • arrow
Share This