Category: Scoop.it

 

Cluster-Scale Performance on a Single Large-Wafer Chip

Video is here

Programming a cluster to scale deep learning is painful. It typically requires dozens to hundreds of engineering hours and remains a practical barrier for many to realize the value of large-scale AI for their work. On a traditional GPU cluster, ML researchers – typically using a special version of their ML framework – must figure out how to distribute their model while still achieving some fraction of their convergence and performance target. They must navigate the complex hierarchy of individual processors’ memory capacity, bandwidth, interconnect topology, and synchronization; all while performing a myriad of hyper-parameter and tuning experiments along the way. What’s worse is that the resultant implementation is brittle to change, and this time only delays overall time to solution. With the WSE, there is no bottleneck. We give you a cluster-scale AI compute resource with the programming ease of a single desktop machine using stock TensorFlow or PyTorch. Spend your time in AI discovery, not cluster engineering.

Learn more

 

 

Designed for AI

 

Each core on the WSE is independently programmable and optimized for the tensor-based, sparse linear algebra operations that underpin neural network training and inference for deep learning, enabling it to deliver maximum performance, efficiency, and flexibility. The WSE-2 packs 850,000 of these cores onto a single processor. With that, and any data scientist can run state-of-the-art AI models and explore innovative algorithmic techniques at record speed and scale, without ever touching distributed scaling complexities.

 

1000x Memory Capacity and Bandwidth

 

Unlike traditional devices, in which the working cache memory is tiny, the WSE-2 takes 40GB of super-fast on-chip SRAM and spreads it evenly across the entire surface of the chip. This gives every core single-clock-cycle access to fast memory at extremely high bandwidth – 20 PB/s. This is 1,000x more capacity and 9,800x greater bandwidth than the leading GPU. This means no trade-off is required. You can run large, state-of-the art models and real-world datasets entirely on a single chip. Minimize wall clock training time and achieve real-time inference within latency budgets, even for large models and datasets.

220Pb/s

 

 

High Bandwidth – Low Latency

 

Deep learning requires massive communication bandwidth between the layers of a neural network. The WSE uses an innovative high bandwidth, low latency communication fabric that connects processing elements on the wafer at tremendous speed and power efficiency. Dataflow traffic patterns between cores and across the wafer are fully configurable in software. The WSE-2 on-wafer interconnect eliminates the communication slowdown and inefficiencies of connecting hundreds of small devices via wires and cables. It delivers an incredible 220 Pb/s processor-processor interconnect bandwidth. That’s more than 45,000x the bandwidth delivered between graphics processors.

Read the full article at: www.cerebras.net

 

Despite being larger than the original Starlink satellites, the new “Mini” version is fainter, meeting astronomers’ recommendations.

 

 

SpaceX launched their first batch of second-generation Starlink satellites on February 27th. These spacecraft are called “Mini,” but they are only small in comparison to the full-size satellites that will come later. The 116 square meters of surface area make them more than four times the size of the first-generation spacecraft.

The Minis’ large dimensions were an immediate concern for professional and amateur astronomers alike because area usually translates to brightness. However, SpaceX changed their physical design and concept of operations (conops) in order to mitigate their brightness. The company developed a highly reflective dielectric mirror film and a low-reflectivity black paint, which are applied to several parts of the spacecraft body. The mirror-like surface reflects sunlight into space instead of scattering it toward observers on the ground. In addition, the solar panels can be oriented so that observers do not see their sunlit sides.

 

 

The brightness mitigation plan sounded promising but measurements were needed to determine its effectiveness. So, a group of satellite observers began recording magnitudes. Scott Harrington recorded the first data point visually on March 14th. He has since obtained 125 additional magnitudes from his dark-sky location in Arkansas. Meanwhile, Andreas Hornig developed software to process video observations. He derived 108 magnitude measurements recorded from Macedonia on the night of April 12th alone. In all, we have acquired 506 brightness measurements for our study.

 

SpaceX launched three additional batches of 21 or more Mini satellites in April, May, and June. These spacecraft ascend from low, orbit-insertion heights toward their eventual altitude at 560-km (350 mi). Until May, we were observing Mini satellites at all heights without knowing whether they were operating for brightness mitigation. Then Richard Cole in the UK noticed that some spacecraft had leveled off at 480 km. He reasoned that these satellites might already be in mitigation mode and suggested that we prioritize them.

 

We found that the Minis at that height were several magnitudes fainter than those at other altitudes. SpaceX sent us a message on May 16th confirming that Richard was correct. Now that we could distinguish between mitigated and unmitigated spacecraft, we began to characterize the brightness of each group, prioritizing measurements for those satellites that were already operational.

Observed brightness indicates how severely satellites impact celestial observations. The average magnitude for mitigated Mini spacecraft in our database is 7.1, just below the limit set by astronomers’ recommended guidelines. So, most of them are invisible to the unaided eye and do not interfere greatly with research.

Read the full article at: skyandtelescope.org

An autonomous drone has competed against human drone-racing champions — and won. The victory can be attributed to savvy engineering and a type of artificial intelligence that learns mostly through trial and error.

 

First-person view (FPV) drone racing is a televised sport in which professional competitors pilot high-speed aircraft through a 3D circuit. Each pilot sees the environment from the perspective of their drone by means of video streamed from an onboard camera. Reaching the level of professional pilots with an autonomous drone is challenging because the robot needs to fly at its physical limits while estimating its speed and location in the circuit exclusively from onboard sensors.

 

Here the authors of this paper introduce Swift, an autonomous system that can race physical vehicles at the level of the human world champions. The system combines deep reinforcement learning (RL) in simulation with data collected in the physical world. Swift competed against three human champions, including the world champions of two international leagues, in real-world head-to-head races. Swift won several races against each of the human champions and demonstrated the fastest recorded race time.

 

This work represents a milestone for mobile robotics and machine intelligence, which may inspire the deployment of hybrid learning-based solutions in other physical systems. An autonomous system is described that combines deep reinforcement learning with onboard sensors collecting data from the physical world, enabling it to fly faster than human world champion drone pilots around a race track.

Read the full article at: www.nature.com

A team of experts at University College London are working to make drugs more palatable – and at a speed never seen before

Read the full article at: www.telegraph.co.uk

Illinois passed a law to protect child influencers. Advocates are cautiously optimistic more states will follow.

Read the full article at: www.nbcnews.com