Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Current Articles
July 13, 2020
-
Wear 4100+ Gets 12nm Upgrade
The Wear 4100 is a strong upgrade from the Wear 3100, increasing performance while decreasing power consumption. It also adds a new feature: support for a 16MP camera.
-
Stratix 10 NX Adds AI Blocks
Intel has redesigned the Stratix 10 DSP block to focus on neural-network inference, adding 15x more multipliers to greatly boost performance for custom AI accelerators.
-
Editorial: Arm Wins Big in Apple, HPC
Arm’s design wins in next-generation Mac computers and the world’s fastest supercomputer prove the CPU architecture can compete for big systems as well as little ones.
July 6, 2020
-
Cooper Lake Limps to Market
Intel’s 3rd Generation Xeon Scalable processors target high-end systems, particularly those that perform AI training, but offer few other advantages relative to current Xeon products.
-
Qualcomm Equips AI-Powered Robots
The RB5 is a configurable developer kit powered by a Snapdragon 865–based QRB5165 processor. It comes with mezzanine cards that include cameras, motor controllers, sensors, and cellular radios.
-
Xilinx Alveo Makes Video Zoom
Xilinx has developed two new real-time-server reference architectures based on its Alveo U30 and U50 accelerator cards. Both designs target live-video transcoding using adaptive-bit-rate technology.
June 29, 2020
-
Lakefield Shrinks x86 SoCs
Intel’s Core Hybrid processors (Lakefield) are the first x86 chips to employ Foveros die-stacking technology and a cluster of large and small CPU cores. Targeting slim notebooks and tablets, they compete mainly with Arm-based processors.
-
Snapdragon 690 Lowers 5G Entry Price
Qualcomm’s newest mid-premium 5G processor brings 7-series features to the top of the 6-series lineup. It’s the first 6-series chip with Cortex-A77 CPUs, replacing both the Snapdragon 675 and 720G.
-
Qualcomm Delivers 6GHz Wi-Fi
Qualcomm added 6GHz support to its Networking Pro chipsets for Wi-Fi access points and to its FastConnect 6700 and 6900 combo chips for next-generation smartphones.
-
Certus-NX Extends Lattice Lineup
The second member of Lattice’s Nexus FPGA family brings the benefits of 28nm FD-SOI technology to a broader range of applications, adding hard cores for Ethernet, encryption, and A/D conversion.
-
Tom Halfhill Retires, Aakash Jani Joins
After 20 years, Tom is retiring from his senior editor role at Microprocessor Report but will join the editorial board. New senior editor Aakash Jani is joining Microprocessor Report from AMD.
June 22, 2020
-
Pliops Makes Storage Computational
Computational storage uses new system architectures to move some processing closer to storage. Startup Pliops is shifting more of the stack into its storage processor, multiplying SSD performance.
-
EnSilica Drives Automotive Radar
EnSilica has delivered its eSI-ADAS intellectual property for radar-signal processors. It also offers eSi-Medisense, an IP platform that enables wireless vital-sign monitoring in ultra-low-power wearables.
June 15, 2020
-
Pensando, Xilinx Debut Smart NICs
Seeking to increase demand for merchant smart NICs, two new entrants plan to serve the broader market beyond hyperscale data centers.
-
Editorial: US Sanctions Squeeze Huawei
New sanctions that block TSMC and other foundries from supplying ASICs to Huawei will hurt that company’s competitiveness and revenue but will also have unintended consequences.
-
Dimensity 820 Shows True Colors
Just a month after releasing the Dimensity 800, MediaTek has upgraded most of the design’s capabilities to create a new processor that appears in smartphones selling for as little as $225.
June 8, 2020
-
Nvidia A100 Tops in AI Performance
The first GPU to implement the Ampere architecture, the A100 delivers huge performance increases to data centers running AI, data analytics, and high-performance computing (HPC) workloads.
-
Nvidia Ampere Delivers AI Innovation
This deep dive into Nvidia’s latest data-center GPU architecture explains the new training data types, structured sparsity, cache compression, GPU virtualization, and other features.
-
Kirin 985 Upgrades AI Performance
The Kirin 985 gives Huawei a competitive edge in the 5G mid-premium segment while maintaining a healthy distinction from the premium segment, which the Kirin 990 5G occupies.
June 1, 2020
-
Cortex-X1: A Bigger Arm
Cortex-X1 debuts a series of Arm CPUs that provide a new level of smartphone performance. The company also released Cortex-A78, which increases efficiency for the mainstream line.
-
Mali-G78 Raises Premium Performance
The new Mali-G78 GPU strengthens Arm’s hold on licensable IP for premium phones. It outperforms the Mali-G77 by about 15%, and its larger maximum core count enables even greater performance gains.
-
Ethos-N78 Boosts AI Efficiency
The new Ethos-N78 is Arm’s second-generation deep-learning accelerator (DLA). In its largest configuration, the N78 can deliver peak performance of 10 trillion operations per second (TOPS).
May 25, 2020
-
Andes Plots RISC-V Vector Heading
Andes has delivered the first CPU core to support the new RISC-V vector extensions (RVV), even though RISC-V International (formerly the RISC-V Foundation) won’t freeze the v1.0 specifications until August.
-
Innovium Debuts 25.6Tbps Switch
Startup Innovium is first to announce a 25.6Tbps Ethernet switch chip using next-generation 100Gbps PAM4 serdes. Teralynx 8 doubles the per-lane data rate of its predecessor, yielding up to 64x400G Ethernet ports.
May 18, 2020
-
Intel Takes Desktop PCs to 10th Gen
New 14nm processors based on Comet Lake-S offer 10 cores at the high end, clock speeds up to 5.3GHz, and Hyper-Threading for all models, delivering greater value for most customers.
-
Prophesee Detects Rapid Movement
Prophesee is a French startup developing event-based image sensors. It’s working with Sony on its fourth-generation device that delivers one-microsecond temporal resolution with HD spatial resolution.
-
Editorial: Wave Goodbye to MIPS
Wave Computing’s bankruptcy filing leaves the future of its MIPS architecture group in doubt. One of the original RISC designs, MIPS once had a bright future but has struggled in recent years.
-
Sima Seeks Greener AI
To bring a new level of power efficiency to embedded systems that require neural networks, Sima is designing a chip called the MLSoC that it expects will deliver an impressive 10 TOPS/W.
May 11, 2020
-
Perceive Reduces AI to a Math Problem
Two-year-old startup Perceive has shattered power-efficiency benchmarks for edge-AI processors. Its first chip, Ergo, needs just 70mW to deliver 250fps on Yolo v3, all without connecting to DRAM.
-
LeapMind Jumps on Binary Networks
A Japanese startup is licensing a hardware design that runs binary neural networks (BNNs) at 28 TOPS per watt. It also offers software tools that greatly improve the accuracy of BNN models.
-
Helio P95 Boosts AI and Graphics
MediaTek is offering a slight upgrade to last year’s octa-core P90 for mid-premium smartphones. It barely stays competitive with similar chips from Huawei and Qualcomm.
May 4, 2020
-
XMOS Xcore.ai Adds Vector Unit
XMOS enhanced its third-generation Xcore design to include neural-network acceleration, enabling smart-home and industrial devices to add AI capability at a chip price of just $1.
-
MaxLinear Disconnects Intel’s Home
MaxLinear is acquiring Intel’s Home Gateway Platform Division, separating well-hidden assets from the behemoth’s Client Computing Group (CCG). It’s getting a mixed bag of goodies from past Intel broadband acquisitions.
-
AMD Targets HPC With Faster Epyc
AMD introduced three new models in its Epyc processor line that demonstrate industry-leading per-core performance. Some also provide more cache memory than previous versions.
April 27, 2020
-
RISC-V Vectors Know No Limits
New vector extensions for the open-source RISC-V architecture are nearing formal adoption. They support almost unlimited register widths, data lanes, and data types to fit any design.
-
Flex Logix Accelerates DSP
On the basis of customer input, the embedded-FPGA developer has applied its NNMax deep-learning accelerators (DLAs) to finite-impulse-response (FIR) filters, achieving strong performance.
April 20, 2020
-
Ceva SensPro Fuses AI and Vector DSP
Ceva’s new SensPro gives designers a versatile and highly configurable set of licensable intellectual-property (IP) cores for scalar and vector signal processing as well as neural-network acceleration.
-
Comet Lake-H Maxes Out Clock Speed
Intel has refreshed its Core H-series with what should be its final Skylake design. The 10th Generation processor lines up against AMD’s new Ryzen 9 4900H.
-
Editorial: Publishing Under a Quarantine
Many technology companies are struggling to develop new products during the unprecedented lockdowns, but The Linley Group delivers online articles and a virtual conference.