Linley Newsletter: April 11, 2019

 weSRCH's Best of the Internet Award
  11th-Apr-2019
 338


Linley Newsletter

Please feel free to forward this to your colleagues

Issue #647

April 11, 2019


Independent Analysis of Microprocessors and the Semiconductor Industry

Editor: Tom R. Halfhill

Contributors: Linley Gwennap, Mike Demler, Bob Wheeler


In This Issue:

- 2nd-Gen Xeon Scalable Adds Cores

- First Optane DIMMs Disappoint

- Intel 10nm FPGAs Add Chiplets

- Qualcomm Amps Up Smart Speakers


2nd-Gen Xeon Scalable Adds Cores

By Linley Gwennap

As part of a massive data-center announcement, Intel rolled out 58 new models representing the second generation of its Xeon Scalable processor family. Although the new design, code-named Cascade Lake, offers little performance benefit at the high end, the company increased the core count for many of its most popular models, delivering about 30% more performance at the same price. The new products, which are socket-compatible with Skylake-SP and fit into the same Purley platform, are in production now.

Cascade Lake also brings new capabilities to improve performance on specific workloads. The new processors introduce Intel's VNNI extension, branded DL Boost, accelerating many AI inference workloads by 2-3x compared with Skylake-SP. They're also the first to support Optane DIMMs, a new type of fast but persistent memory that connects directly to the processor alongside standard DRAM.

At the top of the line, the new Platinum 92xx products combine two die in one package to fit as many as 56 CPU cores per socket. Although the 92xx models double core count, cache size, and memory bandwidth compared with even the 82xx models, they require a new socket and burn as much as 400W. Intel doesn't plan to sell these supersize processors directly, instead offering them to OEMs on custom motherboards.

Cascade Lake uses the same CPU microarchitecture as Skylake and the same process node, limiting the gain in general-purpose performance. The transistors improve slightly from 14nm+ to 14nm++, allowing Intel to wring out a bit more clock speed. For example, the base speed of the flagship Platinum product rises from 2.5GHz in the 8180 to 2.7GHz in the new 8280 with the same core count and power (TDP). Many midrange models receive a bigger boost in clock speed or core count, but often with a commensurate rise in TDP.

Microprocessor Report subscribers can access the full article:

https://www.linleygroup.com/mpr/article.php?id=12122

First Optane DIMMs Disappoint

By Linley Gwennap

After a long wait, Intel's first Optane memory modules are now in production. For more than three years, the company has promised to deliver a revolutionary memory that is persistent, is less expensive than DRAM, and has "near-DRAM" speed. The new Optane DIMMs, code-named Apache Pass (AEP) and officially branded as Intel Optane DC Persistent Memory, mostly meet these goals, but they're considerably slower than DRAM -- so much so that they typically require some amount of DRAM to function as a cache.

Intel offers the new Optane DIMMs in three capacities: 128GB, 256GB, and 512GB. All have lower per-byte cost than DRAM, although the 512GB Optane DIMM sells for nearly $7,000. Since DRAM DIMMs top out at 256GB, Optane allows customers to pack more memory in their servers than ever before: up to 4.5TB per processor. The company previously offered Optane memory only in SSDs and a small PCIe stick.

Although Optane DIMMs plug into a standard memory slot, they use a modified protocol called DDR-T to communicate with the processor. The only processor that supports this protocol, and thus Optane DIMMs, is Intel's new second-generation Xeon Scalable, known as Cascade Lake.

The new DIMMs can improve performance in two ways. First, they allow larger memory systems at a lower cost per byte than DRAM. The larger memory system boosts performance for big data and other memory-intensive workloads. At a system level, the cost savings are about 20%, enough for only a small memory increase. Second, by retaining their value during a power loss, the Optane DIMMs can greatly reduce boot time. For customers that focus on minimizing downtime, faster reboots have a significant benefit. NVDIMMs, which combine DRAM with a battery backup, offer the same function.

Microprocessor Report subscribers can access the full article:

https://www.linleygroup.com/mpr/article.php?id=12123

Intel 10nm FPGAs Add Chiplets

By Tom R. Halfhill

Intel's next-generation Agilex FPGAs diverge from Xilinx's new Versal devices in important ways. Although both companies want to expand the FPGA market by augmenting their programmable logic with faster processing and I/O, Intel favors multichip packages tailored for particular applications or customized for special customers. Xilinx, which disclosed its Versal family last year, favors more chip-level integration and preconfigured product tiers. Both vendors plan to sample their new devices later this year and begin volume shipments in 2020.

Agilex (code-named Falcon Mesa) is Intel's first new FPGA family since acquiring Altera in 2015. Enough time has elapsed since the $16.1 billion acquisition -- Intel's most expensive -- to forge a new strategy that counters Xilinx and exploits the strengths of the merged companies. Although product details remain sketchy, the strategy is clear. Intel will use its new 10nm FinFET process and copackaging technology to build FPGAs that combine a programmable-logic chip with one or more chiplets. Some Stratix FPGAs and x86 processors already employ this packaging technology, but Agilex expands it and allows customers to develop their own chiplets.

Chiplets can add high-speed transceivers, memory-coherent host-processor interfaces, PCI Express interfaces, or custom logic. Intel will offer chiplets with 112Gbps serdes, PCIe Gen5, High Bandwidth Memory (HBM), and 3D XPoint memory controllers. Its 2018 acquisition of eASIC allows customers to harden part of their programmable-logic design into a tightly coupled chiplet that delivers higher performance and power efficiency.

Thanks largely to the process shrink from 14nm to 10nm, Intel says Agilex will deliver up to 40% more performance than existing Stratix 10 FPGAs or consume up to 40% less power. Peak floating-point performance soars to 40Tflop/s. Sampling is scheduled for 3Q19, although we expect the company will roll out the entire product family over several quarters.

Microprocessor Report subscribers can access the full article:

https://www.linleygroup.com/mpr/article.php?id=12125

Qualcomm Amps Up Smart Speakers

By Mike Demler

Qualcomm's new smart-audio platform enables in-device speech recognition of more than 100 phrases. It includes four models of the QCS400-series application processor, targeting products ranging from speech-enabled mini-speakers (e.g., Amazon's Echo Dot) to display-equipped devices (e.g., Google's Home Hub). The processors work with the CSRA6640 direct-digital-feedback amplifier (DDFA), which integrates an audio DSP and a class-D amplifier that delivers up to 20W per channel to stereo speakers, making it suitable for small sound bars. Alternatively, the CSRA6640 can bridge the output drivers to create a single 40W amplifier.

The QCS400 chips are new designs, but they repurpose major elements from older Snapdragon smartphone chips. They provide a new wireless baseband that implements 802.15.4 Zigbee, which is popular for smart-home applications. The other wireless basebands are the same as in premium Snapdragons, including Bluetooth 5.1 and 802.11ac Wi-Fi. The latter implementation supports 1x1 and 2x2 MIMO. Qualcomm withheld the process technology it uses to manufacture these chips, but we expect it's 28nm.

To ease product development, the company offers two system-on-module (SoM) development kits, which integrate a QCS400 processor, a WCN3980 Bluetooth+Wi-Fi RF combo, and a PMS405 PMIC. The SoMs are user configurable, but Qualcomm designed them to be production ready. An optional speaker casing contains two speakers and up to eight microphones. Software support includes a mobile app for music streaming and connection to Alexa voice services.

Microprocessor Report subscribers can access the full article:

https://www.linleygroup.com/mpr/article.php?id=12124

About Linley Newsletter

Linley Newsletter is a free electronic newsletter that reports and analyzes advances in microprocessors, networking chips, and mobile-communications chips. It is published by The Linley Group. To subscribe, please visit: http://www.linleygroup.com/newsletters/newsletter_subscribe.php

Domain: Electronics
Category: Semiconductors

Recent Newsletters

Linley Newsletter: August 8, 2019

Linley Newsletter Please feel free to forward this to your colleagues Issue #664 August 8, 2019 Independent Analysis of Microprocessors and the Semiconductor Industry E

08 August, 2019

Linley Newsletter: August 1, 2019

Linley Newsletter Please feel free to forward this to your colleagues Issue #663 August 1, 2019 Independent Analysis of Microprocessors and the Semiconductor Industry E

01 August, 2019

Linley Newsletter: July 25, 2019

Linley Newsletter Please feel free to forward this to your colleagues Issue #662 July 25, 2019 Independent Analysis of Microprocessors and the Semiconductor Industry

25 July, 2019