Enlarge / Some current Arm licensees view the proposed acquisition as highly toxic. (credit: Aurich Lawson / Nvidia)

In September 2020, Nvidia announced its intention to buy Arm, the license holder for the CPU technology that powers the vast majority of mobile and high-powered embedded systems around the world.

Nvidia's proposed deal would acquire Arm from Japanese conglomerate SoftBank for $40 billion—a number which is difficult to put into perspective. Forty billion dollars would represent one of the largest tech acquisitions of all time, but 40 Instagrams or so doesn't seem like that much to pay for control of the architecture supporting every well-known smartphone in the world, plus a staggering array of embedded controllers, network routers, automobiles, and other devices.

Today’s Arm doesn’t sell hardware

Arm's business model is fairly unusual in the hardware space, particularly from a consumer or small business perspective. Arm's customers—including hardware giants such as Apple, Qualcomm, and Samsung—aren't buying CPUs the way you'd buy an Intel Xeon or AMD Ryzen. Instead, they're purchasing the license to design and/or manufacture CPUs based on Arm's intellectual property. This typically means selecting one or more reference core designs, putting several of them in one system on chip (SoC), and tying them all together with the necessary cache and other peripherals.

Enlarge / The GPU Grinch doesn't care about your lists or whether you've been naughty or nice. (credit: Aurich Lawson / Dr. Seuss / GettyImages )

Some time last week, thieves stole a large number of Nvidia-based RTX 3090 graphics cards from MSI's factory in mainland China. The news comes from twitter user @GoFlying8, who posted what appears to be an official MSI internal document around the theft this morning, along with commentary from a Chinese language website.

Roughly translated—in other words, OCR scanned, run through Google Translate, and with the nastiest edges sawn off by yours truly—the MSI document reads something like this:

Ensmai Electronics (Deep) Co., Ltd.
Announcement
Memo No. 1-20-12-4-000074
Subject: Regarding the report theft of the graphics card, it is appropriate to reward

Explanation:

1. Recently, high unit price display cards produced by the company have been stolen by criminals. The case has now been reported to the police. At the same time, I also hope that all employees of the company will actively and truthfully report this case.
2. Anyone providing information which solves this case will receive a reward of 100,000 yuan. The company promises to keep the identity of the whistleblower strictly confidential.
3. If any person is involved in the case, from the date of the public announcement, report to the company's audit department or the head of the conflicting department. If the report is truthful and and assists in the recovery of the missing items, the company will report to the police, but request leniency. The law should be dealt with seriously.
4. With this announcement, I urge my colleagues to be professional and ethical, and to be disciplined, learn from cases, and be warned.
5. Reporting Tel: [elided]

Reporting mailbox of the Audit Office: [elided]
December 4, 2020

There has been some confusion surrounding the theft in English speaking tech media; the MSI document itself dates to last Friday, and does not detail how many cards were stolen or what the total value was. The surrounding commentary—from what seems to be a Chinese news app—claims that the theft was about 40 containers of RTX 3090 cards, at a total value of about 2.2 million renminbi ($336K in US dollars).

• 

  The 2020, M1-equipped Mac mini. [credit: Samuel Axon ]

Apple is crazy, right? The Mac just had its best year of sales ever, and Cupertino is hitting the platform with a shock like it hasn’t had in nearly 15 years—back in a time when the Mac was not having such a good year. Apple is beginning the process of replacing industry-standard Intel chips with its own, custom-designed silicon.

In a way, we're not just reviewing the new Mac mini—a Mac mini is always a Mac mini, right? We're reviewing an ARM-based Mac for the first time. And this is not exactly the same story as all the other ARM machines we've looked at before, like Windows 10 on ARM—a respectable option with some serious tradeoffs.

Sure, longer battery life and quick waking from sleep are already out there on other ARM computers. But as you may have seen in our hands-on earlier this week , what we're encountering here is also a performance leap—and as you'll also see in this review, a remarkable success at making this new architecture compatible with a large library of what could now, suddenly, be called legacy Mac software.

On Thursday, an Amazon AWS blog post announced that the company has moved most of the cloud processing for its Alexa personal assistant off of Nvidia GPUs and onto its own Inferentia Application Specific Integrated Circuit (ASIC). Amazon dev Sebastian Stormacq describes the Inferentia's hardware design as follows:

AWS Inferentia is a custom chip, built by AWS, to accelerate machine learning inference workloads and optimize their cost. Each AWS Inferentia chip contains four NeuronCores . Each NeuronCore implements a high-performance systolic array matrix multiply engine, which massively speeds up typical deep learning operations such as convolution and transformers. NeuronCores are also equipped with a large on-chip cache, which helps cut down on external memory accesses, dramatically reducing latency and increasing throughput.

When an Amazon customer—usually someone who owns an Echo or Echo dot—makes use of the Alexa personal assistant, very little of the processing is done on the device itself. The workload for a typical Alexa request looks something like this:

1. A human speaks to an Amazon Echo, saying: "Alexa, what's the special ingredient in Earl Grey tea?"
2. The Echo detects the wake word—Alexa—using its own on-board processing
3. The Echo streams the request to Amazon data centers
4. Within the Amazon data center, the voice stream is converted to phonemes (Inference AI workload)
5. Still in the data center, phonemes are converted to words (Inference AI workload)
6. Words are assembled into phrases (Inference AI workload)
7. Phrases are distilled into intent (Inference AI workload)
8. Intent is routed to an appropriate fulfillment service, which returns a response as a JSON document
9. JSON document is parsed, including text for Alexa's reply
10. Text form of Alexa's reply is converted into natural-sounding speech (Inference AI workload)
11. Natural speech audio is streamed back to the Echo device for playback—"It's bergamot orange oil."

As you can see, almost all of the actual work done in fulfilling an Alexa request happens in the cloud—not in an Echo or Echo Dot device itself. And the vast majority of that cloud work is performed not by traditional if-then logic but inference—which is the answer-providing side of neural network processing.