D2D: More questions than answers
A look at the x86 cross license and what it would mean for an acquisition of Intel, and a look at Arm and Qualcomm results.
In which we explore Arm’s and Qualcomm’s latest results, take a look at changing automotive semi trends and the strong utility of MCUs, as well as lots of small movements in telecom software and standards.
Just a reminder that we operate our newsletters on a paid model. Paid subscribers will get three newsletters a month including our China Deep Tech notes. Paid subscribers will also get early access to the newsletter as well other benefits coming soon. Please subscribe and support our work.
Highlights from our Blog
One under-explored aspect of any potential acquisition of Intel is the fact that AMD has the right to effectively block any deal through their cross-license agreement for x86. We think any such deal is unlikely, but if it did happen AMD has a few options. They could block the deal, but that would lead to its own problems. So what concessions could they ask for to grant the deal? We have a few ideas.
In our ongoing exploration of the business of selling servers, not just chips, we have noticed that almost no one mentions Sales Support Engineers, or Field Application Engineers. These are a critical resource that can often mean the difference between a busted demo and committed purchase order. This blind spot exists not only for outside observers but often for chip companies themselves.
Arm reported a solid quarter, and raised guidance, but left their full year numbers unchanged. We read that as the company tuning the way it issues guidance, because the core business is doing very well. They are steadily increasing their value capture across semis, a process that shows no signs of slowing down any time soon.
Qualcomm reported good earnings leading to a jump in their share price, only to give back those gains two days later. Our sense is that most investors are deeply ambivalent about the company with its lackluster growth prospects. Maybe that changes as we learn more about their prospects in the automotive business, but the Arm lawsuit and the pending loss of Apple remain significant overhangs.
If you like this content, you should check out our podcast The Circuit
Semis, Hardware and Deep Tech
Tom’s Hardware asks if Intel is too big to fail. Tom’s knows their stuff when it comes to chips, but we are not so certain the policy makers see it the same way. There seems to be a growing consensus in the semis industry that the only way for Intel to survive is with a sizable amount of government support. We do not share that view. That being said, policy making is beyond our ability to forecast, especially now. It’s just that we cannot fail to notice that many of the people calling for government support stand to benefit from it in some way.
The next year or two are going to be important in determining the systems used to power automotive semis. Qualcomm is likely to start seeing their ADAS and Infotainment systems start appearing in quantity in production vehicles. The PRC EV makers have multiple chips coming on stream as well. But in this rush for large processors, we think it is important to remember most of auto semis today are based on microcontrollers (MCUs), and in many cases enhanced MCUs may work very well for ‘advanced’ compute. For instance, Renesas unveiled a new family of highly capable auto MCUs last week. These notably include some ADAS functionality. In a related vein, ST Micro also provided some good background on ‘zonal control’ an important design trend for the large MCU makers.
Amazon donated the use of a 40,000 node cluster of Tranium AI chips to academic use. Very generous. Of course, those are Tranium version 1, and Amazon is about to launch Tranium 2.
RISC V is premised on the idea of flexibility. So we should celebrate all the publicly available extensions that are now available. On the other hand, there are now an awful lot of extensions available, how compliant is everyone’s silicon?
We profiled Orca Semi a few months back. They took a fresh approach to a sizable corner of the analog market with what we thought was some good engineering and reasonable expectations. They have now extended their product line to another sizable adjacency.
An accessible primer on how GPUs work and the importance and utility of CUDA.
A close examination of the Performance Cores (p cores) in Apple’s latest M-Series CPU. Key takeaway - these chips can get away with using a really small amount of power.
Networking and Wireless
Remember RCS? This is a messaging system first proposed by the wireless operators seeking to compete against Apple Messages and WhatsApp. It made little progress until Google adopted it for largely the same reasons. From the start, we discounted it because we have seen this movie before (many times, anyone remember WAC or OMA?). Apple finally adopted it last year, probably to head off European regulators. The whole project is essentially now run entirely by Google, so of course it is a mess.
Telecordia, the company that runs mobile number portability in the US is being sold to private equity. This author takes a very alarmist approach to that deal, but the article does provide a good piece of background on this particular bit of critical infrastructure.
Remember Open RAN and Qualcomm’s ambition in the infrastructure market? They have just seen their first live deployment in Vietnam. Given that this is the only customer they have referenced after years of marketing the chip, and just the general state of the OpenRAN market, we imagine there may not be many more operators following suit.
6G is still very far away, but not so far off that it is too early for the geopolitical wrangling to start. PRC regulators recently called for a “unified, global 6G standard”. Light Reading interprets this as a plea to operators to help them navigate the next round of the US-China Trade War. We read it is the PRC reminding the world that they have a big voice in setting the standard and intend to exert that influence as the standard starts to come together in coming years.
An exploration of how Ethernet works at a very low level.
Software and the Cloud
Addition is all you need. Another academic approach at simplifying AI math and thus eliminating the need for so much (expensive) GPU compute. One of these days, one of these approaches may actually land.
We link to a lot of stories about the rising costs of using public cloud infrastructure. This company now has a service to orchestrate migrating data around all the various clouds to optimize pricing - which is both genius and also a perfect demonstration about how something in not quite right in how systems are built today.
Diversions
In the Bay Area a healthy industry exists to liquidate failed start-ups. These first emerged in the early 00’s and the dot.com aftermath, and have been doing steady business ever since. Most notable for anyone looking for some nice but inexpensive office furniture.
Like many people today, we have to deal with a fair amount of cognitive overload. This developer tries manage that with code, useful ways to think about thinking, but also just one more thing to think about.
Image by Microsoft Co-Pilot
Thank you for reading D2D Newsletter. This post is public so feel free to share it.
Oops. Apologies. Here is the correct link
https://blog.codingconfessions.com/p/gpu-computing
Great post as usual but I think the Gremlins got to the GPU/CUDA link.