D2D: Qualcomm versus Everyone
Plus a look at RISC V leader SiFive and Tower Semi
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 and audio recordings of our posts. 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
We did not plan to spend the week writing about Qualcomm, but the news cycle had other ideas.
A few weeks back an investor asked us why Broadcom is able to have a good (or at least normal, commercial) relationship with Apple, while Qualcomm seems locked in a very bad romance with their largest customer. Some of the explanation has to do with Apple’s early ambitions to entirely disrupt mobile, and a lot has to do with the irresistible force meeting immovable object nature of the two companies’ position on intellectual property (IP).
Then Qualcomm announced a RISC V-based chip running a version of Android for digital watches. Is this just a simple product for a small category, or is it the sign of Qualcomm’s bigger plans for RISC V? It is probably the former, RISC V is well suited for IoT devices like wearables. That being said, if the two companies did want to move more of Android to RISC V semis, a good way to start would be building the developer ecosystem in a small out-of-the-way market.
Then we learned that Qualcomm and Google’s relationship is not great. And it occurred to us that Qualcomm seems to have a hard time getting a long with anyone. Maybe we can chalk this up to their long history of having to fight the world to make CDMA a reality, but it seems like those habits are now working against the company. Qualcomm is fighting its largest customer, one of its most critical IP providers, and a host of regulators. They have a minimal relationship with Amazon, and we have a lot of questions about their relationship with Microsoft. We think this is solvable, but the company needs to revamp its org chart, and ensure that its sales people and not its lawyers are the lead point of contact for major customers and partners.
SiFive is one of the leading commercial RISC V providers. We wrote about them a few years ago with a degree of caution. Last week, we attended their analyst briefing and came away much more interested in their prospects. They have rationalized their business model and product offering. We need to start thinking of them as a direct competitor to Arm, and on that basis they are well positioned to benefit from the growing interest in RISC V.
Much has been written about Intel since they abandoned their attempt to acquire Tower Semi, but much less attention has been given to Tower. They have been very active, announcing new products, partners and customers. They are working hard, but probably have a few tough years ahead of them.
If you like this content, you should check out our podcast The Circuit
Semis and Hardware
There are signs that the long smartphone and PC winter is ending, or at least getting close to the end. On the latest episode of the Circuit we talk about TSMC’s commentary that we are “very close” to the bottom”. And there are many other signs coming out of Asia that the companies in the smartphone supply chain are finally starting to restock their component inventory. Our guess is that we will hear more about this in commentary in the coming earnings season, but actual results may not start to improve until the new year.
Like many, we are cautiously optimistic that Intel has turned the corner and is finally able to get its manufacturing process back on track. But we are less optimistic about the prospects for Intel as a foundry, producing chips for others. And some are just outright pessimistic about Intel’s chances in the foundry market, like this AMD executive.
Chip start-up Zero ASIC came out of stealth this week. They are positioning themselves in between fully custom ASICs and FPGAs. Interesting idea if there are enough companies that are big enough to want their own chips but not big enough to afford the work needed to build an internal product.
Canon unveiled its latest lithography solution, nanoimprint lithography (NIL). Not fully competitive with ASML’s EUV, but a reminder that technology evolves and the dominant solution of today may not hold that position forever.
Toshiba is delisting from the stock exchange and after being taken private by a sort-of government-led buyout. Western Digital is also reported to secured financing to acquire memory maker Kioxia. There are many signs that Japan, Inc. is waking up to ‘shareholder value’ and beginning to rationalize its electronics sector. There is immense value waiting to be unleashed here.
One more benchmark of x86 versus Arm for cloud workloads. Arm, in the form of AWS’s Graviton, scores well on many metrics but not all.
Networking and Wireless
Last newsletter we linked to a story about Facebook’s efforts to reposition Ethernet for AI workloads, in opposition to Nvidia’s heavy promotion of its Infiniband. Facebook/Meta is not alone, Arista had a blog post recently extolling the virtues of “Ultra-Low Latency” Ethernet. That low latency bit is what matters in this context.
A list of the Top Ten largest streaming events. Networking is important and challenging for things other than AI. Also, this is a fun list, a lot of Cricket, esports and football (soccer).
Software and the Cloud
A review of AMD GPUs for AI workloads, with some surprisingly positive commentary about AMD’s ROCm alternative to Nvidia’s Cuda. AMD still has a long way to go in cloud AI, but this is a good sign of progress.
A deep look at the requirements and performance of many leading messaging apps. The author has compiled a lot of data, probably too much and it can be hard to read. No easy conclusions. Facebook appears to be the biggest resource user, but beyond a certain threshold the others all catch up.
A look at how GPUs work. This is written for software developers, but is worth reading for anyone interested in how CPUs and GPUs differ, and what each is good at.
An engine built with DNA. No idea what this is good for, but it is deeply fascinating.
Why do hospitals still use pagers? The short answer is they are very reliable, reaching places that Wi-Fi does not, also inertia and ties to legacy software systems.
Thank you for reading D2D Newsletter. This post is public so feel free to share it.