257: Nvidia's 34 Announcements, Taiwan's Engineer Shortage, Datadog, Cadence Software, James Webb, Crowdstrike on the Dark Web, and Clever Statues
"Mister, I don’t sell to fish.”
To doubt everything, or, to believe everything, are two equally convenient solutions; both dispense with the necessity of reflection.
🛀 I wonder what's the total size of data stored on public cloud hard drives. *All* kinds for *all* customers, including redundancy/backups/etc.. just the raw total storage of the whole thing.
While at it, I wonder what’s humanity’s total distributed storage capacity, every single device that can store 1s and 0s, PCs, laptops, phones, servers, USB drives, SD cards, magnetic tapes… *Gary Oldman voice in The Professional* EVERYYYYYYTHIIIIIING
✍️ 📩 📫 Friend-of-the-show 10-K Diver (🤿) has a good thread about the economics of newsletters:
He covers the ‘fundamentals’ of the business model well. But on the more day-to-day operational side, here are a few of my observations:
By far the hardest thing is the consistency/discipline ONCE THE HONEYMOON PERIOD IS OVER. For a while at first, it's super-exciting and everything is new. But that goes away, and most people stop around then, when all that's left is the pain of writing. It's also very hard to have longevity in newsletter/blogging, because few people have an *end goal of writing*. Most stumble into it as a way to get to somewhere else, so few people stick with it for a long time. Either they try to leave, or get offers that pull them out. I think what many people struggle with in any "business of one" is that *everything is on you*. Feels good when it works, but there's nobody else to blame when it doesn't, or to support you when you want a break or life gets in the way. Most outside observers also probably don't realize how much successful writers work at it, evenings and weekends, and how many years they had been building up their skills before they even started their current projects.
🧸🪀 Only after becoming a parent do you realize how *horribly* most kids’ toys are designed.
The vast majority seem to be aimed at being something that a parent can look at on a shelf at Toys “R” Us and look good enough to buy.
Playing with it, having fun? Not part of the equation…
It’s like the old joke about the fisherman at the fishing tackle store:
“My God, they’re purple and green. Do fish really take these lures?”
The salesman said, “Mister, I don’t sell to fish.” 🎣 🐠
Sometimes it feels like the people who design toys have never watched kids try to play with them, or watched parents try to assemble them.
The very worst category: toys that parents would immediately recognize as being total BS, but that non-parents will buy for other people’s kids. 🤬
These are worse than useless, because they clutter your house and when you try to get rid of them, kids don’t want to because they still get attached to them (even though they’re not fun and never get played with).
💚 🥃 Just to keep you in the loop, this is what the total number of subs to the newsletter now looks like, getting close to 6,500.
I like how it just kind of chugs along, without huge spikes (or dips! 😬).
The real engine of growth here is if you tell someone you know, “Hey, I think you may like this, check it out”.
I’m sure the line would go up faster with all kinds of growth-hacking best practices, but I don’t want the most readers, I want the best readers, so I find word-of-mouth to be the most effective discovery method for that ¯\_(ツ)_/¯
Liberty’s Highlights is reader-supported. To support my work, consider becoming a free or paid subscriber. 🧯
A Word From Our Sponsor: ⭐️ Tegus ⭐️
What makes a great investor isn’t pedigree or diplomas 🎓, it’s quality thinking and information 🧠
Tegus can’t provide the thinking for you, but on the information front, they’ve got you covered! 📚
With 25,000+ primary source expert calls covering almost any industry or company you may want to learn about, and the ability to conduct your own calls with experts, if you so choose, at a much lower cost than you’ll find elsewhere (70% lower than the average cost of ˜$400), this is the service to turbo-charge the depth and breadth of your knowledge 📈
Investing & Business
🔥 Nvidia GTC 2022 🔥
Nvidia’s semi-annual GTC conference took place yesterday (this was the AI-centric event, the gaming even takes place in the fall).
There was way too much to mention everything (I counted them, and Nvidia’s news page has 34 press releases about the event), but here are my highlights:
First, I have to say that their intro ‘I am AI” video is quite well-made, and has some old Apple qualities. It’s not just about the technology, but about the intersection of that technology and humanity. That’s a big difference.
Around 16 minutes into the keynote, they show an animated ‘knight’, and I thought the animation was some of the best that I’ve seen for something that wasn’t lovingly tweaked by animators (like in a Pixar film). It felt like the character had weight, inertia, and really interacted with its environment. Really looking forward to what this type of technology can do for games and film/TV VFX.
They mentioned their language model, Megatron, which has 530bn parameters (vs 175bn for GPT-3). I wish they had a sandbox API to play with it, like OpenAI…
Top billing was the H100 chip, the first member of the new Hopper architecture, which is a successor to the crazily successful Ampere (2 years later and still supply-constrained).
To introduce it, Jensen said:
A single H100 can sustain 40 terabits/second of I/O bandwidth.
To put it in perspective, twenty H100 can sustain the quivalent of the entire world’s internet traffic.
The man is a poet with technological analogies. 🌸 He’s the closest that we have to Steve Jobs on that front right now.
80 billion transistors using TSMC’s new 4N process, 4 petaflops at FP8 precision or 1 petaflops at TF32 precision. They also created a Transformer Engine to further accelerate AI workloads on the very popular transformer AI models: “The H100 accelerator's Transformer Engine is built to speed up these networks as much as 6x versus the previous generation without losing accuracy.”
This is less and less a GPU doing AI, and more and more an AI chip that can also do graphics (see also: Tensor cores).
Nvidia also introduced the new HGX H100 where 4 or 8 of these H100 modules are on the same server board. The numbers are just ridiculous… 32 petaflops at 8-bit precision! 640 billion transistors in a box, if you include all the networking and IO chips!
That’s *one* board. Imagine a data-center filled with racks and racks of these.
In fact, Nvidia is doing exactly that, building themselves a new supercomputer with 18 DGX Pods, adding up to 18 EXAflops of computing power at 8-bit precision (9 exaflops at 16-bit) and 230 terabits/second of bandwidth.
They also upgraded their Grace ARM CPU before it’s even on the market, announcing Grace Hopper, which is a combo of Grace with the new GPUs, and the Grace Superchip, which is two Grace CPUs stitched together for a total of 144 CPU cores 🤯
This surprised me, but it’s a good move: They’re opening their 4th generation NVLink Chip2Chip IP to third parties, so that they can integrate their chips with Nvidia’s various families of chips. Very clever.
This one must’ve hurt just a bit (context: Nvidia was recently hacked):
Jensen: Every company must move to a zero trust architecture. Nvidia can for sure use Mopheus [their new cybersecurity product]
They’re really pushing hard on Omniverse, their “real-time reference development platform for 3D simulation and design collaboration”, which dozens of other companies already interface with (Autodesk, Adobe, Blender, Renderman, Ansys, etc.
If you don’t have a computer with an RTX GPU, you can even stream Omniverse from GeForce Now (their game-streaming platform)! And every animation and demo in the Keynote was created on Omniverse, so it’s a nice showcase of its capabilities.
They went as far as to create special data-center hardware to host digital twin simulations with synchronous computing (“OVX will enable designers, engineers and planners to build physically accurate digital twins of buildings or create massive, true-to-reality simulated environments with precise time synchronization across physical and virtual worlds.”).
Around 1h25min into the keynote, they show how Amazon is using Omniverse to help manage its warehouses and robot fleets, including optimizing building design and layout for maximum efficiency. Very cool stuff.
I’m kind of running out of space for their autonomous vehicle and industrial robotics announcements, but suffice to say, cool stuff. The big inflection points there are probably a few years out, but it makes sense for them to invest a ton in the space, because it’s clearly going in the direction of AI-in-everything, and Nvidia could be one of the big winners, as few of the car or industrial companies have the expertise to do it themselves and few big techs are as good at making full-stack platforms.
Great event. Makes me wonder what they’d have announced if they had been able to buy ARM, but alas, that’s only in some alternate timeline… ¯\_(ツ)_/¯
🇹🇼 Taiwan’s Semiconductors Labor Shortage
In the fourth quarter of 2021, there were close to 34,000 chip industry job openings a month on average on 104 Job Bank, a popular Taiwan recruitment platform, about 50 percent more than a year earlier
Although demand for workers has soared, Taiwan - with one of the world's lowest birth rates - has been producing fewer engineers over the past decade. Even fewer are enrolling in the doctoral programmes that best prepare engineers to develop breakthrough technologies.
Taiwan can no longer sustain local industry's needs, and foreign competition will further affect its long-term R&D talent development, said chip designer MediaTek, which this year plans to recruit more than 2,000 R&D employees and double its number of summer interns to lock in talent earlier. (Source)
🐶 Datadog, Complexity & Scarcity Edition
Olivier Pomel on some important trends shaping the industry:
if you look at the way infrastructures are evolving, they're getting more and more complicated. There's more complexity, there are more moving pieces, they move faster. There's -- a single developer is going to manipulate more and more and more different components. [...]
Then the last major trend I would mention on this is that there are just not enough software engineers to hire for everyone, and this is not going away. It's getting worse, actually. The world is transforming. More and more is built in software. There are not enough software engineers, and so they need more help.
They need more tooling. They need more ways to help them understand what's going on and extend their reach, and so we're going to see more and more of that.
I think the pendulum is going to swing more and more on buying versus building just because companies in general, not customers, in particular, will need to redirect their engineering resources to where it can make a difference for them in terms of their products, their offering, serving their own customers, differentiating themselves. So we definitely think that it's going to put more pressure on DIY and build-it-yourself initiatives in the future.
Cadence Software: Making Crucial Tools for Semiconductor Design
If you’re new to the semiconductors industry, this is a great intro and overview, as they really give a lot of context and history on most of the major players and business models.
Great job of going 101 on what is a pretty complex topic — kudos.
Science & Technology
🔭 🛰 ‘First images from James Webb exceed all expectations’ 🌌
Having completed the self-assembly of its 18-segmented main mirror, the telescope has now taken exceptional images of an unexceptional star as a test of its capabilities. The star, known as HD84406, is 100 times fainter than what can be seen with the human eye. The star itself is of little interest, lovely though its image is – instead, astronomers are captivated by the spray of tiny dots scattered across the background. Each is a distant galaxy, and this is the first time we’ve ever been able to capture them. [...]
“I’m happy to say that the optical performance of the telescope is absolutely phenomenal, it is really working extremely well. The performance is as good if not better than our most optimistic prediction.” (Source)
I always love it when scientists and engineers are this happy.
Looking forward to all the new things we get to learn about our universe thanks to this extraordinary piece of equipment.
🏴☠️ Crowdstrike Falcon X Recon — Checking the Dark Web for Threats & Leaked Data
Basically, you or Crowdstrike (depending on what package you pay for) can use it to scan the ‘dark web’ to see if there are threats against your company, or if some of your leaked data is for sale on the black market.
That’s some deep undercover, Donnie Brasco-type stuff right there! 🥸
‘Hacked US companies to face new reporting requirements’
Speaking of cyber-security, get ready for a lot more headlines about ‘companies critical to U.S. national interests’ getting hacked, because it looks like it’ll be law in the U.S. to disclose breaches (hopefully these are then made public, but that’s not part of the law, I think):
The reporting requirement legislation was approved by the House and the Senate on Thursday and is expected to be signed into law by President Joe Biden soon. It requires any entity that’s considered part of the nation’s critical infrastructure, which includes the finance, transportation and energy sectors, to report any “substantial cyber incident” to the government within three days and any ransomware payment made within 24 hours. (Source)
Not a bad way to put further pressure on these companies to improve their defensive posture. It should be enough just to want to avoid breaches in the first place, but we all know that fear of public humiliation is also a powerful motivator for the C-suite.
The Arts & History
Just really clever statues
Absolute $ military expenditure, 2020
In constant 2019 dollars. I think it’s useful to look at the numbers once in a while, because otherwise, if we only do “word thinking” about all this, it can be easy to forget the relative scales. Source.