672 karma · joined November 15, 2019
There's a market for both, but the furniture slop of Ikea is dominant.
This is true, but only for the bigger players. The nature of hardware still fundamentally favors scale and centralization. Every hyper-scalar eventually gets to a size that developing in-house CPU talent is just straight up better (Qcom and Ventana + Nuvia, Meta and Rivos, Google's been building their own team, Nvidia and Vera-Rubin, God help Microsoft though). This does not bode well for RISC-V companies, who are just being used as a stepping stone. See Anthropic, who does currently license but is rumored to develop their own in-house talent [1].
> Extensibility powers technology innovation
>> While this flexibility could cause problems for the software ecosystem...
"While" is doing some incredible heavy lifting. It is not enough to be able to run Ubuntu, as may be sufficient for embedded applications, but to also be fast. Thusly, there are many hardcoded software optimizations just for a CPU, let alone ARM or x86. For RISC-V? Good luck coding up every permutation of an extension that exists, and even if it's lumped as RVA23, good luck parsing through 100 different "performance optimization manuals" from 100 different companies.
> How mature is the software ecosystem?
10 years ago, when RISC-V was invented, the founders said 20 years. 10 years later, I say 30 years.
The nature of hardware as well, is that the competition (ARM) is not stationary as well. The reason for ARM's dominance now is the failure of Intel, and the strong-arming of Apple.
I have worked in and on RISC-V chips for a number of years, and while I am still a believer that it is the theoretical end state, my estimates just feel like they're getting longer and longer.
[1]: https://www.reuters.com/business/anthropic-weighs-building-i...
I often wonder how exactly you'd mitigate this. Further, as a user, I wonder what incentive there is for me to write anything at all online, let alone commenting on forums, if it will just be fed back into an LLM.
Is paywalling or forcing user accounts the solution? That feels antithetical to the reason for the internet at all.
Just musings.
Lip Bu Tan, for instance, has performance targets on a five year timeline, which are all negated if the stock falls below a certain threshhold in 3 years. [1]
Or, ever controversial Elon Musk, certainly has an (also egregious) $1 Trillion dollar pay package, but it has some pretty extreme goals over 10 years, such as shipping 1 million Optimus robots [2].
All in all, we can debate about the Goodharting of these metrics (as Musk is keen to do), but I feel boards of these public companies are trying to make more long-term plans, or at least moving away from tying goals to pure quarterly metrics. Perhaps we can argue about the execution of them.
Note: I own neither of these stocks and my only vested interest is buying the S&P.
[1] https://www.cnbc.com/2025/03/14/new-intel-ceo-lip-bu-tan-to-... [2] https://www.bbc.com/news/articles/cwyk6kvyxvzo
This makes some intuitive sense to me. Money provides freedom, but freedom is not happiness. It's freedom to explore, which is quite scary in and of itself. But if your mindset is proper, your creativity and appetite for exploration will be unbounded.
1.6 Bil in 1969 dollars translates to 13.6 Bil in 2024 dollars.
Oftentimes I hear complaints that today's projects cost too much, or I come across ballot measures where the other side is always like "something something we have no moneys" such as Prop 4 in California [1]. Sometimes reading about the past puts the present in perspective.
[1] - https://ballotpedia.org/California_Proposition_4,_Parks,_Env...
It's really hard to describe what the book's about. It's an epic, through and through, and all epics are hard to detail precisely. Inter-generational trauma? Handling one's "sin"? Making a livelihood after repeated failure, be it yourself or external factors?
Contrary to my first sentence, there is one character that I would describe as pure evil. But I feel that just supports one of the conflicts; however incredibly rare, what can an individual do when they come across a bonafide force of evil?
It is dripping with Biblical imagery, and Steinbeck's prose is rambling and tangential for some (though poetic for me), and his characters are not "realistic" and larger than life (but that's what makes them pop off the page for me and so memorable. I guess it's always a balance).
"Now that you don't have to be perfect, you can be good."
When you had kids (or I guess anyone who is considering having kids), did you ever fear of "passing down" your trauma?
I feel that us tech workers, for all intents and purposes, have "made it," which might best position us in general to overcome and prevent a cycle of trauma to the next generation.
My main barriers are 1) writing discipline and 2) some obsessive need to link/cite every claim I make. This results in me taking months to write an article, be it due to laziness or constant rabbitholes/google scholar searches about an idea.
Any clue on dedicated disciplined writing time or even considering "reducing" the rigor of writing?
Sure, anybody can make a RISC-V CPU, but who really has the capabilities to verify them?
There's a reason the ARM model has succeeded-- that is, providing totally off-the-shelf IP with pre-verified cores (because of their own large verif team). The logical end of RISC-V is that we have custom cores literally everywhere, but verifying them is quite costly.
The (equally) hard part with CPU design is funnily enough not in creating the design, but the verification. (That's kinda one small reason why I think CoPilot-esque tools haven't permeated the hardware design space very much).
• Lindsay Lohan
• Jake Paul
• DeAndre Cortez Way (Soulja Boy)
• Austin Mahone
• Michele Mason (Kendra Lust)
• Miles Parks McCollum (Lil Yachty)
• Shaffer Smith (Ne-Yo)
• Aliaune Thiam (Akon)
I had to doubletake. Never did I ever think I'd see this permutation of celebrities in an SEC report.
Applied to Chat-GPT, a charitable take on this self-aggrandizement would be that the speaker has requires deep knowledge on the model they're attacking, in the same way a reverse engineer generally knows how X system is built. But I'm just being nice.
[1] https://proceedings.neurips.cc/paper/2019/file/7fea637fd6d02...
I like to draw parallels with the '90s and all of the Pentium bugs. These events really really lit a fire under Intel's ass to get their crap working. Cue decades worth of research and $$$ into formal verif, EDA tooling, and perhaps most importantly, organizational ownership.
Software models, where many security paradigms have been explored decently, don't map too well or don't have a great solution, like fuzzing or linters. So it's kinda hard to provide various security guarantees if you don't have tooling to verify them. Plus, the tools that do exist require domain expertise from the engineer, so the thought to look for some MMIO exploit will never occur to a cache designer.
This kinda turned into a ramble, but ultimately, while I agree I wish Intel would do better, I can't fully fault them since I feel like the whole industry is ill-equipped to provide these security guarantees. Spectre has started some efforts, but it's slow to pickup, which is why it's rather patchwork-y still. Frankly, it kinda feels like the only thing that'll kick asses into gear is some lawsuits, F00F and FDIV style.
P.S. A couple questions--
1) Would you happen to have the "EGX" framework you wrote in Appendix A? That seems massively useful.
2) Can I just screenshot the rubber duck NFT? :)
This assumes you'd anticipate rate cuts in the near-ish future. This property tax arbitration could easily go wrong.
One might then ask, do higher-class wages match/exceed inflation more than lower-class wages? This is debatable. But here's some interesting data from the Atlanta FED, where you can track wages by quintile, education, "skill", hourly vs. non-hourly, etc. Make your own conclusions.
Third Q-- What's your opinion on BTI as a possible mitigation? Given it's an v8.5 feature meant for JOPs, and this attack is essentially a speculative JOP, maybe we could use BTI to mitigate and heavily reduce the number of gadgets, speculative or not.
2 questions.
1) it's relatively known that PAC is brute-forcable given its relatively small key space (16 bits, sometimes 8 if TBI is enabled). How does your attack differ from general brute forces? (My impression is just your leveraging of the BTB/iTLB is a bit more stealthy.) Similarly, in your opinion, would a fix be more ISA-level or you think it's more specific to the M1 (given brute forcing in general is a PtrAuth problem)?
2) you mention in section 8 that this took 3 minutes for a 16b key and tons of syscalls. Wouldn't another proper mitigation be to limit the number of signatures per key? 3 minutes is definitely a long time, and some form of temporal separation may be quite helpful.
I would have added that as a potential mitigation in the mitigations section. I think, say, changing the key every so often would be a reasonable task for a kernel to do, especially in the timeframe that this was exploited (about 3 minutes)
Directly from the article.
> Qualcomm would be interested in forming a consortium of sorts to acquire ARM. “We’re an interested party in investing,”
Relatedly, the CEO has mentioned that ARM going public would make him buy a bunch of shares, given his faith in the continued success of ARM. As is the case of any "bull case".
Forming a consortium or buying public shares is not, as the title implies, a singular company buying ARM.
I use Q2 '22 specifically because it's a recent quarter that's most likely to reflect the vision of the current CEO. Given that my thesis is "current Qualcomm is different", I think it's apropos to talk about current results. And my yearly sampling is based off of the assumption that the general business cycle is periodic YoY (i.e. Q2 2020 is roughly similar in growth rates to Q2 2019, barring anything massive like a pandemic haha lol haha).
Furthermore, I also think it's important to look at breakdowns of % of revenues by division, to see how much a company relies on any specific source.
---- | QTL Rev | QTL EBT | QTL % as non-GAAP Rev
FY13 | $7,554 | $6,590 | 69%
-------------------------------
FY14 | $7,569 | $6,590 | 65%
-------------------------------
FY15 | $7,947 | $6,882 | 74%
-------------------------------
FY16 | $7,664 | $6,528 | 80%
-------------------------------
FY17 | $6,445 | $5,175 | 69%
-------------------------------
FY18 | $5,163 | $3,525 | 64%
-------------------------------
FY19 | $4,591 | $2,954 | 58%
-------------------------------
FY20 | $5,028 | $3,442 | 50%
-------------------------------
FY21 | $6,320 | $4,627 | 34%
The pie was rather small back then, and the licensing slice was HUGE. Nowadays, the exact quantity of licensing revenue is actually just reverting to the mean, and the slice is actually decreasing. You can't go by the growth rates you're given there.
Links to FY13--FY19 here. BTW you can skip every other year, since FYXX tends to report FY(XX-1) as well.
https://investor.qualcomm.com/financial-information/historic...
QTL in Q2 '13 had $1800mil EBT with 88% of revenue contributing to earnings. EIGHTY EIGHT PERCENT. Holy cow. QCT in that same year had $600mil EBT with 17%.
For comparison, QTL in Q2 '22 had $1540mil EBT with 74% of revenue contributing to earnings. QCT in that same year had $3340mil EBT with 35%.
I guess really you can only squeeze so much before the market gets annoyed. Given these numbers, I can't see how QCT would be a loss-leader. Why else would the margins on them be increasing, and margins on QTL be decreasing?
My point is, I believe things aren't as lawyer-y as before. The Qualcomm of today focuses a lot more on its QCT division. Licensing is still obviously there, and will always be an incredibly nice moat, but it's not the suffocating "shakedown" it might have been 10 years ago.
I'm curious how long ago this was? I can't find any time period where QTL (the licensing division) made more than QCT (the Snapdragon et. al division). Still, I think your statement's sentiment is the consensus-- that overall, Qualcomm is/has been the "Oracle of hardware" that milks the licensing cow.
Let's look at Q2 '22, Q2 '19, Q2 '16, Q2 '13. I choose 3-year gaps for no particular reason, simply how I personally view company timelines and how long it might take to "steer a ship", e.g. if the CEO wants to change its public perception, it might take a few years. I also choose Q2 simply because Q2 was the most recent quarter. Relatively arbitrary, but still sensible data points when talking about Qualcomm's recent history. Noteworthy also, is that Q2 '22 is about the 1 year anniversary of when the new CEO was instantiated.
($mil) | - QCT - | - QTL -- | QTL % of tot rev
--------------------------------------------
Q2 '13 | $3,916 | $2,057 | 33.6%
--------------------------------------------
Q2 '16 | $3,337 | $2,135 | 38.5%
--------------------------------------------
Q2 '19 | $3,722 | $1,122 | 22.9%
--------------------------------------------
Q2 '22 | $9,548 | $1,580 | 16.5%
Licensing revenue has actually been flat or on the decline recently. Even in business outlooks, Qualcomm usually directly says "our goal is to maintain QTL" while heavily expanding on QCT, somewhat implying an attempt to cast off this "ran by lawyers" reputation.
Time will tell, I suppose.
[1] https://d1io3yog0oux5.cloudfront.net/_ef937ef9314613934ed123...
[2] https://d1io3yog0oux5.cloudfront.net/_ef937ef9314613934ed123...
[3] https://d1io3yog0oux5.cloudfront.net/_ef937ef9314613934ed123...
[4] https://d1io3yog0oux5.cloudfront.net/_ef937ef9314613934ed123...
> [I] would like to see is increased risk-taking in the workplace [...] I want to blow up the scope [1], but they won't allow me
While I generally agree with the sentiment, let me offer a devil's advocate position to counter the whole "woe is me, I am but a 10x engineer suffocated by the evil VCs and PMs of the world who keep asking for dumb jira updates. If they stopped asking for updates, I'd give it to them in 5 years" narrative.
There are similar criticisms in academia-- "The bean-counting administrators need some dumb metrics (num papers, num citations) for tenureship". While these are valid criticisms, the alternative is not one I'd prefer.
Consider someone who does make grandiose claims, but pushed back against those who ask for "updates" or "metrics". Elizabeth Holmes is cited as one such person with "[going] all-in; mega bold and despite the odds stacked against them". But this example actually is a warning that goes against the point of the article, I feel. The "evil VCs" full-heartedly trusted her. Are you telling me that if the VCs just took more risk, she would have succeeded? But instead, they basically margin called her and screwed her over because they were too risk-intolerant?
W.r.t Holmes, I actually place a small amount of fault on her. It's good to dream big. I place more fault on the VCs for not being "bean-countery" enough, because anybody with proper auditing would be able to see the issues there, instead of blindly trusting "oh this $100mil is for R&D, just trust me. You need to take risks for this world changing blood testing methodology."
The takeaway is-- there's a reason scope might be restricted. In academia, there's a notion of a "publon" [2], the smallest "quantum" of knowledge that can be published (ha! I love that name). The idea is, no matter how large a project might be, it is segmentable, and if it is not segmentable, it's not a good project. Furthermore, publishing is THE mechanism of which scientists get feedback, and feedback is critical to our iteration process.
(Again, as I mentioned, there are tons of problems with the "publish or perish" culture, which I could also go on forever about, but preferable IMO to the alternative of publishing a magnus opus every 10 years).
We can generalize that concept to most things. Feedback and metrics are critical inputs to iteration. You can certainly criticize these metrics, but you can't JUST say "take more risk" without providing alternative data to update a VC's investing strategy. You want them to accept your giant ass scope? Give them better data. Help them help you.
For another example-- Consider a project that's been going on for almost 20 years. Mill Computing [3]. They have established some pretty massive scope, and it's a goal I support. But do we (the public) have much insight as to what they're doing, precisely? Not really. They most definitely have their reasons, I'm sure, but I, as an outsider, know very little about how they are achieving their goal. So then why would I invest in them?
[1]: https://timdaub.github.io/2021/06/18/when-scope-blows-up/