2
points
peter_d_sherman
Joined 17,424 karma
Programmer, Student Of Law, Entrepreneur & Comedy Writer.
Open Source, Open Hardware, Transparency & Free Speech enthusiast.
peter.d.sherman@gmail.com www.linkedin.com/in/peter-sherman-a6107a5 https://twitter.com/peter_d_sherman
"The true knowledge consists of knowing that one knows nothing..."
- Socrates
- We know, minimally, from Perplexity's list above, that there exist multiple alternative DRAM manufacturers other than Micron, Samsung and SK Hynix, that have completed at least some of the barriers of entry to the high-end DRAM market, and possibly many...
We also know that the world is full of capital -- as you suggest.
That capital is continually looking for investment opportunities, and DRAM is a huge, huge market...
When capital invests in markets, any barriers to entry are moved, if not outright displaced (i.e., OpenAI, $1 billion, etc.).
Point is, we don't know what the future will hold...
I think it's a good bet that cheaper DRAM, DDR5 and otherwise, will be coming down the pike soon, once production catches up, once supply outpaces demand...
- That's true.
Coke and Pepsi dominate the worldwide drink market, but due to the immense size of the market, there are always up-and-coming competitors.
Go to your local superstore, supermarket, or your local convenience store.
You'll find Coke and Pepsi, lots of it, but you'll also find no-name drinks and sodas from drink companies that are not as well established yet or well known.
That those exist is a good thing, at least for consumers, at least for those that consume, because the root of all consumer prosperity brought on by capitalism (global trade = capitalism, regardless of the names of countries involved) comes from the relentless competition brought on by two or more companies, ideally as many as possible...
We would not have the super high performing desktop computers we have today if it were not for the historic early competition between AMD and Intel (later entered by other CPU manufacturers), and we would not have choice if it were not for competition.
Getting back to DRAM manufacturers, The first three do dominate 95+% of the market as of 2026.
But there might be some interesting up-and-coming smaller companies to watch...
Let's remember that OpenAI came from basically nowhere -- to give Google a run for its money -- as did Google with Microsoft's behemoth 20+ years ago...
What new DRAM manufacturer might be the next up-and-coming DRAM manufacturer in the space?
Well, we don't as-of-yet know... but the space is an interesting one to watch, to be sure!
But you are correct!
The first three do currently dominate 95+% of the market as of 2026...
- >"Using just two components — a
high-speed [1N5711] Schottky diode
and an LED — you can build a tiny “crystal detector” that responds to
2.4 GHz
Wi-Fi, Bluetooth, and even microwave..."
That is interesting!
I never knew that a Schottky diode could rectify at 2.4 GHz -- that's pretty darn impressive!
- I asked Perplexity.ai to list all of the DRAM manufacturing companies that it could. What follows is that list:
Samsung Electronics – https://www.samsung.com
SK hynix – https://www.skhynix.com
Micron Technology – https://www.micron.com
ChangXin Memory Technologies (CXMT) – https://www.cxmt.com
Nanya Technology – https://www.nanya.com
Winbond Electronics – https://www.winbond.com
Powerchip Semiconductor Manufacturing Corp. (PSMC) – https://www.psmc.com.tw
Fujian Jinhua Integrated Circuit (JHICC) – http://www.jhicc.com
GigaDevice – https://www.gigadevice.com
Etron Technology – https://www.etron.com
Integrated Silicon Solution Inc. (ISSI) – https://www.issi.com
Elite Semiconductor Memory Technology (ESMT) – https://www.esmt.com.tw
Zentel Electronics – https://www.zentel.com.tw
Alliance Memory – https://www.alliancememory.com
AP Memory Technology – https://www.apmemory.com
AMIC Technology – https://www.amictechnology.com
Hua Hong Semiconductor – https://www.huahong.com
(Note that I have not researched each of these companies individually... There may be errors in the above list (some may be DRAM resellers, some may be defunct, etc., etc.))
- 45 points
- >"Later this year, Point2 will begin manufacturing the chips behind a 1.6-terabit-per-second cable consisting of eight slender polymer waveguides, each capable of carrying 448 gigabits per second using two frequencies, 90 gigahertz and 225 GHz. At each end of the waveguide are plug-in modules that turn electronic bits into modulated radio waves and back again. AttoTude is planning essentially the same thing, but at terahertz frequencies and with a different kind of svelte, flexible cable.
Both companies say their technologies can easily outdo copper in reach—spanning 10 to 20 meters without significant loss"
This is absolutely fascinating! For the longest time, I thought that optical fibers were the future, but waveguides (of whatever material appropriate) at whatever frequenc(y|ies) appropriate could give optical fibers a run (get it, a "run"? :-) ) for the money!
If we think about it, both fiber and copper cables are both very specific cases of a more broader
waveguide (first) principle...
That is, in theory you could make something that looks like a wire or cable out of any material(s) -- and if the material(s) and apertures and frequencies are correct, then you've created a transmission of path for data from point A to point B...
So, kudos to Point2, AttoTude (and other future companies!) that go down this technological tract! You're increasing both human knowledge (and data rates!) -- which could never be a bad thing!
- Brilliant!
This may just become my next most favorite project on GitHub!
For anyone who would create their own OS, or just experiment with other OS'es, this could be a godsend!
The set of ideas which gives rise to this tool are brilliant, and while I haven't reviewed all of the code for potential security implications (as I would want to if I were deploying it to a production server in a business environment) -- it looks very well thought out at first glance!
Extra kudos for having a flake.nix (for us Nix users!)
(If you're using NixOS or the Nix package manager, you can download it here https://search.nixos.org/packages?channel=25.11&query=quicke... , i.e., "$ nix-shell -p quickemu")
And extra extra kudos for having Alpine, Nix, ReactOS, TinyCore and OpenBSD as downloadable OS choices!
In the future, I'd love to see Windows XP, Windows 2000, and Windows NT too (assuming that Microsoft would permit that!) -- but that would just be the icing on the cake!
Short review: There's potentially something for everyone here! (Well, any OS person! Could Minix 3 be added in the future? :-) )
Long review: Will definitely have to watch this project in the future, to see where it goes!
- >"We also have a contract with AMD to get MI350X on MLPerf for Llama 405B training."
Anything to help AMD (and potentially other GPU/NPU/IPU etc. chip makers) catch up with NVidia/CUDA is potentially worth money, potentially worth a lot of money, potentially worth up to Billion$...
Why?
If we have
a) Market worth Billion$
and
b) A competitive race in that Market...
then
c) We have VALUE in anything (product, service, ?, ???) that helps any given participant capture more of that market than their competitors...
(AMD (and the other lesser known GPU/NPU/IPU etc. chip vendors) are currently lagging behind NVidia's CUDA AI market dominance -- so anything that helps the others advance in this area should, generally speaking, be beneficial for all technology users in general, and be potentially profitable (if the correct deals could be struck!) by those that have the skills to do such assisting...)
Anyway, wishing you well in your endeavors, Tinygrad!
- 2 points
- This is by far the absolute best "how things go from a high-level (AI/LLM/Inference/Python/JAX/etc.) to low-level (TPU equivalent of assembly language instructions)" article that I've read to date!
Upvoted and favorited!
A keeper -- for present and future study...
Well done!
- 3 points
- 1 point
- >"The question isn't just why Nvidia paid $13.1B more than market rate for technology they could build themselves (they have the PDK, volume, talent, infrastructure, and cash). The question is why they structured it this way.
Where the premium was spent:
Regulatory arbitrage:
Non-exclusive licensing avoids years of antitrust review. Structure the deal as IP licensing + talent acquisition, and regulators have no grounds to block it.
This alone is worth billions in time and certainty."
Isn't that fascinating!
Observation: For any given Deal (in business or in life in general) -- there may be one or more legal components -- to it...
But (equal-and-oppositely!) there also may NOT be one or more legal components to it!
What each deal has in some legal components -- it may lack in other legal components...
Conversely, what each deal does not have in some legal components -- it may have in other legal components...
Now, perhaps this may sound like a "self-evident truth", and as such, apparently may not be worthy of a deeper exploration, but it seems that there exists an:
Intersection of Set Theory and Legal Aspects -- applied to Deals
(AKA "Transactions", "Exchanges", "Barters", "Trades", "Exchanges Of Value", etc., etc.) in various jurisdictions (which can be thought about as how contracts, both legal and natural, arising from such exchanges are, or would be interpreted through its courts, through its regional statutes (aka "Laws") IF there are inter-party disputes which subsequently require a court for such interpretation...)
And that intersection -- could well be worthy of further study!
Phrased another way -- it (and this article!) are highly interesting from a legal perspective!
(And also a Set Theory / Set Theoretical one!)
- Interesting, and very cool!
- Very cool!
A few questions:
How was the 696 MHz DDR3 RAM speed determined?
Also, how is the system initialized at startup? Is there boot/startup code, does that exist in a flash chip (or some other ROM or EEPROM or flash type chip) on the system, and can the boot code be inspected/customized/overwritten by the end-user? How does the system boot?
Anyway, looks very cool!
- >"To address these issues, the research team adopted a dual strategy combining a Reversible Host (RH) and a Designed Electrolyte (DEL). The reversible host consists of a polymer framework embedded with uniformly distributed silver (Ag) nanoparticles,
guiding lithium to deposit in designated locations rather than randomly
. In simple terms, it acts like a dedicated parking lot for lithium, ensuring ordered and uniform deposition."
(I'm wondering if some process like this -- might one day replace (or supplement) photolithography for creating chips/IC's...)
- Related:
"What is a Language Processing Unit (LPU)?":
https://groq.com/blog/the-groq-lpu-explained
>"Groq on-chip SRAM has memory bandwidth upwards of
80 terabytes/second
, while GPU off-chip HBM clocks in at about eight terabytes/second. That difference alone gives LPUs up to a 10X speed advantage, on top of the boost LPUs get from not having to go back and forth to a separate memory chip to retrieve data."
- 2 points
- 55 points
- The following passage seems particularly noteworthy:
>"Sec. 3. AI Litigation Task Force. Within 30 days of the date of this order, the Attorney General shall establish an AI Litigation Task Force (Task Force) whose sole responsibility shall be to challenge State AI laws inconsistent with the policy set forth in section 2 of this order, including on grounds that such laws unconstitutionally regulate interstate commerce, are preempted by existing Federal regulations, or are otherwise unlawful in the Attorney General’s judgment, including, if appropriate, those laws identified pursuant to section 4 of this order. The Task Force shall consult from time to time with the Special Advisor for AI and Crypto, the Assistant to the President for Science and Technology, the Assistant to the President for Economic Policy, and the Assistant to the President and Counsel to the President regarding the emergence of specific State AI laws that warrant challenge."
It would seem logical to believe that there will be a number of AI-meets-law legal cases in the future, both in the U.S., it's States, and in the jurisdictions of foreign countries and their respective States/Districts/Regions...
I'm guessing (but not knowing) that the U.N. will have its own similar task force in the future -- as will other countries and their jurisdictional / law-making regions...
It will be interesting (at least from the perspective of a disinterested-in-outcome-but-interested-in-process legal observer) to see what cases (and also what laws/statutes) emerge in this area (Region Vs. Nation, Nation Vs. Region, Nation Vs. Nation, Region Vs. Region) in the future, and how they will be resolved...
(You know, for students of AI, students of Law, and students of The Future...)
- 53 points
- 3 points
- Interesting: >"There is no single, universally agreed-upon definition of microservices."
- 1 point
Observation: One of the great virtues of Python is that typically when someone runs a Python program, they are running a Python interpreter on transparent Python source code files, which means that typically, the person that runs a Python program has the entirety of the source code of that program!
But that's not the case here!
.pyc, aka "pre-compiled, pre-tokenized, pre-digested" aka "obfuscated" python -- one of the roots of this problem -- is both a blessing and a curse!
It's a blessing because it allows Python code to have different interpretation/compilation/"digestion" stages cached -- which allows the Python code to run faster -- a very definite blessing!
But it's also (equal-and-oppositely!) -- a curse!
It's a curse because as the author of this article noted above, it allows Python codebases to be obfuscated -- in whole or in parts!
Of course, this is equally true of any compiled language -- for example with C code, one typically runs the compiled binaries, and compiled binaries are obfuscated/non-transparent by their nature. And that's equally true of any compiled language. So this is nothing new!
Now, I am not a Python expert. But maybe there's a Python interpreter switch which says 'do not run any pre-digested cached/obfuscated code in .pyc directories, and stop the run and emit an error message if any are encountered'.
I know there's a Python switch to prevent the compilation of source code into .pyc directories. Of course, the problem with this approach is that code typically runs slower...
So, what's the solution? Well, pre-created (downloaded) .pyc directories where the corresponding Python source code is not provided are sort of like the equivalent of "binary blobs" aka "binary black boxes" that ship with proprietary software.
Of course, some software publishers that do not believe in Open-source / transparent software might argue that such binary blobs protect their intellectual property... and if there's a huge amount of capital investment necessary to produce a piece of software, then such arguments are not 100% entirely wrong...
Getting back to Python (or more broadly, any interpreted language that has the same pre-compilation/obfuscation capability), what I'd love to see is a runtime switch, maybe we'd call it something like '-t' or '-transparent' or something like that, where if passed to the interpreter prior to running a program, then if it encounters a .pyc (or equivalent, for whatever format that language uses, call it "pre-tokenized", "pre-parsed", "pre-compiled" or whatever), then it immediately stops execution, and reports an error where the exact line and line number, the exact place where the code in the last good source file which called it, is reported, exactly to the end-user, and then execution completely stops!
This would allow easy discovery of such "black box" / "binary blob" non-transparent dependencies!
(Note to Future Self: Put this feature in any future interpreters you write... :-) )