Comments URL: https://news.ycombinator.com/item?id=45186561

Points: 3

# Comments: 0

The bleeding edge uses abstract interpretation to verify code.

Free from NASA: IKOS (Inference Kernel for Open Static Analyzers) is a static analyzer for C/C++ based on the theory of Abstract Interpretation. https://github.com/NASA-SW-VnV/ikos

Commercial: Astrée is a static analyzer for safety-critical software written or gen­er­ated in C or C++. https://www.absint.com/astree/index.htm https://www.absint.com/astree/index.htm

abstract interpretation for static analysis and verification:

Good intro Mechanized semantics, fifth lecture Abstract art: static analysis by abstract interpretation https://xavierleroy.org/CdF/2019-2020/5.pdf

Static Analysis and Verification of Aerospace Software by Abstract Interpretation https://mine.perso.lip6.fr/publi/article-bertrane-al-fntpl15...

Everyone working with the latest process node is tightly coupled with the manufacturer: Nvidia, Apple, AMD, and Qualcomm. Microarchitecture design and optimization for a new process is super expensive and hard.

Legacy processes have settled, have standardized design software and tooling, so everything costs less.

ps. Nvidia and Qualcomm are considering Samsungs 2mm because TSMC's 2-nanometer chip production capacity is extremely limited.

Comments URL: https://news.ycombinator.com/item?id=44593628

Points: 83

# Comments: 3

If you want free software I recommend IKOS - a is a sound static analyzer for C/C++ developed at NASA. Checks: https://github.com/NASA-SW-VnV/ikos/blob/master/analyzer/REA... Numerical abstract domains: https://github.com/NASA-SW-VnV/ikos/blob/master/analyzer/REA...

Commercial tool like Astree https://www.absint.com/astree/index.htm if you have money.

APNewsBreak: US suspects cellphone spying devices in DC https://apnews.com/general-news-d716aac4ad744b4cae3c6b13dce1...

Washington, D.C. mobile traffic is probably the most spied in the world. Especially now when it's run by technological cavemen and overly confident techbros. Israeli, Russians, Chinese, French and everyone.

What a naive fantasy.

Organization of people is much more important than guns. You don't even need guns when you organize. You can stop the state just by collective action. See color revolutions.

When it is guns, you need RGP's, detonators and TNT (and drones), a good underground insurgent logistic chains. You also need commit to life in poverty and eventual death.

United States, Hungary, Turkey, Russia, Serbia have plenty of independent weapons, yet there is no fear of effective armed resistance. Everybody is a rebel in the Internet. When things go tough it's "I have to go to work and eat. My family needs me."

Even without spoke-and-hub model, hubs grow and there are more hubs. Airports are the bottleneck. 500 passengers per landing adds capacity.

While I strongly disagree with them, I feel an affinity and familiarity with their thinking. I have read the same books, seen the same news, and lived in the same era. I understand how they arrived where they are now.

The Neo-reactionary movement seems exactly like what my generation comes up from the right. Dark Psychology of Dark Enlightenment is a cyberpunk sci-fi world as a fantasy. To live with societal collapse, dystopia and decay with low-life and high-tech. Always framing oneself as an independent outsider and a rebel. Sarcasm as a reflex. These guys see themself living in William Gibson's Neuromancer world.

Their idea is to have 44 GB SRAM per chip. SRAM is _very_expensive_ compared to DRAM (about two orders of magnitude).

It's easy to design larger chip. What determines the price/performance ratio are things like

- performance per chip area.

- yield per chip area.

Comments URL: https://news.ycombinator.com/item?id=41706271

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=41706253

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=41219660

Points: 7

# Comments: 1

Speedup is realistic. Multicore SMP or NUMA are not good for memory access patterns they optimize.

Their thick control flow model that should work for exclusive matrix-addition and log-prefix style memory access patterns. In comparison to the baseline the speedup is 150% in log-prefix algorithm, over 190% in fft-style butterfly algorithm, 50-100% in matrix addition and threshold filtering. silicon area and power consumption are estimated to be low.

Light reading material:

Optimizing Memory Access in TCF Processors with Compute-Update Operations Optimizing Memory Access in TCF Processors with Compute-Update Operations https://ieeexplore.ieee.org/document/9150423

The REPLICA on-chip network https://ieeexplore.ieee.org/document/7792877/

Preliminary Performance and Memory Access Scalability Study of Thick Control Flow Processors https://ieeexplore.ieee.org/document/10305463/

Realizing multioperations and multiprefixes in Thick Control Flow processors https://linkinghub.elsevier.com/retrieve/pii/S01419331230005...

This is how most semiconductor startups work. Do better and show innovation that is bought buy bigger players.

As long as the bottleneck is the fab capacity as wafers per hous, the number of operations per second per chip area determines who will produce more compute with best price. It's a good measure even between different technology nodes and superchips.

Nvidia is leader for a reason.

If manufacturing capacity increases to match the demand in the future, FLOPS or TOPS per Watt may become relevant, but now it's fab capacity.