banjo_milkman

This is an interesting conjecture. It could be a cunning Microsoft competitive strategy.

They convince cost-sensitive IT infrastructure groups in other businesses to use Teams by offering it for free, reducing company productivity massively, hobbling product development.

Meanwhile, Microsoft adopt some other pay messaging/video service internally, accelerate product development 10x, become insanely competitive in a whole host of new markets rapidly.

Yes, I use an Onyx Note Air 2 for this purpose. It works nicely, much easier on the eyes than a laptop/Ipad. The Note Air 2 is fairly big - 10.3" - works well for books.

Illumina

They are talking about one 'reference genome'. The variation from human-to-human is relatively small (a few million bases out of 3 billion). The reference genome has historically been some kind of average/mosaic of several individuals (this has obvious disadvantages), good enough to put reads in the right place (mostly), and call 'variants' - the differences that make the test genome unique.

The latest/greatest end-to-end T2T reference is entirely based on 'HG002' an individual from Utah, due partly to new information derived from long read technologies.

Siddhartha Mukherjee's books 'The Gene', 'Emperor of all Maladies', 'The Song of the Cell' all mix biology and history with lots of detail and are easy to read.

The UK has had an industrial intervention policy for ages. Its undeclared but it basically involves supporting finance, real estate, business services and insurance in London above all other stakeholders in the country.

The BBC has some good history ones: In our Time, You're dead to me

The Ted Chiang books are great. Short stories. 'Arrival' the movie is based on one of them.

Liquid biopsy for early cancer detection.

It's not a documentary but the BBC drama 'Micro Men' is very good and AFAIK mostly historically accurate.

It's about the battle between Clive Sinclair (of Sinclair Spectrum fame) and Chris Curry (founder of Acorn Computers, the progenitor of ARM) for dominance in the 80's UK home computer market.

https://www.youtube.com/watch?v=XXBxV6-zamM https://www.bbc.co.uk/programmes/b00n5b92

Turbo codes and low density parity check codes are two classes of iteratively-decoded capacity approaching codes, amongst others.

Ironically, LDPC was actually developed in the 60s by Robert Gallagher, but were reinvented in the 90s when the decoding was practical.

https://en.wikipedia.org/wiki/Low-density_parity-check_code

Fiction:

Hamnet by Maggie O'Farrell. Very moving.

Piranesi by Susanna Clark. Like her earlier books, it had a strong story + original style + was very inventive.

Non-fiction:

The Great Leveller - Walter Scheidel

Sid Meier's biography snuck in after Christmas + I really enjoyed that one.

Raw FLOPs is completely misleading, which is why Nvidia focus on it as a metric. The GPU can't keep those ops active - particularly during inference when most of the data is fresh so caches don't help. It's the roofline model.

In my experience FPGA>GPU for inference, if you have people who can implement good FPGA designs. And inference is more common than training. Much of this is due to explicit memory management and more memory on FPGA.

10GBASE-T was actually standardised in 2006. PHY chips were available that year but reaching 100m was very difficult on unshielded cables - the chips were large, high power, expensive to make, 100m was only guaranteed if users upgraded the cables to Cat6A - and the solutions had relatively high latency due to the need for powerful error correction. After PHY vendors had 100m working - an enormous technical challenge - they were not inclined to release an 'easier'/cheaper 10m/30m version, since that would have enabled more competitors, reduced revenues and partitioned the market. But hardly anyone was deploying 10GBASE-T anywhere anyway. So the technology got stuck with low revenues/high prices supporting 100m reach.

Broadcom were making $$ from 1GBASET and were slow to develop/release 10GBASE-T, they held the market back and encouraged SFP+ since they did sell those chips. Intel/Cisco were inclined to wait for BRCM chips. The startups that developed 10GBASE-T (Solarflare initially led, then Teranetics, Aquantia emerged) were not successful quickly. Eventually Aquantia managed to partner with Intel & survived. Both Solarflare's PHY technology & Aquantia ended up being acquired by Marvell, Aquantia for significant $$ last year. Teranetics circuitously ended up as part of Broadcom. The rest of Solarflare was acquired by Xilinx.

10GBASE-T wasn't a good fit for data-centers due to the high power/latency. So those customers went direct attach / SFP+ / optical. Which drove those prices down, and made them more attractive, further delaying 10GBASE-T volumes. Data centers got used to expensive cables, (relatively) cheap/simple low-power, low-latency transceivers. 10GBASE-T was a solution looking for a problem.

Eventually the 2.5G/5G Ethernet for Wifi back-haul opened the copper market up - those technologies reuse almost everything from 10G. Also automotive Ethernet too. Chip/power scaling and increasing volumes for Wifi/ datacenter deployments has eventually driven down the cost to a point where 10GBASE-T is becoming more widespread/attractive.

The initial sales/marketing strategy for 10GBASE-T failed, sadly, and it has taken a long time to recover.

I think this is driven by partly embedded applications and more directly by the datacenter.

Nvidia already own the parallel compute/ML part of the datacenter and the Mellanox acquisition had brought the ability to compete in the networking part of the datacenter - but they were missing CPU IP, for tasks that aren't well matched to the GPU. This plugs that hole. They are in control of a complete data-center solution now.

There are multiple examples of revolutionary technology from the UK: 1) ARM - dominates embedded systems, >90% of cellphones and IoT, sold to SoftBank, now on the way to NVidia 2) DeepMind - arguably global leader in AI, now part of Google 3) Illumina/Solexa - >90% of global DNA sequencing market, biotech leader, sold to Illumina for $600, Solexa sequencing now the basis for a $50bn market cap

Britain has the problem that too much of the country is focused on finance + the services that surround it, the country bends over backwards to support those industries through deregulation/not enforcing rules (LIBOR scandal et al), intervening with foreign governments to protect banks (see HSBC 'too big to jail') while pretending it is lassez-faire & does not have an industrial policy.

The ironic thing is that the financial industry could help develop these seeds of technology into behemoths but have misguidedly focused on inflating land values + developing dodgy financial instruments.

Another problem is culture - many Brits have the ambition of making $ millions of money and then retiring to a country house. Not many Zuckerbergs have emerged from the UK.

The Great Leveler: Violence and the History of Inequality from the Stone Age to the Twenty-First Century

Seems moderately depressing so far.

if you don't test it then it doesn't work.

DARPA does have some interesting VLSI projects running currently. Some of the ones I've noticed are:

Chiplets: https://www.darpa.mil/program/common-heterogeneous-integrati...

PIPES - optical chip io : https://www.darpa.mil/news-events/2020-03-16

Automated ML chip generation: https://www.darpa.mil/attachments/Real%20Time%20Machine%20Le...

Try Murakami, Knausgård and Ursula Le Guin.

Hard-Boiled Wonderland and the end of the world is the most scifi Murakami. Wonderful stuff. Then read all his others. Knausgård 'My Struggle' is a very compelling personal story (5 volumes!) where you get inside his head. Not a novel. Ursula Le Guin - the Lathe of Heaven. More scifi.

On DNA/genetics: Life's Greatest Secret, Matthew Cobb Siddhartha Mukherjee's books: 'The Emperor of All Maladies' + The Gene are very good as an introduction (start off by watching the PBS documentary)

Jim Baggott's books are good on physics e.g. 'Mass'. Also Jim Holt, though he was trained as a philosopher. Penrose 'The Emperor's New Mind' etc are very good but not new.

Robert Caro's Lyndon Johnson biography (multiple volumes) (idolised by Conan O'Brien see https://www.nytimes.com/2019/04/17/style/conan-obrien-robert... - “I’m a fan to a disturbing level.”)

The Silk Roads: A New History of the World by Peter Frankopan

Paris 1919 and The War that Ended Peace - Margaret McMillan

Postwar: A History of Europe Since 1945 by Tony Judt

SPQR: A History of Ancient Rome by Mary Beard

Origins of Political Order: Francis Fukuyama - how politics + the state developed. Interesting contrast of Asia/China vs Europe/UK.

How Asia Works: Success and Failure in the World's Most Dynamic Region Book by Joe Studwel - fascinating history of the tactics east Asian countries adopted to grow post wwII

https://www.amazon.com/Science-Radio-Paul-J-Nahin/dp/B01JXOE... is quite good for a simple introduction.

My struggle book 6 by Karl Ove Knausgaard. Good, though I think I preferred the earlier books. It's still very compelling, but a little too much Hitler in this one

Underland: A Deep Time Journey by Robert Macfarlane. Poetic landscape history, reminiscent of Sebald. Good stuff.

Click here to kill everybody by Bruce Schneier. Scary.

Machines like me by Ian McEwan. First McEwan I read for a while but I liked it more than I expected. I thought the AI bits were quite good.

The deep history of ourselves : the four-billion-year story of how we got conscious brains by Joseph LeDoux. Bit disappointed in this one, it seemed like there should have been a better story here.

The Tangled Tree: A Radical New History of Life: David Quammen.

The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power by Shoshana Zuboff. Makes CS careers look less appealing.

Many ASICs are tested as 'soft' designs on FPGAs before becoming ASICs, which is essentially the same idea. It is a lot faster than verilog simulators.

This. And a milk frother/warmer.

I think this is more difficult than it sounds. Kid A is very different from kid B. I've found that things that 'work' with my kids don't work on other kids.

Along these lines, I thought this book was interesting: The Gardener and the Carpenter: What the New Science of Child Development Tells Us About the Relationship Between Parents and Children

This is horrifying. Similar practices in the City of London led to the 2009 financial crisis by eroding standards/regulation everywhere as described by Adam Tooze in Crashed.

Some additional info here in Danish: https://www.dr.dk/nyheder/penge/whistleblower-til-dr-finansf...

Calculated Risk has a good record on these calls... really. He says not this year. https://www.calculatedriskblog.com/2019/09/update-predicting...

This ties in nicely with chiplets: https://semiengineering.com/the-chiplet-race-begins/ - a way to integrate dies in a package, where the dies can use specialized processes for different functions - e.g. analog or digital or memory or accelerators or CPUs or networking etc. This would make it easier to iterate memory/CPU/GPU/FPGA/accelerator designs at different rates, and reduce development costs (don't need to support/have IP for every function, just an accelerated set of operations on an optimized process within each chiplet). But it will need progress on inter-chiplet PHY/interface standardization.