* Posts by IJD

43 posts • joined 19 Jul 2016

Intel couldn't shrink to 7nm on time – but it was able to reduce one thing: Its chief engineer's employment


The problem for Intel is that nobody now believes that they can deliver on their roadmaps for any CPUs beyond 14+++++, compared to AMD where nobody doubts this any more as TSMC keep hitting (or beating) all their targets for rolling out new high-yielding process nodes.

Intel stuck with inhouse fabs and lots of different big monolithic chips because they could afford it and this is what always worked for them in the past, and ignored the oncoming train wreck. AMD were forced to go to foundries because unlike Intel they had no choice, and they went to chiplets partly because they couldn't afford to do multiple foundry tapeouts for different monolithic-die SKUs.

In hindsight, these look like two of the smartest decisions AMD had no choice but to make, and two of the dumbest decisions Intel made out of pig-headedness ;-)

China’s preferred Linux distro trumpets Arm benchmark results


The problem we have competing with China is explained by what happened in the USA...

The USA became dominant in semiconductors (and other high-tech areas) by investing huge amounts in fundamental R&D and having a lot of skilled scientists and engineers working on technology, and then a spending a lot more money and having a lot of creative engineers designing on hardware based on it. The fundamental R&D has now largely disappeared because it's expensive and doesn't deliver short-term profits, and most of the engineering talent has moved from hardware into software (or services, or apps...) because it delivers more profits (and higher salaries) more quickly. It's why places like Bell Labs and the other big research centres that used to drive technology are no more.

Meanwhile places like China are pouring huge amounts of money (and talent and engineers) into both R&D and technology/hardware development because a) they've got the money b) they see it as essential to China's future success c) they're not so obsessed with stock prices and short-term profits. Yes they're still behind in several areas (especially process technology) but are catching up fast, and spending money like it's going out of fashion to do so. When I talked to someone from Huawei a few years back, they were planning to expand R&D resources by hiring 16000 new graduates that year alone...

So the sad truth is that losing our technology lead is basically down to short-termism and obsession with the next quarter's stock price, coupled with the fact that software/services is a faster way to make more money than hardware and fundamental technology -- ignoring the fact that software needs hardware to run on and communicate through, somebody else can do that, it takes too long and doesn't make enough money. Having your technology businesses dominated by the desire to make lots of money as fast as possible to pump up the stock price goes directly against what is needed to stay as a technological leader, especially in expensive areas like semiconductors.

Blaming China for this is looking in the wrong direction, the damage is basically self-inflicted by Western business strategy... :-(

Hey, Boeing. Don't celebrate your first post-grounding 737 Max test flight too hard. You just lost another big contract


Re: Learn from the smaller world...

Even with a new type rating and simulator training the Max couldn't have been flight certified without MCAS, because the rules say that required elevator force to preserve attitude must always increase as angle of attack increases, and without MCAS the Max breaks this rule -- it's not that the plane is unstable or will stall, just that the pilot would have to pull less hard on the column and this is verboten.

If Boeing had a proper FBW system like Airbus this wouldn't be a problem because the "feel" is artificial anyway and could easily be corrected, but what they actually have is a 50 year old mechanical flight control system with some electronic bodges added on -- one of these bodges being MCAS...

But there still wouldn't have been a problem if either the MCAS bodge had been done properly so it was reliable (redundant sensors with voting and fail-safe), or it hadn't been designed with enough control authority to override the pilots (4x more than the FAA was told), or the pilots had been properly trained about what to do if MCAS went wrong.

Unfortunately all these options would have cost Boeing a lot of money and possibly sales, so they opted to save the money, design a crappy MCAS, and hide it from the pilots and the FAA -- result, 300 dead passengers. There really is no excuse and the Boeing execs responsible should be jailed for corporate manslaughter...

Beware the fresh Windows XP install: Failure awaits you all with nasty, big, pointy teeth


We lost all power at work a few years ago, all the UPS kicked in fine but they only had about 30mins capacity. We had some long simulations running (a week or so) which we didn't want to abort, so sat and watched the UPS capacity lights gradually ramp down to zero while repeatedly calling the power company. Turned out to be caused by a cat vs. 11kV argument in the local substation just down the road -- leastways that's what they thought it was, it's difficult to identify species after they've lost an argument with 11kV...

SpaceX is about to launch its first Starlink internet satellite sporting a sun visor following complaints by astronomers


"It was the fifth trip into space for this particular first stage, suggesting the days of disposable boosters may well and truly be at an end."

Except if you're Airbus (Ariane) or Boeing (SLS) or Lockheed (Atlas) or Russian or Chinese or...

In other words, at an end for SpaceX, and maybe Blue Origin assuming they can catch up from being about 8 years behind SpaceX...

Ethernet standards group leaves its name in the dust as it details new 800Gbps spec


These 800G links aren't intended for PC to PC or even LAN links, they're intended for use within and between switches in data centres and in the Internet backbone where huge numbers of lower-speed Ethernet channels (400G, 200G, 100G, 50G, 25G, 10G...) are muxed together. In the longer-haul links (100s, 1000s or 10000s of km) many of these channels are multiplexed together on one optical fiber using different laser frequencies -- just like radio channels, except one channel is typically 50GHz wide (or more), the total "radio bandwidth" is about 5THz centred on 190THz, one fiber can carry getting on for 25Tb/s, and one cable (for example under the Atlantic) might have 8 fibers and be able to carry 200Tb/s.

Before anyone says "this is all pie in the sky, why would anyone want this?" these transceivers are aimed at hitting the market in 2022 when the next generation of switch chips (25.6Tbps in one chip) hit the market, all driven by the apparently unstoppable demand for more bandwidth at lower cost, which in turn is largely down to video streaming -- which is now mainly mainstream film and TV, not porn for once. How do I know all this? Because I've been working on 800G transceiver designs since last year...

More than a billion hopelessly vulnerable Android gizmos in the wild that no longer receive security updates – research


OnePlus 3, bought June 2016, last update was to Android 9, security patch level 1 Oct 2019. Seems just fine to me...

C'mon SPARCky, it's just an admin utility update. What could possibly go wrong?


PDP11/44 at uni in the early 1980s which (as a PhD student) I somehow ended up (sort of) running because nobody else would. Users who had to access image capture hardware needed superuser rights (or whatever this was called, it was a *long* time ago). Some users had multiple IDs. One user -- no, definitely not me -- who had hogged too much disk space (2 x 20MB drives shared between a dozen or so users!) decided to save his data to tape and clear up his disk areas, DEL [*,*]*.*;* deleted all files for all users (not just him) including the OS (RSX-11?) -- or at least, deleted all the file allocation tables, the data was still there, but of course no OS commands worked any more since the commands ran from disk. And there were no proper backups, tape drives were mainly used for data storage, people were supposed to save programs on 8" floppies but rarely did. Months of work for multiple postgrads circled the digital drain...

Luckily the system debugger just happened to be loaded (in 64k of RAM!), and could talk to the disk and printer, and could print out (to the line printer) the absolute block address, owner and filename for each disk block. Said idiot user had to sit there and manually reallocate all blocks by hand to rebuild the FATs, took him most of a weekend, sweating all the time because if anything happened or the debugger crashed there was no way back except reformatting the disks, reinstalling the OS, and losing all the data.

Sometimes shining a light on a nuclear problem just makes things worse


Reverse case -- the first silicon of first chip I ever designed (an echo canceller for ISDN) didn't work. Looking down a microscope at it to see if I could see anything wrong (suspect was a comparator inside an ADC) as I moved the chip around on the stage it suddenly started to work. Turned out the problem was offset voltage in the comparator, the light from the microscope injected photocarriers into the input stage which cancelled this out so long as you looked at it *just* right -- you had to within a few microns of the "sweet spot", which was in a different place for each chip...

Cisco slips on a Tolkien ring: One chip design to rule them all, one design to find them. One design to bring them all...


Re: 15 to 1 ????

This is nothing to to with the cheap end-user GbE switches mentioned above, this is stuff that builds the Internet backbone and ships terabits or petabits around and comes with typically a six-digit price tag...

Mozilla locks nosy Avast, AVG extensions out of Firefox store amid row over web privacy


Avast is still free for the basic AV package...

A bridge over troubled water: Intel teases Ponte Vecchio, the GPU brains in US govt's 1-exaFLOPS Aurora supercomputer


Re: We'll burn that bridge when we come to it

The TSMC process dates are the wrong way round -- 5nm (lots of EUV) is first, then 6nm (which is a shrunk 7nm with 1 more EUV layer), then 3nm. Exact dates depend how you define them, I believe some customers already have working 5nm silicon.

Either way there's no doubt Intel is well behind, at least 12 months even if things go well for them on 10nm (7nm TSMC) and 7nm (5nm TSMC) -- they really dropped the ball on 10nm and managed to convert a 1-2 year process lead into a 1-2 year lag, which takes some doing...

That time Windows got blindsided by a ball of plasma, 150 million kilometres away


Re: Not Just Mice

I can beat that -- a chip that didn't work unless you looked at the problem circuit through a microscope and moved it to *just* the right place in the field of view, and this was different for each chip. The light-generated current was adjusting the (random) offset voltage of a comparator to zero if you got the right amount of light falling on each side of the input pair...

Rise of the Machines hair-raiser: The day IBM's Dot Matrix turned


But not as amusing as:

"WARNING -- do not look into the laser with your remaining eye"

-- seen on a big optical bench at the NPL when I went to do some equipment calibration...

Comms room, comms room, comms room is on fire – we don't need no water, let the engineer burn


Re: Hydrogen

More likely to have been a tank of hydrogen fluoride (HF) which is used for all sorts of things in wafer processing and is *incredibly* nasty stuff...

Never let something so flimsy as a locked door to the computer room stand in the way of an auditor on the warpath


Re: so easy to get in

Mate of mine years ago was a security guard for a construction equipment company. Turned up one Monday morning to find everyone in headless chicken mode because an entire tower crane was missing -- you know, those tall things that lift stuff, and are worth several hundred thousand quid each.

When they rang the police they already knew -- the thieves had turned up at the weekend complete with low loaders and lifting gear, loaded up the crane (they're modular, bit like Meccano), and then got the police to hold up the traffic while they got the awkward loads out of the yard...

Tech giants get antsy in Northern Virginia: Give us renewable power, there's a planet to save... and PR to harvest


Re: Also

Thta's one of the excuses the anti-wind-power lobby used in the UK, that tens of thousands of birds would be killed. When it was pointed out that their beloved pet moggies kill tens of millions, that argument went quiet...

Register Lecture: Hidden heroes of Alan Turing's Enigma


You'd think with modern technology it would be possible to do something like streaming the lecture live, or failing that recording it and making it available online afterwards. Millions of vloggers seem to be capable of doing this, surely it can't be beyond the wit of the organisers...

So you've 'seen' the black hole. Now for the interesting bit – how all that raw data was stored


Re: I was just musing the other day that M$ might do this and low and behold!

How about mercury delay line memories, as used on UK's MOSAIC computer? (which my father-in-law Bill Chandler designed and built with Allen Coombs after Colossus). Five tons of mercury in a hundred steel tubes, kept temperature stable to better than one degree...


Airlines in Asia, Africa ground Boeing 737 Max 8s after second death crash in four-ish months


The first "mistake" Boeing made was building a safety-critical system into the 737MAX without enough sensor redundancy, because this would have cost more.

The second "mistake" was not telling airlines/pilots about MCAS because this would have needed pilot retraining which costs the airlines money and puts them off buying the plane, and such a change would have meant re-certification is needed which costs Boeing money.

"Mistake" in inverted commas because it appears that saving/making Boeing more money was the driving factor in both decisions. Hope they're still happy with that given the resulting body count...

Return of the audio format wars and other money-making scams


Loudness wars

I've digitised quite a lot of my favourite old LPs, and apart from the amount of time it takes -- because really you need to get rid of any big clicks/pops/scratches before normalising the volume -- the most obvious difference is how *quiet* they sound compared to any modern CD. Not because the peak level is lower (digital full scale in all cases) but because they haven't had loads of compression and peak clipping applied to raise the average level. End result is you have to turn the volume up significantly, then turn on back down for a modern recording.

And I'm not talking about heavy metal or rock or pop here, I'm talking about acoustic or folk or classical or jazz recordings where you'd think this wasn't so prevalent. But it is, and this applies to every single "old vinyl" album I've transferred compared to any "new CD/digital" album. The difference is obvious when you look at either the waveforms or average levels in any audio tool...

Facebook didn't care if your kids ran up gigantic credit card bills – lawsuit


Colleague at work's 8-year old son ran up a credit card bill for eight thousand quid in one month buying gold and stuff in an online children's game. Was playing on dad's iPad which had (password-protected) credit card payment details stored on it, in-app purchases auto-filled in payment details, needed password but kid has seen dad entering this and remembered it, entered it once and off he went. Of course he didn't think it was "real" money he was spending, he thought it was pretend money.

Cue explosion when credit card bill arrived. He managed to get the money refunded by arguing with the online game firm that there's no way they should have allowed a young child to run up a bill of this size, at least without confirming the spend was authorized by the card holder.

Yes you could say this was the fault of the parent for storing card details on a device that he allowed his son to play online games on, but as far as he was concerned payments were password-protected. For sure the online gaming company is more to blame for having a game targeted at children which allowed -- no, encouraged -- them to spend real money on in-app purchases. And the biggest blame goes to them for not having a system which flagged ridiculously high spends as likely to be fraudulent or at best unauthorized, which I guess is why they paid up -- if it had come to court they wouldn't have had a leg to stand on.

Having swallowed its pride and started again with 10nm chips, Intel teases features in these 2019-ish processors


TSMC 7FF/7FF+ and Intel EUV

Saying that TSMC 7FF is "low-power" and 7FF+ is "high-speed" isn't correct. 7FF has both compact low-power libraries (and metal options) for mobile (which Apple/HiSilicon use) and bigger faster higher-power libraries (and metal options) for CPU/HPC (which AMD use) -- actually, they can be mixed on the same chip. 7FF+ uses 5 EUV layers and new libraries to get 15%~20% area shrink, a small power reduction, and an even smaller (a few percent) speed increase -- and it also has low-power and high-speed libraries, just like 7FF. The main reason for 7FF+ is to pipeclean EUV before TSMC use it in anger for 5nm (due next year), and to get some reduction in die size/cost/TAT.

Intel's 10nm problems are not due to EUV because they don't use it -- they used quad patterned metal instead to push the metal pitch down (problem#1), cobalt interconnect instead of copper for the same reason (problem#2), contact over active gate to save more area (problem#3). All these together screwed the yield, and some or all are being removed from their "new 10nm" process due out next year.

More to the point, they're now more than a year behind TSMC 7FF with a similar process instead of the 3 years ahead that they originally promised...

Boeing 737 pilots battled confused safety system that plunged aircraft to their deaths – black box


In the end, the fault comes back to Boeing. Adding the MCAS system onto the 737MAX was a bodge to get it through qualification after the engine/airframe changes made it unsafe in a stall, but the crash would not have happened if the system had been properly designed against sensor failure (redundancy, voting) and/or the pilots/airlines had been told about MCAS.

The first one would have probably meant design changes to add the extra AOA sensors and cost Boeing more at manufacture (bad), the second would have meant recertification and retraining of pilots which could have meant them selling fewer aircraft (bad). In the end somebody decided that money mattered more than safety -- probably not deliberately, but this is the kind of sloppy "it'll probably be just fine" thinking which sooner or later kills people.

Of course if it comes out that the possible problem was pointed out by engineering but stomped on by management (Ford Pinto, anyone?) and this ever comes out, Boeing will be in deep doo-doo...

Bright spark dev irons out light interference


Spark plugs on not-so-old cars...

On one of the early CAN-bus cars (Fiesta?) the recommended fix when the rear wiper stopped working was to fit new spark plugs...

As the spark plugs wore out the engine management system upped the juice to maintain a good spark, which caused more interference, and the rear wiper was at the far end of the CAN bus and had the weakest signal, so stopped working before anything else did.

I have respect for whichever engineer figured that out, but can imagine it must have been difficult convincing customers that they weren't being ripped off by a garage installing expensive parts for no good reason...

European Commission: We've called off the lawyers over Ireland's late collection of Apple back taxes


Does anybody seriously think Apple's tax dodge was legitimate? Having all EU profits assigned to a shell company theoretically based in Ireland (so no US tax payable) but headquartered in the US (so no Irish tax payable)? No doubt some Apple tax accountant thought this was a splendid wheeze, but it's clearly not what any of the governments intended to happen, and they can't really complain when they get found out and sent a massive bill.

Decoding the Google Titan, Titan, and Titan M – that last one is the Pixel 3's security chip


Nobody's pointed out yet that it's an integrated circuit, so Titan IC... ;-)

Rookie almost wipes customer's entire inventory – unbeknownst to sysadmin


PDP 11/44 (running RSX/11 IIRC) at uni shared between a dozen postgrads (including me), who needed superuser privilege to use the video hardware. One who had several accounts wanted to remove all his old files so lazily typed rm [*.*]*.*;* instead of removing one UID at a time. Everything including everyone else's files and the OS (and all system utilities) disappeared, and of course the last backup was months old. Of course the files were still physically there but no longer accessible.

Luckily I'd been running the system debugger and it was still resident in memory (all 64k of it...) as was the printer driver, so I managed to print out the block allocation of the (20MB) hard drive on the line printer. Said user then had to use the debugger to manually reallocate every block on the disk by hand one at a time to the correct UID (including the OS) and filename. Took him all weekend...

[no it wasn't me, but I was the defacto sysadmin when people suddenly found they hadn't got any files any more]

Juniper prepping for a 400 Gbps Ethernet world


400GE will either use 4 intensity-modulated channels of 100G PAM4 (shortest reach), 8 channels of 50G PAM4 (medium reach), or a single channel of 400G QAM16 coherent optics (longest reach).

Boffin supercharges FPGAs with timing signal tweak


Re: Love the comment...

Or if you go back further, mercury-filled ultrasonic delay lines were uses as computer memory in the 1940s/1950s -- except they were six feet long, with a hundred of them fitted into a rigid anti-vibration steel frame weighing several tons (holding five tons of mercury!) , and had to be kept within a degree of the design temperature...


Now *that's* an engineering challenge...

[my father-in-law was responsible for building it, after doing the same on Colossus...]

Revolutionary Brit-made SABRE hybrid rocket engine to burn in 2020


Re: Excellent!

According to L J K Setright in his "Power to Fly" book, the Napier Sabre was initially dogged partly by some manufacturing problems but mainly by maintenance staff/pilots flogging the engines to death by running them under coarse pitch/full throttle conditions that should never have been allowed. Once these problems were fixed it became a reliable and very powerful engine, probably the highest power density piston aero engine ever built.

Farewell, slumping 40Gbps Ethernet, we hardly knew ye


The simple fact is that 40G ports cost about the same as 100G ports, maybe even more now since 100G volume is rising rapidly and 40G volume is falling. 40G rollout was delayed by several years due to the telecom crash and technical issues, so 100G has overtaken it.

8 out of 10 cats fear statistics – AI doesn't have this problem


If you want to see creative misuse of statistics, try this -- look at the graphic lower down if nothing else...


Bye bye MP3: You sucked the life out of music. But vinyl is just as warped


I've transferred a lot of my favourite vinyl albums to 256k VBR MP3, each time normalising the full scale after removing all the big clicks and pops I could find. Without exception the average level is a lot lower than any modern CD I have, because the peaks were left in the recordings rather than compressing them to get the average level up -- which is trivial to do digitally, so *everybody* does it.

I haven't gone through and collected any statistics, but I estimate the typical difference is around 5dB, and maybe up to 10dB in extreme cases -- this shows just how much the peaks have been crushed down to get the overall level up. I've seen this done in studios both on individual instruments (especially drums) and then the overall mix (pushing down the highest peaks in a track or album), both done to get the apparent volume up so the track doesn't sound quieter than everyone else's -- it's basically a race to the top on volume and bottom on quality...


256k VBR and 320k fixed rate were both indistinguishable from the CD source.

I've done the same test myself on studio headphones and came to the same conclusion -- and 128k (most common rate used) is utter sh*te, which is why lots of people think MP3 is rubbish for sound quality.

FAKE BREWS: America rocked by 'craft beer' scandal allegations


Greene King IPA bloody lovely -- what planet are you one? It's one of the blandest real ales around...

Support chap's Sonic Screwdriver fixes PC as user fumes in disbelief


The Apollo DN660 workstation (washing-machine-sized beast with bit-sliced ECL emulation of a 68000 CPU) had a space on top of the card cage just the right size for keeping a pizza hot...

Zombie Moore's Law shows hardware is eating software



There's a continuous space of power vs. flexibility -- to do the same amount of processing and in the same process node, at one end a fixed-function ASIC is by far the lowest power and die size but inflexible, an FPGA is more flexible but higher power and die size (typically ~10x), a CPU is completely flexible but much higher power and die size again (typically ~100x).

Cost would follow the same trend if volumes were similar but they're not, given the very high NRE cost of the latest process nodes tilts the costs towards FPGAs and CPUs unless your volumes are very high. Saying that doing the same job would cost $100 in a CPU or $10 in an FPGA or $1 in an ASIC is true, if you want tens of millions of them -- bear in mind that the total NRE (design and mask) for even a small ASIC in the latest process nodes is at least tens of millions of dollars, or more than a hundred million for a more complex one, so you need to sell a lot of chips to get this back.

So for most cases CPUs or FPGAs make more sense, advanced process node ASICs (or custom CPUs with custom hardware accelerators) only really make sense where the need to get lower power is absolutely imperative and the TAM justifies the cost. One interesting trend driven from this is that such designs move back into the companies who make the end product like Apple (vertical integration), because getting lower power (or higher speed) by doing your own 10nm chip makes sense if you can clean up the market selling a $600 product with $300 gross margin, but not if you're selling a $60 chip with $30 gross margin at the same volume.

Boeing just about gives up on the 747


I can beat that -- an American pilot friend who used to fly over regularly and stop over with me arrived several hours early once with a big grin -- he'd had a 200 knot tailwind and had just set the fastest commercial crossing record. We had quite a few beers to celebrate...

Avoiding Liverpool was the aim: All aboard the world's ONLY moving aqueduct


Loads of fish and chip shops in the North use beef dripping...

By 2040, computers will need more electricity than the world can generate


Extrapolating into the future predicted that by the early 20th century cities like London would have ground to a complete halt because the streets would be six feet deep in horse shit -- which is pretty much what this prediction is...

If the power consumed by the IT industry ever got close to the power generation capacity of the planet, the IT industry would have to find a way of reducing power or go bankrupt because there'd be no power left to run the industries to make things which pay for the IT industry.

You could just as easily extrapolate the events of the last few weeks and predict that by the end of the year everyone on the planet will be spending 100% of their waking hours playing Pokemon Go...

Tesla's Model S autonomous mode may have saved a life


Avoiding accidents is exactly the same issue as causing them -- there will be cases where an autonomous driver will save a life when a human driver wouldn't, and vice versa, and the cases will be different -- what matters is how often the two occur.

Humans have very good sensors and decision making when used properly, most human driver accidents -- around 94% according to the statistics -- happen when this isn't the case (tired, inattentive, distracted, texting, risk-taking, stupid decisions...).

Autonomous cars have less good sensors and less evolved algorithms so will make mistakes that good human drivers don't, but don't suffer from the problems listed above that cause most "accidents".

So autonomous cars will probably (at least initially) increase the 6% but decrease the 94% -- which means overall they're likely to avoid more accidents and cut the death/injury rate, regardless of whether there were some cases where a human could have done better.

Everyone keeps bringing up the exceptional cases (driving in a blizzard, Tesla under a truck) where a human could do better while ignoring the hugely bigger number (most accidents!) where an autonomous car would be better -- because everyone thinks they're an above-average driver (yes, really...) and that only other people cause accidents, it wouldn't have happened if *I* was driving ;-)

Brit chip biz ARM legs it to Softbank for $32bn


Re: Legal obligations to shareholders

Funny that most companies don't seem to work that way any more, short-term shareholder profit clearly takes priority over long-term success -- otherwise more companies would be spending more on R&D to develop new products for a few years hence, instead of cutting R&D budgets to "save money"...


Biting the hand that feeds IT © 1998–2020