Posts by bazza 2723 posts • joined 23 Apr 2008
UK courts prohibit juries being presented with probabilities. They're rounded up to a 1 or 0, and presented as fact.
So if the prevailing Court rules say "1 in a million = dead cert", the fact that 10 million passed that way is never a matter presented to or considered by the jury.
Indeed yes. I often thought back then that Intel weren't putting FMA into x86 simply to make Itanium look better than it really was.
They did this in other aspects of x86 too. In the Nahlem architecture they added a bunch of maths registers that couldn't be access from 32bit code. This made 32bit code look a lot slower than 64bit code, even though those registers could have been added to the 32bit ISA without breaking existing 32bit code. It's all been a bit artificial. Ok, so no surprises there, but play games likes that too much and someone like AMD can come along and, overnight, make one look foolish...
One of the reasons some people liked Itanic was because it had a fused multiply-accumulate instruction in its vector core, which is the building block for a lot of signal processing routines and a fair few image processing filters. X86/64 didn't. (Power PC did).
That kept Itanium ahead, then Intel finally added the instruction to Xeons in about 2011, after which there was zero reason for Itanium.
I have to say that, considering where they were immediately after launch, getting it there and docked is quite a feat for the flight controllers, crew, their support and other participants in the ground->ISS journey. Hats off, beer well deserved.
The fact that their talents were required in this way is a great pity, considering what it indicates.
Iridium is more than still up there, they launched a new constellation recently. They are going to be pushed further into their niches, because their data offering will be swamped by Starlink and all the other upcoming systems.
But those niches are really quite important eg the Pacific Ocean tsunami warning system relies on one of Iridium's very unique services which no one else is replicating. There might come an awkward point where they need to continue being funded for the sake of those niches, and the US DoD gets to pay for it (a bit like they run GPS and let us use it for free).
I agree with all that but I gather that the stumbling block for the phased arrays has been the phase shifters and the manufacturing cost of the elements. There has been a lot of materials research in recent years... The maths for a phased array is easy and does not have to be done very often.
It's going to be interesting to see who does well. Starlink and OneWeb are inelegant and inefficient systems, spending 60% of their time over mostly unpopulated water. But they could be cheap overall.
However, the very high throughput GEO sats can be launched in very low numbers and immediately offer a comprehensive service, which is a good way of offsetting their high unit cost. Slightly worse latency is an issue, but not for most users.
I'm not convinced that either approach will dominate, but it does look like consumers will be getting a whole lot of good choices. Which is good news unless you're a telecoms company in the USA with a local monopoly...
The USA is unusual in having a lot of people with money served by a worse-than-third-world private comms infrastructure. The upcoming services from Starlink, OneWeb, and the newer GEOs will be able to offer a competitive alternative, without any additional ground infrastructure. That alone is a market worth chasing.
As ever, bug costs vary according to system type. A bug on some unimportant website costs not a lot, probably. Whereas a bug in an airliner flight control computer costs $billions and kills people. Just ask Boeing (the MAX crashes being attributed to faults that were not fixed at the design stage, despite the coding company querying the design).
What's changed sufficiently to suggest that FAA regulations are now overbearing? Does rocket fuel make less of a bang when it explodes? Do light aircraft now bounce harmlessly off tall unmarked towers? Is the environment at less of a risk? Are billionaires more entitled to ride roughshod over regulation these days?
The idiotic thing about Musk is that he picks unnecessary fights, or stages unnecessary stunts, or makes impetuous but incorrect decisions. All he's doing is risking being closed down.
I've no knowledge of the situation in Linux, but what I have seen elsewhere is code generated automagically from documentation (thankfully, well structured documentation). If there are tables of registers accessible by some means in script-consumable documents, this is well worth it. One still has to write code to tweak registers in the right way, but at least the registers are easily accessible without relying on someone manually entering a lot of addresses by hand.
Perfect.
For a while the only printer we had was a teleprinter, and it was pretty slow. Printing out a long listing was something to start before lunchtime and maybe it'd be ready afterwards. I can still hear it: chugga chugga chugga chugga chugga ding chugga wham! (that was the carriage return, you could really feel it happening...) chugga etc.
RS233 was first standardised in 1960, but that merely formalised what had been around for a long time before then (they had teleprinters using the same basic idea back in WW2). I don't know if you still can, but I remember having PCs whose RS233 ports could be configured to support 5bit Baudot code (teleprinters still being just about recently relevant when the first PCs came out).
Amusingly, SATA has more in common with RS232 than it does with IDE.
From the article:
"On a QKD system, the mathematics is in some way intrinsically, and necessarily, linked to the actual physicality of the system. This situation is unavoidable, and we would do well to design for and around it."
The thing is, whilst both quantum mechanics and special / general relativity are both theories that have excellent backing from myriad reliable experiments, they do not wholly agree with each other (especially about gravity). What this means with absolute certainty is that one of these theories is incomplete.
And it might be quantum mechanics that is wrong...
So, if one accepts that quantum physics genuinely is fully described by our theories of it, but someone else has a better theory that just happens to have a loophole, then QMKD is worse than useless. A lot worse, in fact.
Er, the semiconductor junctions used to make up the chips that are in the computer you used to type that comment rely entirely on the quantum behaviour of selectively doped silicon.
We've all benefitted enormously from a quantum something.
Microsoft briefly showed off a version of Office compiled for ARM running on Windows, also compiled for ARM, a very long time ago: 2008? "Oh goodie" we all thought, "MS are getting into this ARM thing heavily, we'll all have ARM desktops before too long (glad I kept the old Archimedes)". And what came out was - tada - WinRT.
What a let down.
So 13 years later they finally seem to be getting the idea.
From the article
Google has said: "Your keys are stored on the same infrastructure that Google uses to store its own keys. Keys are protected by Google's Key Management Service."
Oh goodie. How reassuring...
I have no doubt that Google do an awful lot to ensure the security of such things, and they are probably reasonably competent at it. But is does smell a bit like every single damned egg in one basket, well padded basket it may be... Can you just imagine the mayhem if someone got inside that.
Surely a better way would be to have an interface where the developer holds the keys and provides them to Google as required, to get the same end result. That might be a busy interface, but it leaves the control in the hands of the developer.
The "no email or calendar" bit does sounds like workplace bliss.
And it is a great pity that a company once famed for employing the right people at a time when every other 1930's company was men-only now has a reputation for ageism.
No Email, Permanently
The BBC had a piece some years ago about a few companies that had a no-internal-email rule. If the person you were emailing was on the same site, you had to walk there or pick up the phone. The report indicated that these companies were thriving.
And it's easy to see why; actually dealing with email is a huge f*****g waste of time, mostly. Get rid of it across the board (internally), and the firm has got itself a ton of productivity back at a stroke.
They've got the cargo. That's the hostage. And sure, they can make the ship go anywhere they want. The last thing anyone is going to want is a ship that cannot be manually controlled.
A container ship can have several $billion's worth of cargo on board. That's too much to be lost or even interrupted on a routine basis.
A large part of a merchant mariner's time is spent on maintenance, including maintenance of the cargo (eg refrigerated containers). That will have to be going on regardless. There is also watch keeping.
A lot of the business reason behind such labour is insurance. A well run ship has cheaper premiums. A ship that is crewless is going to be expensive to insure.
Another part of it is piracy. Most ships routinely go through the Malaca strait and the western Indian Ocean to Suez, both places where piracy can be a problem. If word gets out that there's $billions in cargo sailing past with no one on board to defend it, then the pirates are going to be having a terrific time. Their grins may even be enough to make those Canon cameras let them on board.
Depends on the engine and its intake.
Concorde didn't need afterburner to sustain Mach2, but did need it to get there. The SR71 did need AB to sustain supersonic speed, and in fact if heavy it could not even get supersonic in level flight; it needed a shallow dive to get there. Both depended entirely on variable inlet geometry to be efficient at high Mach numbers, gaining in efficiency as speed increased, both being thermally limited rather than power limited, both being hopeless gas guzzlers subsonic.
And, really, nothing has changed. No one is likely to beat the SR71 inlet efficiency because it was the ideal design. And the compressor design in such an engine is still a horrible compromise.
Concorde certainly had a lot of what was then very high tech electronics.
Totally primitive by today's standards of course, but strangely enough that gave it far better longevity than anything modern. So long as someone somewhere is making discrete transistors and passive components of almost any specification, the electronics could be repaired. It's far harder to keep modern electronics in service.
Risc-V is the underdog, but I fear that it always will be.
I just can't see why any company would take it on as a tool to try and achieve market dominance in any existing CPU market segment. To achieve that would take a lot of money and time, and a lot of nerve, unless they already were a large incumbent in the industry.
Look at how long it's taking ARM to get into the server market; years and years of trying by various ARM licensees, and there's still very large customers who still won't even look at AMD (who are at least compatible with Intel), never mind giving a whole new ISA a try.
Nor can I see an academic collective being able to do things fast enough for it to challenge a well resourced, well practised company.
People who sell or run software for a living are collectively very conservative. Even if someone came up with a compelling, gonna-be-market-beating Risc-V offering, a large part of the existing software owners / runners would be content to wait for Intel / AMD / ARM to catch up and stamp all over it.
I'm not convinced. What a supercomputer these days needs is SIMD cores closely coupled with fast memory interfaces and fast low latency interconnecting fabric. The CPU ISA that is used to orchistrate it all is comparatively unimportant, so long as it does not produce much heat itself.
So if RiscV wants to get into the HPC business, it'll need those things. And the problem is that getting software math kernels for SIMD cores right take a lot of very specialised care and expertise...
Where this won't help is with running closed source software that they also consider essential. It's no good having a domestic IT ecosystem where every machine is binary incompatible with the rest of the world's if you sometimes need to run their software (eg CAD, or silicon design tools, big important software for big important jobs).
So they can't achieve full independence without somehow reimplementing the software tools they will still rely on which cannot be got simply by cloning a repo.
I've had a reasonably good rummage around, but have been unable to find anything definitive about whether or not the M1 silicon has the equivalent of Intel AVX. Anyone out there know?
Little tidbits like this article suggests not. I don't know if Blender uses AVX or not. Given how Rosetta is supposed to work you'd think that a statically translated Blender would, if the M1 silicon were up to it, be at least comparable to running on Intel hardware, but it wasn't. So perhaps Blender does use AVX, and perhaps M1 is rubbish in that department.
AVX512 is a seriously chunky piece of kit. I know that similar SIMDs are bolted to ARMs (Fujitsu do for their latest supercomputer), but I don't know if Apple has bothered to or not.
I can't see what they can possibly do, other than something like a brutal big bang "Office 2022 is now .NET Core", or whatever. This is the kind of thing that Apple just do; announce a breaking change, and go through with it no matter the consequences. I've been constantly amazed at that eco-system's ability to just keep dying a little, everytime Apple does it.
It's clearly untennable to leave Office stuck with .NET Framework when MS are actively encouraging the entire .NET world to go to Core. Also, saying "Can't they eat Javascript" has got to be some sort of perverse joke; cross platform it may well be, but then WTF is .NET Core supposed to be all about?
Anyway, I thought that all this COM stuff was supposed to solve this kind of thing. So far as I can see, the only way it can be this screw up is if the COM Add-ons are (as the article hints) run purely in-process. But the Wikipedia article on COM says that "COM objects can be transparently instantiated and referenced from within the same process (in-process), across process boundaries (out-of-process), or remotely over the network (DCOM)." So does MS have a really simple option, changing the add-on architecture to use out-of-process instantiation without anyone really having to change that much of their existing code bases? If that were even mildly feasible, surely doing that ASAP is a good idea?
The Register - Independent news and views for the tech community. Part of Situation Publishing
Biting the hand that feeds IT © 1998–2021
