* Posts by Steve Todd

2644 publicly visible posts • joined 19 Sep 2007

Getting to the bottom of BMW's pay-as-you-toast subscription failure

Steve Todd

Re: if you tolerate this then your chilled air will be next.

They have been doing this for years. When I was a kid the local Polytechnic bought a Univac 1110, which turned out not to be fast enough for their needs. They paid for an expensive upgrade, the service technician turned up and found that the jumper he was going to move for the upgrade had already been moved.

Steve Todd

Re: don't forget

Think yourselves lucky. The US don’t have amber turn signals, they just flash the break light on the side they are going to turn. Chances are that they just squished the break light enough to fit an amber light into the cluster.

While Intel XPUs are delayed, here's some more FPGAs to tide you over

Steve Todd

Erm, 10nm has been renamed Intel 7 as it’s roughly equivalent to TSMCs 7nm process

Intel 4 is ready now and Intel 3 is due mid year. The next gen FPGA parts are likely going to be on one of those two.

Non-binary DDR5 is finally coming to save your wallet

Steve Todd

Re: Apple is testing 24GB stacks. Maybe.

Erm, Apple use HBM stacks in their M series processors, not DDR5.

Qualcomm talks up RISC-V, roasts 'legacy architecture' amid war with Arm

Steve Todd

Where’s the ARM A7x competitive RISCV chip?

Never mind something newer and more powerful. Eben Upton was asked about RISCV in the Raspberry Pi and that was his response. They are fine for microcontrollers, but there’s nothing yet on the horizon that can compete with even slightly modern ARM cores, and it’s a 100+ man year project to produce something, even starting with an experienced team and an existing design.

Long story short, Qualcomm aren’t going to be able to roll a replacement non-ARM desktop/laptop class core any time soon.

GCC 13 to support Modula-2: Follow-up to Pascal lives on in FOSS form

Steve Todd

Re: Shouldn’t be too hard to reproduce the OberonStation

I’ve no doubt that trying to sell these would be a losing proposition, but as an open source core on off-the-shelf hardware it might generate some interest. I’m also interested in the fact that it is described as running on a “RISC5” core, I wonder if this is in any way related to RISCV?

Steve Todd

Shouldn’t be too hard to reproduce the OberonStation

It seems that it was just a port of the Verilog published in the 2013 book by Wirth and targeted at a Digilent Spartan 3 Dev board. Best as I can figure this is way outclassed by the £50 Sipeed Tang Primer, which has everything you need on a modern board (HDMI, USB 2 OTG, 128MB of DDR3, Ethernet, SD card slot and built in JTAG/USB serial).

Intel offers Irish staff a three-month break from being paid

Steve Todd

Re: Are they allowed pursue other paid employment during that time

Erm, you do know that the leading manufacturer of chip making tools, ASML, is a European company? Also the biggest user of said kit is TSMC, based in Taiwan.

If Apple's environmental rhetoric is meaningful, Macs and iPads should converge

Steve Todd

Apple made it possible to run iPad apps under MacOS already

It doesn't work too well as the input methods are different. Don't expect it to work well the other way either, Tablets and Laptops are very different in the ways that users expect to interact with them.

Intel’s first discrete GPUs won't be a home run

Steve Todd

It will be interesting to see if they have the same problem as the A370

The review I’ve seen says that it suffers from fairly noticeable glitches in frame rate, and it NEEDS a CPU able to handle resizable BAR. That and the fact that older games (using DX11 or prior) perform poorly in comparison. Intel still seem to have a way to go with their drivers.

Apple's new MacBook Air: Is the jump to M2 silicon worth another $200?

Steve Todd

Re: Is the jump to M2 silicon worth another $200?

So for you the answer is no at any price. You do not represent the entire world though.

Personally I have an M1 based Mini, and I’m happy with that for what I use it for (it will transcode 1080p video to h265 at around 200fps, while running near silently for example). I’m not the target audience either.

There are however many people for whom the answer is Yes. If there aren’t enough then expect the M2 models to be discounted or dropped to that of the M1 and the M1 discontinued.

Russia cobbles together supercomputing platform to wean off foreign suppliers

Steve Todd
FAIL

Re: Fab plants?

Even if China were stupid enough to try to invade Taiwan, and TSMCs fabs survived un-damaged, they still wouldn’t be able to make leading edge steppers. ASML, the only company in the world currently making EUV steppers, is based in Europe and the US of A. The next best manufacturers are Japanese.

Also these machines need a continuing supply of parts and consumables. The existing machines would stop functioning in a fairly short period of time if taken by China.

Steve Todd

Re: Fab plants?

The best that China can currently manage are 90nm lithography machines. India isn’t even that far along. Neither of them are going launch Russia into the stratospheric reaches of the top 500 supercomputer tables.

Intel updates ATX PSU specs, eyes PCIe 5.0 horizon

Steve Todd

Re: 600W for a GPU?

It’s worse than that. It’s 50 amps at the PSU, but more like 500 amps at the GPU. It’s pretty much like a dead short electrically speaking.

Steve Todd

Re: Still not seeing the point of ATX12VO

What do you think uses 5v and 3.3V power these days? All the heavy lifting in current systems is done by VRMs stepping 12v down to whatever the required voltage is (and at pretty high orders of efficiency). Putting extra VRMs on the motherboard for 5V and 3.3V for the (comparatively) low demands on these rails allows for a simpler and more efficient PSU, at the expense of a slightly more complex and slightly more expensive motherboard. It’s a reasonable trade.

C: Everyone's favourite programming language isn't a programming language

Steve Todd

Sorry, someone who found CORBA in any way praiseworthy?

It was a godawful mess that shouldn’t be touched with a 10 foot pole.

I’m not convinced that any inter-language low level communication protocol (to, for example, let an application talk to the operating system) is going to be much cleaner without introducing huge performance penalties (which is what CORBA did in spades)

Nvidia CEO: We're open to Intel making our chips

Steve Todd

Re: Live off the Supply Chain

You mean they can’t fab at the leading edge.

Intel have renamed their 10nm process “Intel 7” as it’s roughly in line with TSMCs 7nm process. TSMC have however moved on to 5 and now 4nm. Intel are at least 18 months behind, possibly 3 years or more.

AMD: Our latest, pricier mega-cache Epyc processors leapfrog Intel’s

Steve Todd

Re: AMD catches up to Cray

By definition all cache memory is static. Having to cope with RAS/CAS/refresh cycles would ruin their performance. You literally present an address to them and either get a response back or a cache miss signal predictably a couple of cycles later.

Another Massive Display as AMD hails 'outstanding' 2021, teases Genoa and Bergamo chips

Steve Todd

Re: China

Xilinx is a fabless design house. They don’t have any plant. They have offices in Shanghai, which is I suspect why they needed Chinese permission, but are headquartered in the US of A

Final PCIe 6.0 specs unleashed: 64 GTps link speed incoming... with products to follow in 2023

Steve Todd

NVMe SSDs are already saturating PCIe 4.0

and PCIe 5 SSDs are already starting to appear. GPUs aren't the only things that chew bandwidth.

Add to that CXL is paving the way to unified memory access, where GPU RAM and CPU RAM are effectively shared, so the faster the link the better.

RISC-V CTO: We won't dictate chip design like Arm and x86

Steve Todd

Re: No there weren't

To answer that question, no, there is no requirement to contribute back. RISC-V is provided in the form of a spec an the tools to verify that an implementation meets the requirements. Nothing needs to be shared about how a core is actually implemented (and indeed the fab manufacturers of many of the more advanced process nodes forbid the sharing of technical details of how the nodes work, so that detail wouldn’t be available anyway).

Think small, score big: India details subsidies for chipmakers

Steve Todd

China wafer manufacturing?

There’s an interesting problem that the Chinese have, most of their wafer production depends on equipment made by foreign companies.

The best that Chinese native hardware can manage at the moment is 90nm, and much of their existing capacity is for older 150mm or smaller wafers. American trade restrictions mean they can’t lay their hands on the equipment for advanced nodes. They are claiming to be able to start manufacturing 14 and 28nm chips with their own equipment next year, but in practice they have only demonstrated very small chips at that scale in the lab (this while IBM are demonstrating 2nm chips) and in practice it’s likely to be at least 3-4 years to translate this research into practical hardware and start volume production.

Meanwhile the cost of building an advanced fab is becoming astronomical (from memory it’s circa $10 billion for a 5nm plant at the moment), so the Indian investment isn’t going to get them much in the global scheme of things.

Mediatek unveils its first ARMv9 smartphone chip for advanced handsets

Steve Todd

Odd, I thought they’d just done that

But they are now calling it “Intel 7” (take a look at the specs for the Alderlake CPUs)

TI will splash out up to $30B on wafer fabs

Steve Todd

So what process node are these targeting?

It's all very well saying that they are going to be using 300mm wafers, but without knowing what node they are targeting we have no idea what market segment they are after. There's a lot of demand for older nodes (12nm or larger), but these sound like 5 or 3nm given the cost and timescales.

Amazon hasn't launched one internet satellite yet, but it's now planning a fleet of 7,774

Steve Todd

International law holds that anything in orbit is extraterritorial. It does however state that radio transmissions out as far as the moon are regulated by the ITU, and the member nations under it. So SpaceX can fly satellites anywhere it wants, but to broadcast to a station on the ground it needs permission from the host country to use the specific radio band.

A Windows 11 tsunami? No, more of a ripple as Microsoft's latest OS hits 5% PC market

Steve Todd
WTF?

Re: The Windows 11 sausage machine conveyor belt...one billion PCs -> Landfill Crud.

Sorry, what has the lifespan of an OS version got to do with not being green? Older machines will quite happily run newer OSX builds (last time I looked you could load the latest version onto 7 year old machines) and security updates continue for some time beyond the OS.

Nobody cares about DAB radio – so let's force it onto smart speakers, suggests UK govt review

Steve Todd

Re: Dud DAB

The problem with DAB is that it’s a very old standard. It uses MP2 audio compression (so, low bandwidth stations sound really bad due to the old and poor compression) and little to no error correction on the signal (resulting in a kind of burbling noise when you have poor reception). DAB+ improved on it by moving to HE-AAC compression and adding Reed-Solomon forward error correction code to the signal. The down side is that older DAB radios cannot receive the newer signals, and in the UK at least, there are no plans to move to it until the majority of radios can handle it.

Steve Todd
WTF?

Sorry, why is this about DAB?

DAB is just a means of delivering content to listeners. Smart speakers won’t be forced to include DAB hardware, all they’ll need to do is stream internet radio, which most if not all stations broadcasting on DAB do anyway. So what it comes down to is not “force the users to listen to evil DAB”, but rather “make internet radio available to anyone who wants it on their devices”.

Windows 11 in detail: Incremental upgrade spoilt by onerous system requirements and usability mis-steps

Steve Todd

I’m guessing you’ve never seen an Alto running

Apple may have used the ideas of WIMP (for which they paid Xerox in shares BTW), but MacOS looked and felt nothing like the Xerox machine.

Report details how Airbus pilots saved the day when all three flight computers failed on landing

Steve Todd

Re: Threshold?

Runways over 4000 feet long have aiming point markers that are 150 feet long 1000 feet from the threshold. They were a little long, but not horribly so.

Steve Todd

Re: Only on landing?

As with all software, bugs are categorised. If a bug (a) happens only occasionally and (b) can be worked around then pretty much anywhere will give it a low priority and ship the fix as part of the normal update cycle.

They have found the bug, pilots have been made aware and can deal with it if it occurs. It’s not something that’s going to kill people if it does happen again (*cough* Boeing 737Max *cough*). As such it’s still safe to fly pending a properly tested fix.

Please, no Moore: 'Law' that defined how chips have been made for decades has run itself into a cul-de-sac

Steve Todd

Actually I was going to state the opposite

It’s full of glaring holes. As examples:

Current (non EUV) processes rely on a technique called multi patterning. They create features smaller than the wavelength of the light being used by using multiple exposures with different masks designed to produce interference patterns. Current EUV systems can do the same job faster as they need many fewer masks/exposures. They are however horribly expensive (which limits the number that any given manufacturer can buy) and in short supply, even for those with deep pockets (ASML, the manufacturer, can only build a couple of dozen per year)

Second, no modern process exposes the whole wafer in one go. They expose one chip at a time, stepping from one to the next. This places one of the limits on the maximum size of a single chip, known as the reticle limit. The EUV (or, more correctly, soft X-ray) laser needs a lot of power to produce an exposure, but keeping an even level across the reticle is a solved problem.

Steve Todd

Re: How about a focus on efficiency?

The two are not necessarily opposed. The Apple M1 gets high IPC at low power, for example. X86 has a disadvantage in superscalar design because of the variable length instructions it uses (up to 15 bytes IIRC), which makes the fetch/decode steps complicated.

Modern X86 designs translate native instructions to RISC like micro ops and then execute these on a RISC like core. Getting this core up to 8 wide to match the M1 is doable, but then feeding the micro op pipeline becomes a problem. ARM definitely has the upper hand here.

Steve Todd

Re: magnetocaloric effect ?

The important factor you’re ignoring is the speed of light (or, more correctly, the speed of electric fields in a wire). The current generation of CPUs are already running so quickly that a signal doesn’t have time to make its way from one side of a chip to the other in a single clock cycle, never mind switch a gate at the far end of that wire.

CMOS transistor pairs draw power when they switch state as there is an instant when they are both, at least partially, on when the logic level changes. The current per gate is small, but multiplied by billions of gates it becomes significant. There are other semiconductors that switch faster (gallium nitride for example) and thus run cooler, but you can’t make them as dense so silicon is faster for large scale processes.

The second problem is in disposing of the heat in a home/office environment. Data centre CPUs are already past 200W using industrial cooling solutions. Standard desktop systems run closer to 135W. You CAN run them faster with the right cooling (anyone remember Intel showing of a high core CPU at over 5GHz, only for it to be found out they were using a high power industrial chiller). They don’t get that much faster, not because of temperature but because the signals don’t move fast enough across the chip.

Steve Todd

Re: What do we do with all this processing power?

Treating every variable as if it were a variant? You’re using the wrong languages. Some tasks are not speed critical, so ease of programming wins out (JavaScript or Python being classic examples). Where reliability and speed are vital then strongly typed languages rule.

Steve Todd

Re: Moore core

Moores law states that the number of transistors on a given sized piece of silicon will double every 2 years. The speed of those transistors wasn’t mentioned.

Clock speed has, of late, gone up rather more slowly, and the number of transistors in a CPU core has hit diminishing returns, so the way manufacturers have chosen to use the available space is to replicate their cores a number of times (in addition to adding custom hardware blocks to offload common tasks from the cores to hardware better suited, like vector processing or encryption/decryption)

Given the demands of a modern OS to run multiple concurrent processes (applications, drivers, services etc) then this makes perfect sense, even if developers are bad at using multiple cores for a single application.

BOFH: When the Sun rises in the West and sets in the East, only then will the UPS cease to supply uninterrupted voltage

Steve Todd

Re: Reminds me...

Kind of depends. My grandfather was a chemical engineer working on one of the first UK Nylon plants. When they had finished tuning it, it ran as 110% of original design capacity (reliably 24/7 that is).

Visual Basic 6 returns: You've been a good developer all year. You have social distanced, you have helped your mom. Here's your reward

Steve Todd

Re: Nice

VB.NET may have had little to do with VB6, but at least MS used the opportunity to fix the faults and limitations of the language (Interface only inheritance, archaic error handling, 254 controls on a form, inconsistent array bases (some arrays are zero based, some are one based) to name a few).

The problem was that VB.NET was so different that developers may as well have just learned C# instead, and some of the capabilities of C# weren’t provided in VB.NET.

Appeals court nixes online blueprint sharing ban on 3D-printed 'ghost guns'

Steve Todd

Re: Why bother with 3D printing

I guess you haven’t tried searching eBay for lathes recently. You can pick up a brand new metalworking lathe for less than $1000, far less than that for a used model. You’ll need to put more time and effort into learning how to use it, but it’s not rocket science to pick up.

Pat Gelsinger’s Intel will evolve from lone wolf to touting modular systems-on-packages with third-party foundry collaboration

Steve Todd

Re: I guess he needs more time

There isn’t any standardisation when it comes to talking about process sizes, so Intel 10nm is comparable to TSMC 7nm, and Intel 7nm looks like it will be comparable to TSMC 5nm. That’s still well late to the party though (TSMC have been in volume production since last year with 5nm and are expected to have 3nm in volume next year).

SpaceX wants to slap Starlink internet terminals on planes, trucks, and boats – but Tesla owners need not apply

Steve Todd

Re: 1 million users?

Currently licensed in the USA. They have applied to increase that number to 5 million, then there’s the rest of the world to think about...

GPS jamming around Cyprus gives our air traffic controllers a headache, says Eurocontrol

Steve Todd

Re: ILS?

Not so much of a need to augment Galileo signals, they can manage a precision of +/- 10cm without help.

Steve Todd

Re: ILS?

Larnaca already has VOR-DME, which will let you determine the heading and range to the airport. ILS is far too directional to be of use beyond flying the approach. These older technologies are being phased out in favour of GNSS procedures as they are expensive to install and maintain, plus systems like ADSB/TCAS (which are required for most commercial flight these days) will still be reporting the aircraft’s position to ATC/other aircraft based of what GPS says.

The 40-Year-Old Version: ZX81's sleek plastic case shows no sign of middle-aged spread

Steve Todd

Re: ZX81 option for Raspberry PI?

A Raspberry Pi with an onboard RP2040 microcontroller is what I’m looking forward to. The Cortex A series MPU used by the Pi is far from ideal for control, but the RP2040 is short of compute horsepower. The two in combination should be a winning proposition.

Steve Todd

Re: Retro-Wreckers

The decision to use the 8 bit bus version of the 680000, the 68008, is really what hobbled it. Microdrives were a bit slow and clunky, but once you had code and data loaded the handicapped CPU felt a bit slow and clunky compared to the full 16 bit competition.

Steve Todd

Re: Ha, latecomers

Well, other than the fact that the max on board RAM was (IIRC) 384 BYTES, and the terrible INS8060 CPU had no stack pointer and could only handle 4K of RAM at one go without paging ...

Steve Todd

Ha, latecomers

I built a MK14 (look it up), it made the ZX80 look positively palatial in terms of resources.

Steve Todd

Different problem. The ARM2 didn’t have HARDWARE FP, but had plenty of space and speed to do it in software.

Steve Todd

That was the reason for the larger ROM. The original ZX80 didn’t have space for FP routines in the only 4K of ROM space it had.

Copper broadband phaseout will leave UK customers with higher bills and less choice, says comparison site

Steve Todd

Re: Emergeny calls

“And why does SCADA, DCS, etc need gigabits/second to poll its RTUs in pumping stations or whatever? Sometimes those things need very low (or at least very predictable) latency, and the need to packetize serial data into IP packets can get in the way. IP is trendy though.”

Who said you needed to buy gigabit links? They are a available if you want them, but lower speeds are available. Yes, there is more overhead compared to simple fixed links, but packets can contain small blocks of data if responsiveness is key.

“Figures. Like a parallel universe, where for some weird reason switching from message parsing in software to message parsing in (e.g.) FPGA to save a millisecond or so ”

Now you’re demonstrating your own ignorance. You’re talking about High Frequency Trading, which normally co-locates hardware in the exchange data centre and connects to their core network. Any leased line is extremely short (between two routers in a rack).

I was talking about near real time reporting and recording, where a single exchange can produce hundreds of thousands of messages per second, which needed to be converted to a standard format and merged with the converted message traffic for all other supported exchanges. Industrial quantities of data here.