the hot new programming language

Inflation does the opposite. It's a tax that is proportional to the holding time.

Some of the power dissipation issues could be solved, if new materials tolerating higher temperatures could be found. This would increase the junction to ambient temperature difference, enhancing heat transfer.

For super computers it might even make sense to immerse the whole system into distilled deionized water. The component case will be slightly above 100 C, when a lot of heat is removed by vaporization. Of course, new cold water needs to be added to replaced the vaporized water. I have not heard of vapor cooling electronics for decades, but some big transmitting tubes used vapor cooling.

Of course if you use a single synchronous clock to drive everything on a system, there are going to be a lot of time-of-light problems. However, splitting the system into multiple "islands" (such as processors+local memory) and running every island with private clocks should help a lot, using close proximity within an island. Of course, the communication between islands need to be asynchronous e.g. using self clocked messages (such as cache lines).

If a full wafer is used, using high density locally clocked islands with plenty of space between islands would help heat removal as well as allowing space for running lines between the islands.

While thermal design must be done assuming 100 % computing load on each island, the average power consumption would be less, when every processor not doing useful work at the moment can be stopped, saving a lot of power.

MUMPS (or M) might be the next, since it is widely used in health care and is the core for new systems. It is also used in some banking applications. The youngest MUMPS programmer I know of are about 55 years old, so the supply of programmers doesn't last very long.

One of the strangest job advertisement that I have seen was for a maintenance programmer fluent in MACRO-11 (PDP-11) assembler. The job was for maintaining a control system into year 2050 on a spent fuel rod cooling pool in a nuclear power plant in Canada.

Blimey - I though I was a dinosaur because I dabbled with Macro-32 (the VAX variant) a few times!

I wonder if they are still maintaining actual PDP-11 hardware or whether it's all running on an emulator?

No. It does the opposite 0f the opposite, it devalues cash. In effect increasing the cash value of inverstments.

only if you're holding cash.

I'm just waiting to see if Fortran-II comes next (or IBM 1620 SPS assembler).

I might be competent, but I'll be 104 years in 2050.

I would not t all be surprised if they would have been running a real PDP-11/34 or 11/70, which use 74 resp. 74S series TTL chips, finding replacement chips should not be too hard for a few decades. Anyway, the strict certification requirement in nuclear industry might have made it cheaper to run the old system than certifying some PDP emulator.

In industrial applications that might run decades or even a century, it is common to replace the control system every 20-30 years. For many systems they talk about mid-life updates.

When a fuel rod has been in the reactor for 1-3 years it is removed and it is moved to a cooling pool for at least a decade or until a final deposition site can be found. The fuel rods in the pool will generate heat for at least a decade after the plant has been finally closed down. This might explain the year 2050 requirement, even the reactor is planned to be closed much earlier. IMHO, the company should have recertified a new replacement system for the next 35 years.

Is MUMPS still in use? It was a spinoff of the FOCAL interpreter.

FOCAL was great when all you had was paper tape and a few kilobytes of memory. Barbaric in absolute terms.

We used paper tape back when I was in trade school going through exploritory in machine shop for the CNC's...

Yes, it wsa barbaric but it was the thing back then :)

Jamie

But it does nothing to dampen the trading feedback loop. It doesn't matter what the frequency is, the tax is the same.

No, investments are the same. John's proposal was to dampen oscillations caused by automated trading programs. A tax proportional to time does nothing, here. It has to penalize such transactions to have any effect.

It isn't just the maximum temperature that's the problem, it's the CTE mismatch, stress due to temperature gradients (aka the Hot Dog Effect), and thermal cycling. Chip sizes are already limited by thermally-induced stress on the corner balls, despite very stiff underfill and a lot of work on the pad metal.

IBM glass-ceramic MCMs were actually processed so as to deliberately crack the copper lands free from the top layers of glass-ceramic, for this reason. (The old alumina/refractory metal MCMs never had a single field failure throughout > 20 years of manufacture, because the metal was always in compression. In glass-ceramic bricks, the copper is in tension, which caused no end of trouble. They stuck with it largely for political reasons, but saving somebody's job cost the company a _lot_ of money.

IBM Z-series are water cooled, or were a few years ago when I saw my last one. But that's water-cooled heat sinks, maybe with silicon microgrooves. You don't want to just immerse them in water, because you can get vapour locks and melt things. (That's the classical way for overheated boilers to blow up.) Gotta keep pressure and flow on.

That's been done for years and years--since the '80s that I know about. Big chips haven't been globally synchronous for quite awhile now.

No. It's a huge waste of very expensive silicon, and the yield problems would be horrible. That was what killed Trilogy back in the day.

Again, that's already done, and has been for a long time.

Packaging is hard.

Cheers

Phil Hobbs

At *least* the early '70s. The islands were fairly large by today's standards, though. It's all a matter of speed and size.

Along with scaling the voltage and clocks on islands with lighter load.

The bleeding edge is hard.

MUMPS is still actively marketed as a part of EPIC

The original MUMPS on early PDPs was nice, since it handled local variable and disk stored variables in nearly the same way.

These days (Open)VMS/Linux/WindowsNT use memory mapped files for the same purpose. The programmer sees a multipetabyte array (on 64 bit processors) and you can read or write any byte and the OS uses the very effective page fault mechanism to load pages from physical disks as well as saving updated memory locations to disk.

On Sun, 05 Jul 2015 21:57:41 +0300, snipped-for-privacy@downunder.com Gave us:

Imagine that.... MUMPS never got eradicated. :-)

it reduces the gain, once the gain is below 1.0 short term trading is pointless.

bank deposits count as "cash". invest in someting better performing.

Slow light - is that found in the slow glass of SF fame?

On Mon, 06 Jul 2015 01:59:46 -0700, Robert Baer Gave us:

Ceramic with gold pins? Don't do that!!