OT: Where do I find...

64 bit applications have existed ad least since the early 1990's e.g. DEC Alpha running OpenVMS or OSF/1.

In those days, very few organizations could afford multiple gigabytes of physical memory, so in those days, the only real application was handling big data bases. In a programming language, you simply declare a multiple terabyte array and doing an assignment statement like C = Arr[i] will cause a page fault and a page load into physical memory, using the normal page fault loading from one of the disks in a big disk farm, if the page had not already been loaded into physical memory.

Conceptually, the main memory is just a cache (some would call it L3. others L4) and the actual data is stored somewhere on the rotating disks.

Reply to
upsidedown
Loading thread data ...

Thanks for the information and the history here. I never had any experience with these chips (or similar "big" systems), so I only read about them - and my memory is apparently a bit hazy.

Reply to
David Brown

I just came off of almost five years programming Itaniums. As you say, it's a over-simplification. Still, that's not too bad a way to visualize it if you're not going to study the instruction set in detail.

- Bill

Reply to
Bill Leary

That is a very interesting assertion. They are still making them. YCLIU. It never mad it in the (m)ass market, but lots of long lived servers use it and it has its own legacy SW market.

instructions.

>
Reply to
josephkk

The implementation wasn't the problem. The fundamental assumptions were, viz: - compilers could reliably extract the amount of parallelism necessary to keep the hardware running at full tilt - at the point at which everybody realised a key parameter was how much power cpus dissipated (often expressed in terms of MIPS/W), HP opted for massive speculative execution in which most of the speculated results were thrown away together with the energy used to create the irrelevant results

The first assumption was false, the second was a valid but blind alley.

Reply to
Tom Gardner

I thought there was also significant market resistance because Itanium's x86-32 emulation was much slower than a real x86-32 of the era. That's what let amd64 into the door.

Reply to
Paul Rubin

That emulation concept was always at 90 degrees to the rest of the universe. My guess, based on hot air and other fantasies, is that conversations in one or more campanies went something like this...

- Intel needs to fab the Itanic - Intel needs a 64-bit successor to the x86, running 32-bit and 64-bit x86 code - Itanic is 64-bit - Itanic will be the successor to the x86, running 32-bit and 64-bit x86 code

But as I say, that is supposition.

It is, however, true (as far as I can see) that intel's x86-64 was delayed because the Itanic was to be The Only

64-Bit Intel Processor.
Reply to
Tom Gardner

I think that is inaccurate. The picture was:

- Intel had a 32-bit processor, the x86, now revisionistically called x86-32, but at the time no x86-64 was even envisaged, at least in public. - Intel needed a 64-bit processor and designed Itanic, that was supposed to also run x86 code. It got poor acceptance in part because it ran legacy x86 programs too slowly compared to real x86's. - AMD didn't have access to Itanic designs, so it did amd64 (now called x86-64) instead. AMD 64-bit processors also ran x86-32 code very fast, much faster than Itanics of the era. - Result: amd64 took off, Itanic sank, Intel took up the x86-64 design, though AMD actually stayed competitive in the high end CPU market for a number of years. Sadly, they have mostly dropped off since then.

Reply to
Paul Rubin

This is where you deviate from reality.

AMD64 didn't "take off" until *after* Intel adopted an x86 compatible 64 bit extension. Until then it was largely ignored.

AMD being competitive in the x86 market was only at certain times and only for a short while when they were able to carve a technical advantage which made them the fastest processor on the block. I know this very well because I watched it closely from the perspective of the stock market. Most of the time AMD is not only unprofitable, but spectacularly unprofitable. They are currently loosing a quarter of their market cap each year!

These days AMD has little hope of regaining profitability. They may find some margin in the ARM processor market, but it is very crowded with a large number of vendors. How much money can you make this way when you had a hard time making money against one competitor?

--

Rick
Reply to
rickman

Intel envisages and develops many thing that aren't seen in public. Of particular relevance to this are Andy Glew's postings in comp.arch.

HP designed the first Itanic, starting in 1989. Intel didn't get involved until 1994, and HP was the driving force in the first Itanic released 2001, and was still a significant force in 2002.

The real advantage of amd64 was the memory architecture and HyperTransport links, which were reportedly inspired by DEC's Alpha implementations.

Agreed, though not entirely for the reasons you have stated.

Reply to
Tom Gardner

Right, I bought a lot of AMD64 servers because of this performance advantage, not because of 64-bit capability. We did run native

64-bit apps, and IIRC some commercial DBMS SW used 64-bits to good effect well before Intel responded with competitive offerings.

Too bad that AMD couldn't stay competitive; looks bleak...

Reply to
Dave Nadler

extension. Until then it was largely ignored.

You seem to believe that the Wintel market is the only market for the X86. That may be the largest market, but Sledgehammer took off in the server market while HP/Intel was still pretending that the Itanic was The Way Forward and hence there was no need for i86-64.

Reply to
Tom Gardner

Yes, you learn fast. It is the only market that matters.

Great, where are they now, losing a quarter of their market cap every year with no hope of change in sight? AMD had some insights into how to build processors that gave them small edges for a short time on several occasions. In each case when Intel came out with their response, AMD had to eat dirt and watch their company shrink.

AMD is on the path to oblivion. There is always a chance that Intel will screw up so badly that the ARM will take over from them. There is always the chance that AMD will somehow pull a rabbit out of their hat and ride the wave of ARM processors to the top. But this is not at all likely.

Don't focus on a few things AMD did right and think they are a great company.

--

Rick
Reply to
rickman

already

This is starting to sound a bit like moving the goal posts to me. The very first Itaniums were 32 bit (mostly server machines) in a 16 bit consumer world. The biggest difference is that x86 was never designed to go to 32 bit or 64 bit but the Itanium is(was from the start). Also the Itanium was the fashionable design at the time (VLIW internal architecture).

?-)

Reply to
josephkk

There were never any 32-bit IPF machines, or even a defined 32-bit subset of the ISA (unless you want to count the x86 emulation).

Reply to
Robert Wessel

It's a flop. It was a flop from the start, and it remained so ever since. You can ask Oracle.

As you say, it gets used in long-lived servers. And arguably it is quite good for such systems, along with chips like IBM's Power family.

But the Intanium was designed to take the "normal" server and workstation market as well as the big-iron market. It almost completely failed in these areas - it was only with HP's intensive pushing (in particular, pushing it onto VMS customers who were perfectly happy with Alphas) that it got anywhere at all. And since these are long-lived servers, of course they still have them, and still need upgrade and maintenance plans.

Reply to
David Brown

AMD started its current (long and steep) downhill slide because Intel finally woke up and saw that AMD were producing far better chips at a lower price. Intel copied AMD's good ideas. And since Intel are bigger, richer, have always had a bigger market share, have better manufacturing facilities, and have a large, well-funded and very clever design time, the end was inevitable - Intel made better chips than AMD by using AMD's ideas.

The main ideas Intel copied from AMD were (in no particular order):

  • x86-64 ISA

  • Integrated memory controller

  • Decent multi-core

  • Doing more per clock cycle, rather than just higher clock rates (i.e., killing off the P4 line)

  • Doing more per watt (the "Core" architecture, from which current Intel chips evolved, was derived from the mobile Pentium M)
Reply to
David Brown

with no hope of change in sight? AMD had some insights into how to build processors that gave them small edges for a short

response, AMD had to eat dirt and watch their company shrink.

screw up so badly that the ARM will take over from them. There is always the chance that AMD will somehow pull a rabbit out

at all likely.

Since you have changed the points you are making, I am in broad agreement.

To paraphrase what you are saying: - Intel screwed up and started going down several blind alleys - AMD avoided the blind alleys and went in a better direction - Intel realised its mistakes, recovered, and started going in a very similar direction to AMD - Intel has negated AMD's leads

Never bet against Intel's process technology, but in other respects their technology is as fallible as other companies. e.g. - iapx432 - i960 - i860 - 8086 memory addressing (much better since the 80386) - clock speed is everything => pentium 4 - Itanic (good salesmanship by HP :) - de-emphasising power consumption as a key metric

Intel's mastery of process technology has been a key factor in their success; other aspects look like passing good fortune!

Reply to
Tom Gardner

That's an exaggeration - but only a very small exaggeration.

Not a good point: other factors are dominant w.r.t. Oracle

Agreed.

Of course HP pushed it (!), but your statement makes me think you come from a DEC background. HP's target market was the next generation of HP-UX on PA-RISC machines.

Some industries (think nuclear) are advertising for people to look after systems that will run until the 2050s -- on the PDP-11!

Reply to
Tom Gardner

Yes.

No. Are you confusing Itaniums with Spectrum/PA-RISC machines?

Reply to
Tom Gardner

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.