[hapless]

Context is important.

All of those competing architectures were becoming prohibitively expensive to enhance.

Alpha never paid for itself, ever, the market share was too small. POWER and SPARC had multiple failed projects with enormous capital costs. MIPS hit a performance dead-end with such low market share SGI saw no way to rebuild. HPPA was bleeding HP dry.

Itanium may have hastened their demise but all of these archs had the same core problems with investment returns. The writing was on the wall, and Intel was offering something people desperately wanted. (Unfortunately for everyone involved, it did not pan out.)

[hakfoo]

I wonder how much of that was each one being largely one-vendor operations. Even high-margin markets like servers and HPC can't compete with selling a billion cheap chips for white-boxes.

If there had been an "Alpha Inc" akin to the ARM Ltd model, I wonder if the platform could have survived longer. It might be able to tap firms that want to play in the high end market, but weren't interested in buying from a direct competitor (picturing those big beige Dell Poweredge PII cubes, but with Alphas in them), or they might have gotten a better deal out of stuff like AMD using the EV6 bus for the Athlon.

Aside from that, I suspect there was a significant aspect of vapourware to the Itanium strategy. Peak Intel had great manufacturing process and an endless bankroll; it was easy to assume that they'd deliver a product that would be impossible to compete with, may as well give up on another architecture. By the time the Itanium product shipped and everyone saw what a lemon it is, it was too late to reallocate the resources and make up for years of lost effort.

[mepian]

Actually there had been an "Alpha Inc": https://techmonitor.ai/technology/alpha_processor_inc_offici...

[panick21_]

I that understand that context. But even so.

POWER and SPARC were continue to be developed. And its hard to argue that either of them should have given up on their architecture. POWER is still developed. It probably made sense for Sun to stop working on SPARC eventually but not when Itanium was announced. Sun problem was just that they spend quite a bit of money, they were just never actually very good at designing processors.

SGI was the most aggressive on telling everybody that Itanium would be the future. And they paid for that. They massively delayed MIPS upgrades (only to then start it again once they realized it would take a while for Itanium to come to the market). For a while after this they still made their money on MIPS and all their attempts to push Itanium pretty much fell flat. So arguable it was smart for them to plan to eventually dump MIPS but now how and when they tried to do it (Also I don't disagree that MIPS was a deadend).

The deal HP got would have been fantastic if Intel had adopted PA-RISC 64-bit instead of Itanium. Given what happened HP wasn't really competitive with Sun 64-bit SMP servers and Sun made a killing on those things before the bubble. Once they bought Compaq and for a while were selling SMP servers based on Alpha.

HP really should have continued to push Alpha after the bought Compaq. The already had VMS ported to Alpha and a large captured base willing to pay a little extra. They aggressively ported VMS to Itanium and thanks to the insane deal they made Intel had to manufacture them processors for a decade+.

They are a large enough company and Alpha next few version had the chance to be really great, with things advanced vector extension and they had basically had the best processor team in the world.

So yes, it was correct that all these companies wanted to drop their development cost, but doing so before you know really about Itanium and how good it was, that's questionable and messed up all the strategies. There was just an assumption that Itanium was gone be amazing. Granted, VLIW processors were all the hype. Sun also waste a bunch of money on VLIW processor technology in the 90s and unlike HP, they didn't it on Intel.

Gordan Bell tried to hook up Alpha with Intel and make Alpha the 64-bit architecture that Intel went with. That would have been quite a different history. But for various reasons this didn't happen, and instead Intel went with HP and their next generation VLIW idea.

P.S: It would have been pretty smart of Sun went in on Alpha in 1992, then they could have saved on development threw out the 90s. DEC had built up a huge fabs to handle Alpha but almost nobody bought it. They were really looking for a high value costumer and couldn't find one.

Edit: > (Unfortunately for everyone involved, it did not pan out.)

Well it wasn't unfortunate for everybody. Sun made quite a bit of money for a few more years with SPARC. And so did IBM with Power. I think those two companies were happy with the Itanium failure.

[lizknope]

I remember when Rick Belluzzo became CEO at SGI. He immediately started pushing Windows NT and Itanium abandoning IRIX and MIPS. A lot of people wondered why and it seemed like he was on the Microsoft payroll. Then he left after about 1.5 years in 1999 to join Microsoft. The whole thing still seems fishy to me.

https://en.wikipedia.org/wiki/Richard_Belluzzo

[panick21_]

He pushed the same thing at HP as well. And the left Microsoft soon after as well.

I think its more like that he was brought in by the board because those were his opinions.

The board was sick of paying for development of MIPS and IRIX. The believed their future was massive multicore Itanium Windows IT systems. And because they wouldn't do any of the development but would make the same kind of profits, they would just print money.

This of course backfired in a whole bunch of different ways.

[hapless]

Sun's last successful SPARC project was the UltraSPARC II, in 1997. Their subsequent projects, like "Rock" and "Niagara," were disasters. (USIII and USIV were little more than core shrinks of USII)

If Sun had given up the day Itanium was announced, they mighta come out ahead!

[panick21_]

Well technically successful maybe. But lots of people still bought their serves even after the bubble. And those server were expensive with good margin.

Giving up instantly when Itanium was announced would have just made them a company that waste a lot of money a lot of money on porting things to Itanium (and Sun had quite a lot of software that's not exactly easy to port). Only to not have a 64-bit system to build servers from. Essentially at best offering good x86 32-bit servers for the next couple years.

To be sure x86 workstations and servers should have been a big part of their strategy already in the late 90s. But that's not the same as adopting Itanium.

SGI did exactly what you suggest, give up on their chips and OS. This was terrible choice. Because Itanium was very late they had to restart MIPS development and were not able catch back up. Their Itanium products all didn't sell well even when they finally arrived.

[sillywalk]

They also didn't really invest in a next-generation graphics system. Just a rehash of the InfiniteReality.

SGI also did have X86 workstations, first with a proprietary SGI chipset that required as special HAL for NT/2K, and then just a standard PC workstation.

[hapless]

SGI was already moribund in 1997. Executives made a conscious choice to pivot. It didn't work out, but it is real hard for me to believe they could have done better with a redoubled focus on MIPS development they could not afford.

[sillywalk]

Yeah, in 1996 SGI bought Cray for $740 million. They got CrayLink/NumaLink, but also a big 64way SMP SPARC machine, which they sold to Sun for "significantly less than $100 million." Which became the Sun E10K, which made a lot of money.[0]

Then a bunch of the graphics people (who did the GPU for the Nintendo 64) went to found ArtX [1], which got bought by ATI.

There was also the ill-fated Fahrenheit project.

[0] https://www.forbes.com/2002/05/06/0506sun.html#703713c16a5e [1] https://en.wikipedia.org/wiki/ArtX

[panick21_]

Pretty sure 'CrayLink' was actually developed by SGI and just branded 'Cray'.

[panick21_]

They were forced to redouble on MIPS because Itanium was so late anyway. And even after Itanium came out making products around it took a while.

And I'm not saying not reacting to Itanium means go all in on your existing ISA and your existing business model. That wasn't my argument.

My argument was don't go all out on everything that currently makes you money and your costumers expect for a totally unproven architecture that even if it works out means that you are just gone be a commodity provider with commodity software. Something totally outside of the whole history of Silicon graphics.