Oracle ordered to pay HP $3B in Itanium case
reuters.com
reuters.com
The revisions to the sales forecasts from market research companies were ugly https://en.wikipedia.org/wiki/File:Itanium_Sales_Forecasts_e...
And it killed MIPS and ALPHA architectures on the server, but I guess ended up pushing everyone towards x86 while ia64 was delayed until AMD jumped in an released x86-64.
About the only (major) vendor that uses Itanics is HP, and when Intel decides they no longer want to build them, HP is in trouble, unless they manage to get The Machine ready first, which I am not overly optimistic about (but happy to be proven wrong).
Once Intel decided to switch to x86-64, it does not look like they paid a big price to catch up (thanks to cross licensing, they could easily get the IP, and they still had their manufacturing strengths), so it's not clear to me that Intel faced ANY disadvantages in the long run (as opposed to their partners).
Far as their failures, I think the reason is that they always do something simpler and better but then add some weird stuff that screws it up. Gotta maintain backward compatibility where possible plus make compilers easy. A great example of doing it right would be i960 w/out the BiiN crap, with C compatibility, and x86 emulation a la Transmeta or Loongson. Instead, they did no x86 or C compatibility plus Ada support. Countering market demand killed it even if market was being short-sighted.
Better approach is Cambridge's CHERI which provides both benefits and tooling compatible enough for FreeBSD:
https://www.cl.cam.ac.uk/research/security/ctsrd/cheri/
Another good example of doing it right is Hardbound which experimented on x86 ISA:
https://www.cis.upenn.edu/acg/papers/asplos08_hardbound.pdf
These two both drastically improve an important area of CPU's, mainly security or reliability, while maintaining as much compatibility as possible with existing investments. The CPU's and compilers do most of the work actually. Many of the good designs also modify only a small part of CPU investment and/or compiler to reuse those too. So, my recommendation for Intel is to stop trying to get clever (Itanium, SGX) in favor of copying stuff that works, makes sense, is backward compatible, and has major reuse of x86.
I wonder if deprecating and eventually removing the ugly parts of the architecture over time would work for Intel (even if they had to bifurcate the architecture).
What doesn't work for Intel (and its customers) is huge revolutionary new archs with straight-out-of-research advanced features and no compatibility.
http://dslab.epfl.ch/proj/cpi/
In prior comment, I illustrate with Hardbound they can do a lot more than this. It remains compatible at source-level with x86 apps with good performance while enforcing spatial safety. I could imagine a split version where certain components, esp binaries, run totally unsafe just for compatibility while it enforces safety on whatever you can compile. Alternatively, people wanting backward compatibility with no source could run with safety off whereas anyone (or vendor) with source could compile a version that works. Even the binary stuff might be isolated a bit with techniques that decompile, instrument, and recompile the binaries. Or separate, assembly-level methods.
So, Intel could do a lot more than it's doing without totally scrapping backward compatibility or performance. They just... don't. I think they're afraid of loosing billions again or just don't know about the stuff I'm mentioning. Or both. (shrugs) Love to talk to them about it sometime.
If it's x86-64, the processor running your laptop is AMD tech. Period.
Part of it made it's way into the AMD Athlons, the EV6 bus they used are from the Alpha.
Not too practical, though, unless aiming for low-cost given other holdovers (POWER, SPARC) have more software (esp FOSS OS's) plus state-of-art implementations.
It would take some time to sort of the issues, but given the Wintel stronghold on the desktop, everyone would be moving screaming into Itanium and feel happy about it later on.
But AMD gave people an alternative, so Itanium died.
The Itanium architecture and processor development was happening for a very, very long time (code name "Merced", IIRC) and didn't have much to show for due to incompatibility with x86.
Intel would announce x86 EOL and legacy apps would be emulated or JiTted on Itanium.
Their problem was that Intel wasn't the only company producing x86 chips.
So, there's plenty precedent that your scenario could've played out. Don't forget alternative, though, where Centaur/VIA could just receive a gigantic shift in business their way. Or AMD just keeps accelerating x86. Actually, AMD64 seems inevitable in this scenario.
Everyone would be forced just to switch architecture, they would not have any other option if they wanted to stay on Windows, lets say.
Of course this wasn't at all possible, given that there were companies like AMD and VIA producing x86 chips as well, thus making way for an escape route in such scenario.
We do enterprise (DAX level) consulting, so I do get to ear such stuff.
http://www.oracle.com/us/corporate/features/itanium-346707.h...
It's weird that companies that companies who compete so fiercely in the same markets really ever manage to work together. Once Oracle became a systems manufacturer HP had to know that they would boff them.
From a technical perspective, the stuff that HP has in Nonstop, OpenVMS, and Super Dome are really impressive. I read that they sold OpenVMS to a third party that is busy porting to x64. Looking forward to running it in VM's for fun...because I have a weird sense of fun. Stop judging me.
"Don't make the mistake of anthropomorphizing Larry Ellison"
So does this mean they actually continued to work on it, but just told HP there was no market and had stopped work?
It was an extremely effective tactic by Oracle to get customers to move away from HP... extremely dirty, but effective.
/ˈSHädənˌfroidə/
noun
pleasure derived by someone from another person's misfortune.
Schadenfreude is an obvious borrowing from German. His question is why did they have a word for it in the first place.
On the other hand, I think I remember reading a couple of years ago that per CPU core, Xeons were already beating Itaniums in performance for database workloads. The main advantage apparently is that you can stuff lots of CPUs into an Integrity server, but mostly a cluster of smaller x86_64 machines has a good chance of being faster and cheaper.