SGI: Discussion

I'm well aware of it - in what way does that change the fact that Columbia is still very technically advanced? Or that SGI would not have landed the deal if it was O3k instead of Altix?

I'd be very surprised if any of the bidders submitting quotes for Columbia asked for even 50% list price.

Once you leave the mass market toys and get to the big iron (16 way and upwards) no-one, ever, pays list price. At all.

'Headliner' systems like Blue Gene/L and Columbia are always *heavily* subsidised by the vendor. TOP500 is bragging rights, nothing else.

However, SGI could have given the kit away - for free - if it was MIPS based, and NASA still wouldn't have gone for it.

There's no loyalty in IT, and no-ones going to install a system that's significantly slower than the others on offer - especially not for a major installation like that.



You appear to be confusing SGI with a mass market player like HP, Sun, or IBM.

They are not. They never have been, and they never will be. They are a niche player. They rule that niche, but it's still a niche.

Itanium is a niche CPU - anyone who believed it would be a mass-market solution can come to talk to me - I have a bridge to sell them. The market for Itanium is tiny - just as the market for MIPS or vector based systems is tiny.

The Earth Simulator was more than 10% of global vector-based systems sales when it was installed - doesn't mean it wasn't technically impressive, or that NEC were giving away kit out of desperation to stir the market.

High end system sales are very different from bulk desktop or 4 way server sales. Trying to quote numbers or systems shipped is just a nonsense in that space - that market doesn't work like that.



You don't honestly believe that? I mean, that's some serious tin-foil-hat paranoia going on there.

A post in a web forum does not a business decision make. Definitely not when it would involve hundreds of millions of dollars.

IBM will not buy SGI. At all. It's just not going to happen. Quite apart from the fact that SGI is dwarfed by IBM - an utterly insignificant player - IBM already have products in every area that SGI do.

Why on earth would IBM spend money to buy a non-competitive niche player when they already have higher volume sales from competing products?

Do you think IBM sales lose sleep at night when thinking of Altix vs. their POWER5/Linux business?

There's no business justification for it at all.

And that's not even touching on all the regulatory pain IBM would incur from the US goverment. Look at the hopes SGI had to jump through when buying Cray.

Governments don't like to see their technology suppliers being bought - it doesn't give them the warm fuzzies. They'd much rather bail them out with lucrative research grants and large scale projects. Look at how badly HP have screwed government departments over with their laughable Alpha 'road-map'.

The only people who might have had an interest in buying SGI would have been Sun - they need the high-end scalability. But not even Sun were willing to do a complete about face and swallow Itanium as their server CPU - so they jumped into bed with Fujitsu.



Which was what my post was all about.

SGI are still innovating, they're still ahead of the pack, and they therefore still have their little niche cornered. The day SGI stop innovating, you'll know about it, because they'll go under.

In the meantime, their stuff is technically excellent. The fact that it's no longer MIPS/IRIX should not blind any of you to their continuing innovation and - most importantly - their continued survival in a very harsh market.

Cheers,
TOM

Prism uses SGI's new Tio I/O chip which supports multiple AGP 8x interfaces.
Onyx4 UltimateVision used PCI to interface AGP cards at 1x.

It sounds like Dorado is just a workstation version of Prism.

Unless you need the NUMAlink kungfu, a dual Opteron + dual PCI-E 16x board will probably do just as well...

Alexis Cousein did some posts in the filth that is comp.sys.sgi detailing why this was.
It basically boiled down to the long development lead time for stuff like this (ie. AGP 8x was there and usable, PCI-E was a glint in some working group's eye) as well as stable and well-tuned chipsets surrounding the AGP implementation of the ATI cores.
I believe there was also some development issues/costs associated with the XBOW->PCI->AGP bridges - as in AGP is just more crap on top of PCI, whereas PCI-E is fundamentally quite different.

Cheers,
TOM

Well, personally I'm looking forward to Dorado. Looks like it's gonna be the cheapest SGI yet, and they did a pretty good job with O2 as the cheapo, who knows what they can do with this! I may just have to buy one.

I notice a lot of people will scream for Tezro, but will slag off the IA64 machines. Hey, they're powerful as hell, and rumour has it that the caches on the new Itaniums will be HUUGE! OK, so Dorado is supposedly gonna have 3MB cache per chip, but that's cool for now. I'm just as pissed that we've lost MIPS and IRIX on the roadmap, but that doesn't mean I can't still use em. I still got 4 perfectly working IRIX/MIPS machines here, and just cos the new machine will be IA64/LINUX, doesn't mean they'll stop running.

I for one will look forward to having a play with Dorado, and if I like it, I'm having one ASAP!

Go ahead, flame away, I can take it!

Good attitude!
Cheers!

Actually for the time being AGP is still more optimized than PCI-E. This is especially true for Linux. AGP has a facility, the aperture that can be marked as "write-combined" with video memory and system memory that has been allocated to the aperture. PCI-E lacks the aperture function which makes it perform slower even at 16X in certain situations. PAT (Page Table Attribute) support is making it's way in to the 2.6 kernel but has not been released yet. PAT functionality can decrease the system bottleneck associated with heavy PCI-E/memory bus dependent traffic. Once it's fully supported PCI-E will be a lot faster than what it is now.

I've got to ask though... is anyone actually considering getting one of these things?

Should find out more about it when SGI show it on Tuesday as I understand it

I'm interested in seeing what kinds of software you can load on it, and how much it costs, before I say whether or not I'd want one...

As I understand it, SGI is pretty bad at listing specific applications that you can run on your hardware... It's almost as though people just get SGI hardware because it's awesome

At least Sun lists an included productivity suite

Do you think that the release of this new Dorado workstation might end up shaking out some of the newer Octane2's and older Fuels? Even shaking out a few O2+ units at the very end of what some people probably consider useable in a professional or university environment would be a good thing for the hobbyist community.

didn't several early fuel configs showed up in the 2nd hand market recently? also i believe it's only a matter of (short) time until fully loaded octane2's show up more often. quite some of these should have been replaced by tezro's till now.

the question is, will there be enough quantities of these machines to allow prices to drop into the "affordable" range in general? i somehow have a hard time to believe they sold alot of their newer desktops. but then, it was/is surprising to see how many low-end octane2's were dropped when considering that a lot of the irix->pc transition already took place in the late nineties.

so i'll try to stay optimistic. i'd prefer a dual-600 octane2 all day over a fuel for several reasons though, hope that is the more cost-effective route as well


edit: as for ia64-based machines: to me these seem not very interesting as of now. only reason for me to use irix/sgi is software-availability anyway. makes a nice addition to the windows software world.
a lot of graphics related software has vanished already. i don't see them coming back onto a niche platform unless sgi manages to offer something truly unique (again). and computers without software, well... that's something for the collectors, if at all.

Unless SGI offers something super compelling, some awesome reasons to buy one, I wont. I just don't see the benefit (yet).