Cheap PPC Linux Machines From IBM

[Floid]

CodeSmith wrote:
Although OF is a pretty good idea, you're a bit limited in the gfx cards that you can use (ie "Mac" gfx cards), and they are usually more expensive.  It might be a good idea to add a UBoot driver to AROS, so you can have both.  With UBoot you can use cheap "PC" gfx cards (it's got a built-in x86 emulator so the card's BIOS can run on that).  Since it's on SourceForge (or if not, should be by the time AOS4 is released), it should not be too difficult to find out what you need.

It depends on the OF, obviously; whoever wrote Genesi's has included support, so far as I know.  The question will be whether IBM follows suit, or decides these are destined for the datacenter anyway, and don't make such concessions.

The 'rack/deskside model' would have to have a graphics option of some form - either a generic onboard chipset, if it's a pricey pizza box, or a high-end solution if it's designed for visualization work.  (I'd be most surprised to see a high-end 'consumer' chipset, but the PPC division *have* collaborated with a certain ATI in the recent past...)

Also, as far as I know, there's not much stopping the appropriate emulator (or just direct driver code) from running in software.  Apple doesn't want to do it because it's a kludge, and would further discourage vendors from producing Mac versions of their product ('Works with Mac!' shelf presence is an important advertising tool, as always)... but if you don't mind keeping the original card in for the first display/boot-time diagnostics/OF-hacking-or-poking-with-sticks, I don't think there's any barrier to firing up a 'generic' Wintel card after/while the OS loads.  (Seriously, anyone know if this works under, say, YellowDog or Debian kernels on a Mac?  I could swear it does.)

If it ends up PCI-X and AGPless, as many 'server' boards are appearing these days, it's not like you'll lose much of anything picking one slot over another.  (As an example, here's an Opteron board with a diagram of its slot layout.)

Now, the question is whether OS4, or MOS for that matter, will have SMP support worth the port...

[Floid]

Hm, addressing one of my own points, if IBM's recycling any of Apple's chipset work:

"Three 64-bit PCI-X slots let you add one card running at 133MHz and two cards running at 100MHz. Three 32-bit PCI slots allow you to add three 33MHz cards." -- From the Mac G5 'Expansion' page.

But then, they've got AGP 8X onboard.  Yet again, no way to predict if IBM will follow... or even if they'll cripple the functionality just to please Apple, if they use the same chips.

As contrast, the Rage XL in the Tyan is hung off the PCI bus, since it's just a server design, using AMD's 8111 (HT->PCI + integrated peripherals) and 8131 (HT->PCI-X), without the 8151 (HT->AGP) companion chip.  (Diagram of AMD's architecture here; Via and NVidia are taking more integrated paths, since they haven't had to rush about it, and don't need to demo Hypertransport's ability -- Via's design uses their proprietary V-Link anyway, and I've no idea what NVidia's ever been up to.)

[Floid]

CodeSmith wrote:
The voodoo3 works because 3DFX were a really great company and provided all sorts of info needed to write a complete driver.  ATI is like that too (hence the OS4 Radeon drivers from Forefront and the MOS drivers you're using), but there are many other companies (nVidia comes to mind) that refuse to hand out hardware documentation unless you promise them loads of cash first.  For those cards you need to be able to run the routines in their BIOS.

Well...  For one thing, it's a non-issue, because SciTech's taken care of it no matter how obfuscated they've made it (yes, only for 2D, but that's what you need to 'run' an OS), and in any case, I'm fairly certain XFree86's support - again 2D - extends to the level necessary, even if it's not the prettiest 'documentation' to follow.  That said, I haven't looked at what, say, FreeBSD or NetBSD do to probe and initialize the cards, as it's a bit too 5AM for that, and I wouldn't know what to look for anyway.

XFree source here.  It's one thing to claim it's technically impossible, it's another thing to say "sure, it's technically possible, but who in their right mind would go through the trouble when the vendor's 'hostile?'"

I can muster some sympathy for NVidia, since unlike ATI, they're still something of an upstart, and I gather their financials have yet to even out - but yeah, until they learn to play fair, I'll continue wishing Matrox had a product up to par.

Point is, recycling an old TNT2 or something for 2D while you look for a 'supported card' should at least be possible, and if there'd be issues with Linux on the IBM hardware, they'd be wise to solve that right quick; at least IBM's PC division is still a customer.

As to Mai and Marvell... are you serious?  While there's always the possibility- and it seems like the PowerPC divison wouldn't mind farming out development to such third parties if they could keep up- the 970 is, to put it lightly, "hella new," and unless either company have been involved in the R&D of same, Apple and IBMs chipsets will be the most advanced, and the reference for others to follow.  (Something that never quite worked out for PReP/POP, but it smells like the PPC guys has been willing to learn something from AMD's recent rise.)

Remember, these things don't use the MPX bus of the earlier PowerPCs - it's as big a leap as from Socket 7 to the Alpha EV7 bus used on the Athlon, or from EV7 to Hypertransport on the Opteron/A64 - and as far as I can tell, no civilian knows whether Apple, IBM, or someone else's name is on any of the support chips used in the G5 Macs.  (Heck, from MacOSRumors.com -- "For most Mac users, the biggest news was that estimated shipping dates on PowerMac G5 pre-orders have changed from 'On or Before September 2nd, 2003' to 'On or Before August 29, 2003'" -- looks like we've still got a while to find out.)

---

@mschulz:  Thanks for the info; I wasn't aware (or just plumb forgot) that anyone'd bothered with the emulation before this round of boards.