Page 1 of 2

SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Fri Jan 27, 2017 5:24 am
by Dodoid


Enjoy!

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Fri Jan 27, 2017 11:12 am
by Shiunbird
I love the 3D printed logo!

"More sluggish than a Macintosh." --ouch

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Fri Jan 27, 2017 11:22 am
by foetz
it's too short for covering those years properly. there's too much missing

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Fri Jan 27, 2017 12:15 pm
by Dodoid
foetz wrote:it's too short for covering those years properly. there's too much missing


If you feel I should release an addendum covering things I missed, that would be considered. What in particular do you think should be included? I try to keep my videos short(-ish), but if I missed something important I would cover it. Mind, 1996 products (O2, Origins, Onyx2) are to be covered next episode.

I know the InfiniteReality was released, but I tend not to cover individual hardware components unless they are really important later on (I mention the R8000 and IMPACT because the R8000 had floating point errors in 1998 and IMPACT is repurposed for the Octane and resulted in an Indigo2 color change. WebForce was later (1996, from what I can tell).

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Jan 29, 2017 1:09 am
by vishnu
Well I liked it! Maybe you could have said something about SGI founder Jim Clark bailing out at the end of 1993 for (insert multiplicity of reasons here), but that strays into board politics and your focus is technology, so never mind... :)

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Fri Aug 18, 2017 2:57 am
by DDT
Dodoid wrote:
foetz wrote:the R8000 had floating point errors in 1998


What kind of errors ?

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Fri Aug 18, 2017 9:03 am
by Dodoid
DDT wrote:
Dodoid wrote:
foetz wrote:the R8000 had floating point errors in 1998


What kind of errors ?


https://vizworld.com/2009/04/what-led-t ... chapter-3/

Though this is unfortunately not the original source (the original archive has been deleted and Wayback Machine doesn't have it), it quotes an archived email from inside an Australian university where this is discussed.

SGI has offered to upgrade Octopus from its existing 90MHz R8000 CPU
configuration to 195MHz R10000 CPU's for no cost to JCU. The R10000 is the
current generation processor used in all SGI's high end machines. SGI have
made the offer to two reasons:

o First to ensure that a rare and specific bug in floating point
calculation that has been found in some older R8000 systems does
not become an issue at JCU. I have run tests on Octopus and
it would seem that this bug is not an issue for us at JCU.

o Second to upgrade the Power Challenge so that remains at a modern
level of performance and compatibility.

All in all this is a generous offer from SGI (valued at ~$300,000) that we
should gladly accept.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sat Aug 19, 2017 6:21 am
by dexter1
There also is one hint in the manual page of 'mipscheck':

Code: Select all

2. The Irix 6.2 (and later) operating systems for r8000s will
        automatically patch any running program to remove the prefetch
        instructions.  This will not affect the performance on an r8000 but
        it will avoid the r8000 prefetch problem.  In rare cases, the
        kernel will not be able to avoid the problem and will request that
        the user run the binary through r8kpp to execute the repair
        permanently.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 12:32 am
by DDT
Dodoid wrote:
o First to ensure that a rare and specific bug in floating point
calculation that has been found in some older R8000 systems does
not become an issue at JCU. I have run tests on Octopus and
it would seem that this bug is not an issue for us at JCU.


Uhm... I assume we are talking about the mysterious "prefetch bug" described in the intangible "R8000 errata sheet" (no one seems to have a copy of).
Note that prefetch instructions are not technically FPU instructions. Maybe the bug will affect FPU calculations if prefetch instructions are used. It would be great to have a copy of the errata sheet.

The other "problem" of the R8000 FPU is that the MADD instruction is not compliant with IEEE specs. The reason being that it is too precise. This is actually not a bug but a feature, but might cause problems when strict IEEE compliance is required.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 12:50 am
by Irinikus
The R8000 seems to be an interesting chip! I would seriously like to have a system with one installed someday.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 4:18 am
by dexter1
Irinikus wrote:The R8000 seems to be an interesting chip!

Except that it is not actually a chip but a processor board with two MIPS CPU's, cache and glue logic. Have a look at
R8000.jpg

That is a picture of an actual R8000 processor board.

I would seriously like to have a system with one installed someday.

Well, an R8000 certainly isn't a boring system and they can be a bit temperamental. Also they are rare: You find them only in Indigo2 systems (IP26) and Onyx/Challenge (IP21) desksides and rackmounts. The prefix "Power" is commonly used for these systems.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 5:19 am
by Irinikus
dexter1 wrote:Except that it is not actually a chip but a processor board with two MIPS CPU's, cache and glue logic.


That's an interesting card, very different to the other CPU options available for the said machines.

I'm already in the process of trying to acquire one!

If I manage get it, I will install it in an Indigo2 Impact machine, when I eventually acquire one.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 5:36 am
by dexter1
Irinikus wrote:If I manage get it, I will install it in an Indigo2 Impact machine, when I eventually acquire one.

Alas, you cannot "sidegrade" an R10000 Indigo2 Impact (IP28 motherboard), or an R4400 Indigo2 (IP22 motherboard) for that matter, with an R8000 CPU-card. The motherboard must be an IP26 a.k.a. "Teuton" and its PROM revisions only support pre-Impact graphics such as Newport(XL) and Express(XZ/Extreme)

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 5:41 am
by Irinikus
Ahhh!!!, the seller is also selling an IP26 motherboard, I guess that I will have to buy that of him too then. :shock:

I suppose that I can install the motherboard and CPU my current Indigo2 then.

Re: SGI History Part 3 is here! This episode covers from 1993 to 1996.

Posted: Sun Aug 20, 2017 8:09 am
by dexter1
Irinikus wrote:I suppose that I can install the motherboard and CPU my current Indigo2 then.

Yes and that's pretty much it. The Power Supply is relatively interchangeable between R4K systems and R8K, see http://www.oocities.org/siliconvalley/h ... upply.html

The rest can be swapped over from the R4K Indigo2. Don't forget to handle the R8000 CPU carefully, since it's sensitive to static discharge. Oh and make sure you secure the R8000 CPU correctly on the motherboard with all screws in place, since the standoffs provide power and ground to the module, see viewtopic.php?f=14&t=16726745&p=7351542&hilit=miod+r8000#p7351542