[spiroyster]

The reason the Visual Workstations were mediocre at best when they were released is because the project got delayed so bad, by the time they were released, PC's with discreet graphics cards were almost as powerful, but way cheaper.  Within days of the VW release, the entire department was canned and the project canceled.

At about the same time, my department was developing PC based Image generators for the next product.  Which got canceled when they started having other financial issues. The Fuel owes some of it's existence to that work.

Due to the lawsuit SGI won over Nvidia for stealing their graphics technology, we were getting the highest performing graphics cards cherry picked from manufacturing to go into those systems. A year later most of the SGI graphics engineers were working at Nvidia (in fact a ton of SGI folks ended up at Nvidia and many are still there).

Has anyone heard of the Fahrenheit project? It was another legacy of Rocket Rick.  He forced SGI to work with Microsoft to combine OpenGL and their graphics API's and when things got close to being done Microsoft pulled out of the deal and took all of the IP with them. Not too long afterwards, Rick ended up at Microsoft.

https://en.wikipedia.org/wiki/Fahrenheit_(graphics_API)

^^The meetings I was in contradict some of the information in that wiki page.

I'm so glad you posted .

I heard rumors of the entire VW team being 'let go' either before or just after release of the VW on nekochan eons ago, but have been unable to find anything in relation to it. It's good to hear it confirmed. Shame really... a 320 was the first SGI I got (mainly for Maya 2.0 on Windows)... ironically my first job was then working on a O2/Irix platform. o.0

Did you work on it? Is there any techincal information about Cobalt anywhere?

I would be interested to know what contradictions there were. M$ were also working on their 'Talisman' prior to Fahrenheit... which fell by the wayside. Unsure how these two fit in together as I think they were around the same time.

It would be nice if we could have a subforum for the VW, the poor SGI/M$ bastard love child of that time... Seems to get little love from SGI fans, and x86 fans alike, but I think it is iconic being one of the first integrated graphics systems around, with outstanding 2D bandwidth... and a sexy as f*ck case

Any information you have would be greatly appreciated.

[airbozo]

I did not work in the same department. At the time I was in the Advanced Graphics Marketing Division and we were responsible for the higher end products and labs. We did work with them on some of the demos and marketing stuff to support the roll out and tradeshow logistics, but that was it.

I ended up with a brand new 540 when I was going through my second layoff. It lasted for about 6 months, then things started breaking and I couldn't afford to replace things since CPU's had to be bin matched and tested with VR's until it worked right.

When I was working at the SGI refurbishing company we could not give away the systems but the memory, CPU's and VR's were always in high demand.

I know it's taboo, but I ended up modifying an empty chassis to fit a standard motherboard. I used it as my CAD machine for years before I tossed it, because it no longer fit in my home office. I think the case is still in my garage...

[MojaMonkey]

I know it's taboo, but I ended up modifying an empty chassis to fit a standard motherboard. I used it as my CAD machine for years before I tossed it, because it no longer fit in my home office. I think the case is still in my garage...

Let me know if you ever dig that case out and decide to sell it

I need help, I'm trying to get a 540 up and running, PA8600 in this thread was amazing and shared all the required driver files.

However after upgrading to PROM 1_0006 I cannot install windows 2k. I get an error message saying my PROM  is too old for Windows 2K.

I'm now stuck with no OS and no way to install one.

If anyone has any old VS 540 PROM files they can share that would be incredible e.g. PROM1_1005.exe.

Thanks so much in advance if you can help.

[Irinikus]

I know it's taboo, but I ended up modifying an empty chassis to fit a standard motherboard. I used it as my CAD machine for years before I tossed it, because it no longer fit in my home office. I think the case is still in my garage...

Let me know if you ever dig that case out and decide to sell it

I need help, I'm trying to get a 540 up and running, PA8600 in this thread was amazing and shared all the required driver files.

However after upgrading to PROM 1_0006 I cannot install windows 2k. I get an error message saying my PROM  is too old for Windows 2K.

I'm now stuck with no OS and no way to install one.

If anyone has any old VS 540 PROM files they can share that would be incredible e.g. PROM1_1005.exe.

Thanks so much in advance if you can help.

Make contact with Ian Mapleson, he will be probably be able able to help you out here.

I did receive a DVD from "pauliedweasel", who's also a member here, containing lots of programs for the Visual Workstation, but unfortunately, it's down in Cape Town along with my 320, so I unfortunately won't be able to help you out with this at the moment.

[GL1zdA]

Due to the lawsuit SGI won over Nvidia for stealing their graphics technology, we were getting the highest performing graphics cards cherry picked from manufacturing to go into those systems. A year later most of the SGI graphics engineers were working at Nvidia (in fact a ton of SGI folks ended up at Nvidia and many are still there).

This is often overemphasized. The truth is, only the VR3 had a significantly faster chip. Since I have the V3, VR3 and VR7 I can tell you how they compare to "standard" cards:

V3 - the GPU is clocked at 120 MHz and RAM at 150 MHz - reference values for a GeForce DDR
VR3 - the GPU is clocked at 140 MHz and RAM at 166 MHz - this one is faster then a Quadro DDR. A ELSA GLoria II DDR is 135/166 MHz.
V7 - don't have this one, but looking at the VGA BIOS it should 200/183 MHz which are reference GeForce 2 clocks
VR7 - the GPU is clocked at 250 MHz and RAM at 230 MHz - compared an ELSA GLoria III it has faster RAM (ELSA clocked it at 200 MHz) and same GPU clock. It's basically a Quadro2 with GeForce 2 Ultra clocks

I ended up with a brand new 540 when I was going through my second layoff. It lasted for about 6 months, then things started breaking and I couldn't afford to replace things since CPU's had to be bin matched and tested with VR's until it worked right.

I bough a nice VW540 several years ago (with SDI I/O and flat panel connector). I've maxed out RAM and I'm currently running it with 3 900 MHZ Pentium III Xeons (still looking for a fourth VW540 Pentium III Xeon heatsink). I've even forced Windows 2000 Professional to work with more than 2 CPUs And out of nowhere recently some RAM failed :/. I regret not having bought spares.

As much as I wanted to like the VW540, it really feels unfinished and it's a huge system. I ended up using as my retro PC mostly the plain VW550. It's a PC, but mine has at least Arrows F1 livery to make it special - apparently it was once used by this team.

If anyone has any old VS 540 PROM files they can share that would be
incredible
e.g. PROM1_1005.exe.

Thanks so much in advance if you can help.

Here https://www.dropbox.com/s/vztubc4ac8kzlb...5.exe?dl=0

[Irinikus]

Seeing that this discussion has reopened and I'm in Cape Town, and have access to my 320, I thought I would share some pics of the machine:













No matter what anybody says, these are really cool little SGI's and well worth having!

[mapesdhs]

So much inaccuracy and FUD being posted here about the 320/540, probably because many like to imagine they were worse systems than they actually were due to their not being MIPS/IRIX. The reality is very different. I've personally seen a 320 run a hundred times faster than what on paper should have been a far more powerful PC (P4/2.4 GF4 Ti4600), the cause being the task in question (urban modelling and the use of large 16K composite textures, an application called Realax, used to prototype what was then run on a ReaityCentre). A 500MHz 320 ran the test scene in question at 10fps, whereas the PC could barely manage one frame a minute. The researchers were moving to normal PCs because SGi's CPU/RAM upgrade prices were crazy and there was no roadmap to a proper IVC replacement (otherwise, like the UK MoD had said, a more powerful IVC2 would have been very popular). So the dept. bought cheap Viglen P4 PCs with GF4s. Performance initially sucked because of the composite textures, so the devs had to rework all the models with separate textures, oodles of LODs, etc. in order to match the limitations of the bus-based PC design; one couldn't ask for a better example of how the PC bus arch could limit and define how a problem could be tackled, a technical ball & chain legacy the world has been saddled with for decades.

The 320/540 offered an architecture akin to O2 but with far greater pixel fill rate and geometry potential. Even for conventional 3D tasks, a max spec 320 could outperform potent PCs of the day at launch, and should have maintained this lead over time with better CPUs. I ran a post grad lab of a dozen 320s for several years so I dealt with these issues first hand. The 540 was particularly popular with defense companies, a custom model came very close to beating an Onyx2.

Where it all went wrong was terrible marketing, stupidly secretive tech specs, price gouging by resellers (the ultimate death blow to all things SGI I believe) and a failure to (as Irinikus said) stick to the original promised roadmap of a complete base tech refresh every 18 months. With no IVC2, the custom arch (and its associated advantages) was doomed.

I am of course at heart a MIPS/IRIX nut, but the 320 was the first Windows machine I ever saw that could load and display a 50GB 2D image file as fast and as responsively as an Octane (at the time I had finally obtained my own R12K/300 MXE).

Someone mentioned MIPS vs. IA64 adoption; this is a circular argument, because Intel was only able to develop IA64 once it had the talent pinched from DEC, HP, SGI and others. I had many conversations with SGIand other people about it back then (most not public). I was told by key people that their own estimates were that what SGI's SNx plans would be about a third faster than IA64, but a lot more expensive (this assuming of course IA64 came out on time). What Intel pulled was a clever and effective bluff; I can't fault them for that, as a business strategy it was genius, and it worked. Losing key staff held up MIPS development by several years (note the huge delay between R10K/195 for Indigo2 and R10K/225 for Octane) and of course knocked away Alpha and PA-RISC rivals.

I've often said myself that if only SGI had held out... but could they? We know now that they were having problems combining the Cray and SGI businesses, so despite the theoreticaly awesome combination of Cray's vector tech, MIPS V ISA and MDMX, we can never know whether the project would have come to suitable fruition. Rather, we choose to believe it would have because we want that to be the case, but really who knows. Besides, riding above it all is SGI's rotten reseller sales model which continued to hurt SGI to ever increasing degrees as the commodity market took off. Without a doubt, resellers were not interested in cheaper SGI products, the very thing SGI needed to be able to offer to compete.

We can speculate about whether the switch to a conventional bus arch for the 230/330/530 was a deliberate move by a CEO hell bent on bashing the company, but we can never really know for sure, not without some historical leaks of key documents, emails, etc. Although it certainly looks suspicious, it's not as if SGI wasn't already damaging the 320/540 lineup with insane marketing, crazy upgrade pricing and a refusal to disclose key technical information that's critical to any x86 purchasing decision (eg. max supported CPU config). IIRC it was an impatient SGI employee who eventiually spilled the beans about the dual-PIII/1GHz and single Celeron 1.4. SGI did this again with Tezro, not making the quad-1GHz properly public until after 18 months of pressure from the admin of sgi.com and myself finally got marketing to at last change their mind (ditto the R16K/900 for Fuel).

So with or without management machinations and dodgy plots, SGI was digging its own grave by then anyway with dumb marketing and a sales model that was not fit for the end of the 20th century, never mind the 21st. The debacle of the Octane III showed that all too well. Even I spent six months trying to obtain any quote at all for an Octane III but in the end gave up; a (new) sales lady at the UK SGI office told me she reckoned at least half the management and a third of the sales staff would have to be fired in order for a direct sales model to be viable, given the way the likes of Dell runs their setup.

Oh, and it didn't help the VW line that MS screwed SGI over XP support, which nerfed the Firewire hw and hindered continued use of the 320/540s as a whole.

I really like the 320/540 systems (I have several, including a 540), they are at their core an expression of the SGI ethos, of doing something different, of helping people to solve problems without having to be slaves to the limitations of the hw. Crapping on them because they're not MIPS/IRIX is not an argument. They at least tried to do something different, something that did offer major performance advantages for key markets like GIS, medical imaging, urban modelling, VR, video, etc. The mistake was trying to sell them to markets where raw CPU/RAM performance was more important, such as animation rendering, that made no sense at all. By contrast, the PC market in general has done nothing fundamentally new since the 440BX.

Ian.

[Raion]

I just don't care for the VWS because it's kind of boring for me. I'm not interested in CGI and never has that been the draw for me. I'm interested in IRIX, and if IRIX ran on HP PA or Itanium I'd be singing praises there instead in all likelihood. The VWS line takes the hardware and puts a gimped Windows on it. No DirectX. No consumer grade Windows, just NT/2000. Pretty gimped for anything I'd do with ancient Windows.

It's also undeniable that despite the great technical achievements of the VWS, it was a commercial failure that doomed SGI's in-house fabrication. The IA-64 move was a lifeboat for the company when they exhausted all of their money for in-house fabrication. And a shitty one, as if they had releases the R18000 I would have been far more optimistic for their future.

Hence why I haven't focused on these and why I won't be adding VWS sections or serving their driver discs. It's not part of the core focus on IRIX. Hence why I didn't choose to buy a domain with "SGI" in it (not to mention, don't want to get confused with that nichiren cult).

[airbozo]

Good write up Ian, thanks!

When I was running the engineering reality center in B6, then B43, we tried to get some of the VW's to add to our line up of systems for testing and development, but got shot down every time. We ended up "stealing" 3 of the 540 demo units and were able to hide them for a couple of months while we ran benchmarks and some performance testing using the code that was developed by the Advanced Graphics team (names withheld on purpose), and the numbers were impressive. We were in the process of validating the performance against the Onyx 3 and Octane 2's when the hammer came down and we all got called into a meeting with our VP and got read the riot act for diverting the VW's to our labs. We all acted dumb and told them we thought they were sent to us specifically to test. Most of our testing and code was handed over to the engineer in charge at the time and it seems to have disappeared. We all thought it was due to someone in the O-3000/Onyx 3 and Octane groups trying to block the information because it made their projects look less appealing.

When I worked at MCE, we ended up with the largest inventory of VW parts and systems in the world. Most of that inventory ended up at 3 companies who were bidding for the spares. One of them was SGI who tried to gobble up all of our parts and keep the inventory for themselves, driving up the prices (We had a love hate relationship with SGI. They hated that we were one of the largest refurbished SGI resellers in the world, if not the largest, and SGI had to come to us all the time for parts). We stole so much business from SGI refurb department they tried to get us put out of business. After Rackable bought the SGI IP, we bid on building their new systems (mostly based on Supermicro) but when certain people found out who we really were (we stopped using Mini Computer Exchange and started using MCE), we were not invited back.

I know we've met Ian, but I am not sure if it was face to face or not. I ran the Engineering labs for the the Advanced Graphics Marketing group under Janet Matsuda and was responsible for the VR labs, the Engineering reality center and setup and tear down for most of the trade shows.

[GL1zdA]

So much inaccuracy and FUD being posted here about the 320/540, probably because many like to imagine they were worse systems than they actually were due to their not being MIPS/IRIX. The reality is very different. I've personally seen a 320 run a hundred times faster than what on paper should have been a far more powerful PC (P4/2.4 GF4 Ti4600), the cause being the task in question (urban modelling and the use of large 16K composite textures, an application called Realax, used to prototype what was then run on a ReaityCentre). A 500MHz 320 ran the test scene in question at 10fps, whereas the PC could barely manage one frame a minute.

As I said, the VW320/540 were good in certain scenarios where the huge texture memory was important or the shared memory did matter. It's basically the same as with the O2, which could be faster in certain tasks than an Octane.

The researchers were moving to normal PCs because SGi's CPU/RAM upgrade prices were crazy and there was no roadmap to a proper IVC replacement (otherwise, like the UK MoD had said, a more powerful IVC2 would have been very popular).

Popular for how long? SGI did nothing when the cadence of new hardware releases changed. John Carmack summed it nicely up:

The big issue is the pace of progress -- SGI aimed for a new graphics family every three years with a speed bump in between. The PC vendors aim for a new generation every year with a speed bump at six months.

So an IVC2 would just lengthen the agony. The problem was highlighted by John R. Mashey already in 1995 (from the classic NT on Indigo discussion):

Hardware companies who sell PCs spend 2% on R&D, and make money by selling large volumes. SGI spends 10-13% on R&D

They just couldn't keep up with nVidia or ATI, who released products for a much broader market.

So the dept. bought cheap Viglen P4 PCs with GF4s. Performance initially sucked because of the composite textures, so the devs had to rework all the models with separate textures, oodles of LODs, etc. in order to match the limitations of the bus-based PC design; one couldn't ask for a better example of how the PC bus arch could limit and define how a problem could be tackled, a technical ball & chain legacy the world has been saddled with for decades.

The VW still have a shared FSB, so you are still limited to 800 MB/s (in practice it's lower on the VW) CPU <-> RAM transfer rate, regardless whether you are using 1 or 4 CPUs. You have to work with very specific problems to saturate the whole 3.2 GB/s memory bus on a VW. Again, Carmack said that in late 1999:

The super-memory-system wasn't all it was touted to be. It worked well for sharing the load between the graphics and the cpu, but the cpu didn't actually see any better bandwidth than on a standard intel chipset. The cpu write bandwidth was actually about 10% LOWER than a consumer machine.

The 320/540 offered an architecture akin to O2 but with far greater pixel fill rate and geometry potential. Even for conventional 3D tasks, a max spec 320 could outperform potent PCs of the day at launch, and should have maintained this lead over time with better CPUs. I ran a post grad lab of a dozen 320s for several years so I dealt with these issues first hand. The 540 was particularly popular with defense companies, a custom model came very close to beating an Onyx2.

Carmack disagrees:

I am typing this on a loaded SGI 320.

When it debuted, it was a very good all around performar, and it had the highest fill rate of any intel based system.

Now, an Nvidia GeForce is just plain superior in almost every aspect. Higher fill rate, even in high res, 32 bit, trilinear modes. Faster, more capable geometry acceleration.

Where it all went wrong was terrible marketing, stupidly secretive tech specs, price gouging by resellers (the ultimate death blow to all things SGI I believe) and a failure to (as Irinikus said) stick to the original promised roadmap of a complete base tech refresh every 18 months. With no IVC2, the custom arch (and its associated advantages) was doomed.

18 months was not enough and, unless they've targeted a much broader market for bigger sales, they wouldn't be able to keep up with the GPU vendors. Jim Clark saw it already in the early 90s that focusing on the workstation market was not sustainable, but others ignored it.

Oh, and it didn't help the VW line that MS screwed SGI over XP support, which nerfed the Firewire hw and hindered continued use of the 320/540s as a whole.

Well, lets stick to the dates. Windows 2000 was released in February 2000 and by June 2000 the VW320/540 vanished from SGIs website. Windows 2000 support was already lacking (no Direct3D, problems with FireWire) so there was hardly a reason to develop XP for the VWs. XP was released in October 2001 and by that time Pentium III was to slow for the workstation market. Even the VW230/330/550 were never sold with XP.

By contrast, the PC market in general has done nothing fundamentally new since the 440BX.

440BX was a popular chipset but in no way pivotal. And there were technologies which actually changed the architecture of PCs a lot like the integrated memory controller or PCIe.