The 320 has the same basic architecture as SGI's earlier O2 system (called UMA in O2, referred to as IVC in the 320/540), indeed they should have used the Cobalt core in the O2+. People misunderstand the 320 in so many ways. It isn't a clone, it isn't a hybrid and it would only appear odd to someone not familar with SGI's existing tech, coupled with the idea that a PC is a PC is a PC, same basic design year after year with little to no real innovation (that didn't really get going until the crossbar in the 400BX and the GF256, both of which were based on SGI's ideas). The 320 perfectly reflects what SGI has always been about: processing *data*, designing system architectures to enhance how one can do this. The point with the 320 is that people who are used to standard x86 systems don't know how to even think about how the system works, they're too focused on CPU performance. The heart of the 320 is the Cobalt chip, just as the heart of the O2 is the CRM, likewse the thumping monster in an Octane is the XBow; all of these are data bandwidth engines, with the ones for O2/320/540 also performing main memory access functions and some of the gfx functions (everything not related to geometry, more or less). It's wrong talking about the Firewire ports as if SGI screwed over customers somehow; the reality is that MS scewed SGI with FW support, because SGI had been expecting WinXP to support the 320/540 but instead MS ditched HAL support entirely (MS' line to SGI re driver issues had been, don't worry, we'll fix it all with XP). WIth the system limited to Win2K, that basically nerfed FW as a function of the system. 7:20 - The fact that the out-of-box Ethernet driver in Win2K doesn't work properly isn't really an issue since the first thing one is supposed to do after installing Win2K is then install the SGI-specific updates and patches (these shipped with the system). These changes are especially critical for the graphics and video features. LVDS was a standard that was used by several vendors, not just SGI, but as always with such things a company takes a gamble choosing a particular direction and in this instance SGI didn't pick the "winning" connector, irony being the 1600SW flat panel was very successful indeed, won numerous awards and is still in demand today (I have a buyer waiting list). It would be easy to criticise SGI for the RAM choice (indeed I did at the time), but there were reasons for it (btw, I have RAM available. ;) Where SGI went wrong (they often did this) was simply making the RAM upgrade costs too expensive. Well, this is one of the things they did wrong, the biggest I'll mention later. 11:10 - Again this is typical of what I mean, you're focusing far too much on the CPU, which isn't the real power behind what the 320 can do (definitely matters for 3D since the CPU does triangle setup and geometry, but anyway...) Let me put it this way, I was at that time far more used to dealing with MIPS/IRIX SGIs and the PCs I'd dealt with before were standard designs; no PC until the 320 had ever impressed me, always slow and clunky. Then the 320 came out and I got a new sysadmin job running an SGI-based VR centre which also had a dozen 320s, used for VR, urban modelling and imaging. It was the first time I saw an x86 system load and display a large 50MB 2D image file as fast as an Octane, and it can do it because of how Cobalt works. Like O2, one can then do real-time 2D imaging operations on such a file (pan, roam, zoom, rotate, sharpen, blur, convolution kernels, etc.) without being limited by graphics memory like a normal PC, because main memory *is* graphics memory. Doing complex operations is frequently trivial because no data needs to be moved, one just passes a pointer, whereas on a normal PC there can a lot of data flying about between the gfx card and a separate video card, over the PCI/AGP buses, etc. For urban modellng, a tax with limited geometry but enormous amounts of photographic texture data, this is a dream come true. What SGI fundamentally did wrong though was not stick to their original promise: to release a completely new system with the same architecture and updated Cobalt every 18 months, maintaining a *balanced* architecture, something frequently not possible with normal PCs. By that time SGI was infested with management and marketing people who didn't care about the company's traditional ethos, they were selling off the family silver and breaking promises, the worst being the O2+ which was a total farce. So instead of Cobalt2, the diluted SGI tried to keep their hands in the x86 pie by releasing later systems (230, 330 and 530) based on a conventional PC design with a VIA chipset - crazy idea, they could never compete with Dell or HP in that space, and they had utterly the wrong kind of marketing and sales structure to do it anyway. By contrast, the *customer* demand (especially from the defense sector) was for a proper followon, the UK MoD wanted a new 540 with 8 CPUs, faster Cobalt and a much higher max RAM. Sometime around 2000, Intel, MS, Virtual Presence and SGI put together a custom quad-XEON 540 (using 4MB L2 CPUs) which came very close to matching an Onyx2 deskside, the first time any "PC" had come close to such a feat. So the later gen VWs were just normal PCs, nothing special about them at all, and of course far too expensive. I stuck my nose in the fray at one point, made some public comments, caused a row which even had the CEO asking people to discuss matters, and it did result in a 13% price drop, but that was still nowhere near enough to sustain the product line. Even if they could match the pricing, SGI's reseller sales model was all wrong for the x86 market.
But don't sell the 320 short, nor its unique design. When my dept. upgraded, they bought a bunch of P4/2.4 PCs fitted with GF4 Ti4600 64MB cards. An initial transfer of the apps and data resulted in performance that was a hundred times slower than the 320, this because the modellers (not limited by the nature of the 320 platform) had been using large composite texture images to store all the image data. This had no performance impact on the 320 at all since the data isn't actually moved or copied anyway, but on the normal PC it killed performance stone dead, via dreadful thrashing between the VRAM on the gfx card and main memory, constantly trying to load and reload huge images in a manner the 320 never had to do. So the modellers had to spend ages redoing all their models, using LODs, etc. to *make their task match the limitations of the platform*. And that, in its basic essence, was what SGI in its heyday was all about, helping artists, designers, engineers, animators and other professionals to perform their tasks without having to be limited by the nature of the system they were using. SGI designed tech to help people solve their problems; the PC was and still is entirely about imposing and maintaining a standard system arch on the entire technical world regardless of whether it's a sensible match for individual tasks or not. Customers have to work round or just live with the limitations of the platform, amid an industry mindset that has long accepted this as normal. SGI lost their ethos in the later years though, and they missed numerous opportunities for broader markets, especially consumer gfx (though again their sales model would have made that difficult or impossible), though of course ex-SGI people and their transferred IP kicked off consumer gfx anyway, a la the GF256, N64, etc. 12:10 - Not true since, as I mentioned above, CPU upgrades speed up the gfx automatically. A 320 with two PIII/1GHz can outperform a GF2 GTS, which is very impressive. Again though SGI mucked this up by not even being open and public about what the max possible upgrades were, nor the required jumper positions (though a later firmware update removed the need to bother with jumpers). That was fixed by an employee just making the info public at some point, something that happened several times over the years, employees making info public when it was clearly beneficial to customers. What SGI basically did here was create something genuinely new which, to be successful, required management to stick with it as promised for the long term, marketing to not lie to customers about the strengths and weaknesses of the platform (great for data visualisation, bad for animation rendering or straight compute!), and for technical data to be easily available. The company failed in all three areas. 12:42 - And all of those alternatives would suffer from the same limitations of a conventional bus-based machine of that era. Your reference to HP still making workstations is not an argument, the 320 didn't fail because the arch was a bad idea, it failed because they didn't know how to market it properly, management didn't stick to their roadmap promises and they didn't make it a technically open platform in terms of technical data. 15:35 - You have an old version of the PROM, you should upgrade. 15:55 - Blame MS for that. 17:00 - Do you know about the command monitor interface you can access from the PROM for debugging, etc.? It looks like you don't know it's there. Check the FAQ on my site: www.sgidepot.co.uk/vw/ Btw, the sweet thing about the 540 is that it has a dedicated PCI64 bus *just* for the system disk, so there is never any bus contention for traffic to/from the system disk. Plus of course, the default system disk in a 540 is normally SCSI anyway. 19:50 - That highlights one of the problems I moaned about with the marketing of the machine at launched. I never believed 256MB total RAM was enough as a minimum for any customer, but resellers were happy to sell them with 256MB if it meant securing a sale. I never operated that way, I told clients if they couldn't afford a 320 with 512MB then they would be better off looking at a standard PC, because buying a 320 with underspecced RAM was not a good idea. 22:00 - It's funny you talking about Quake3, reminds of GN focusing on Quake for their O2 video. When it comes to games, who cares? It isn't a games machine. When the UN was providing support operations in Bosnia during the civial war, a dozen pre-release 320s were sent over to handle the sat imaging analysis tasks as it was the only system that was up to it (the alternative would have been far more costly Octane2 VPro systems which lacked the arch advantages for defense imaging). And as I say, the system was not meant to be hw-upgradeable (except for CPUs); think again about how it might have been if SGI had done what they originally promised to do, release a genuinely new system with Cobalt2 18 months later, something staggeringly faster than the competition, using standard CPUs (and perhaps standard RAM by then, since initially ordinary PC RAM wasn't up to the task so I was told), though with MS messing about with XP support maybe that was doomed anyway, who knows. As is so often the case with SGI's tech, it's difficult today to think about why they built what they did, in the way they did, because the tech mindset of today is very different. Modern computing systems have been based on the same derived x86 arch for more than 30 years, whereas SGI's entire ethos was about enabling customers to solve problems now that simply couldn't be done either effectively or at all with current tech. They did speculative designs, tried different things, but so often what they did was ruined by poor management, marketing/PR mistakes and more than anything else an adherence to a fundamentally incompatible sales model. The 320 though, like the O2, because of how it was designed, enabled those in a field such as medical imaging to save lives, because it was far more effective for volumetric imaging than a conventional PC. Even with the O2, a much slower machine (Cobalt is about 5x faster than CRM), one doctor told me the system helped cut CAT scan analysis times from an hour down to 10 minutes, saved time being vital. Today tech is a brainwashed race of sameness. AMD vs. NVIDIA, AMD vs. Intel, CPU, GPU, on and on, but the basic platform hasn't changed in decades, the dominant Windows OS plods on with all its inherited limitations, nobody tries anything genuinely new because we blunder on with mountains of accepted compatbility. Milan is the first thing I've heard of that sounds in any way novel, but even that will still be bolted to same legacy structure in the end, and the same OS. I still like the video btw. :)
In high school and college the place I worked for bought these to run Pro/ENGINEER. Solid machines, they were great work horses. The price v. performance wasn’t the best but they proved to be fairly upgradeable and were always rock solid and had fantastic support. We ended up adding a second cpu, then a couple of years later put in two of the fastest CPU’s possible and eventually the maximum RAM. They were in service for a good 10 years before they were scrapped. It turns out the price v productivity over the machine’s lifetime was fantastic. I ran ProE v20, then 2000i, 2000i2, and finally 2001 on a 320 until they moved me to an SGI 330. The performance was pretty good and it got better as we upgraded the cpus. We didn’t require any texture support so those features were wasted on us. By the time the nVidia Quadro cards based on the GeForce2 architecture came out and Pentium 3 CPUs above 600 MHz were available it started to make more sense to purchase a PC workstation. However it took a couple years for them to catch up. We handed our 320s down as newer machines were brought in. Eventually the engineers were all using SGI 230/330 machines which were just boring PCs. The 320s were eventually moved out to the shop floor for machinists. They took a real beating and really demonstrated how well built physically they were. The door you think was a gimmick kept a lot of crap out of the optical drives and Zip drives we had on our machines. They were almost always kept shut as the machines were never really powered down and everything was typically accessed over the network. Bottom line is that they were pricey, “PC guys” didnt understand their value, and they wouldn’t make sense for a home computer, but when it came down to getting work done in certain applications they were absolute workhorses. Probably ended up being one of the best workstation purchases the company ever made. After the SGI 230/330s the company went to Dells.
I had this workstation in 1999-2000, it worked well, xpt FireWire port not working just because Windows NT had no support for it, so for about 6 months SGI promised a driver for it, but in the end they just gave away free firewire card to every owner. Other than that-no complains...
I bought one when it came, it was great as a developer machine, running Visual Interdev and later Visual Studion and .Net 1.0. But so freaking loud, I placed mine in the bedroom and ran long cables to my work desk I actually have mine in its original box with mouse and keyboard, but I sold the CRT screen long ago.
There is a Granite SGI monitor on eBay at the moment for only £250! These things are rare as anything and real expensive... if only I had the room for it - it's a 30kg monster! :(
Doesn't NTLDR on all non-ARC platforms contain an ARC emulator? I know NT on PPC has this, too, and MIPS and Alpha run ARC, so they don't need one obviously.
Do you happen to have copies of the original CDs that came with the system? I just purchased a 320 with a broken NT installation (looking for DISPLAY_DRIVER.DLL) and it memory dumps. I need to reload and would love to put W2k on it. Can't figure out how to make this boot off of the CD using a generic W2K (WinNT) disk that I have. Much thanks.
sgi's vs pc's. Pc's would be the fastest runners but they would have their shoelace tied together and stumble a few meters up the track. SGIs where decent but slower runners, but they never ran out of breath and you could fill their backpacks with bricks. That was the point of these machines, they got sh*t done, even when you weren't looking.
