Thanks for the interesting analysis.

I do have a problem with the hardware compared though.

As much as I agree that the Asus motherboard is the basis of a great x86 workstation, when I read the title of the article, I was hoping to have a real analysis of a high end UNIX workstations from either IBM/HP/SUN/SGI or other professional workstation vendors, against a high end x86 based workstation (Apple/Dell/HP/Alienware/etc).

As a graphic professional who went to Art school in the late 80's early 90s I was trained on SGI Irix workstations and on some SUN workstations.

Recently I was amazed to discover that the SGI Indy machine I used daily with Softimage in 93 has been shrunk and reduced and is almost the same machine as Sony's PSP game console. (processor, graphics etc --not all the graphics bandwidth but still.)

When I first discovered this I wanted to know how much change has happened in the high end UNIX worstation market vs the X86 based market.

Aren't there specialized "Dassault Systémes" workstation with unimaginable numa memory bandwith being used at Boeing and Airbus to design planes in real time?

The thing is that even though x86 has now in speed and prowess more then enough for most advanced consumer and even media processing,
an Autodesk "Flame" or "Inferno" don't run on x86; an IBM POWER workstation, Sun's Niagara or even INTEL's own ITANIUM based workstation have a totally different set of performance values which would be very interesting to compare with a skulltrail/high end x86 based workstation.

What is the present value proposition of a modern $50 000 workstation vs a $10 000 x86 based workstation.

Someone should write about this. (This is what I thought the article was about.)

PS: what ever happened to amazing specialized graphic subsystem/workstation maker Evans & Sutherland?

So computers are faster and cheaper now than in the last decades. WOW!!!

Look at what passes for "news" nowadays.

I was lucky enough to work with SGI equipment in the early 90s and they were certainly very impressive at the time. However, I started to work in the PC industry after that and the demand for 3D graphics and faster processing for gaming put the PC on a path to kill graphics workstations. Today most so-called graphics workstations are nothing more than glorifies gaming PCs with hand-me-down graphics cards like the Nvidia Qadro series. It was inevitable - the market segment for professional graphics was never big enough to justify the R&D required to keep up with PCs.

PC's, whether we want to admit or nor, have largely consolidated under the X86 banner. Whether that;s good or bad depends on your point of view. I remember the days when SGI and Sun were some of the big names, and if you used any proper OS it was some variant of Unix.

Today those formerly hideously expensive boxes trade on Ebay for nothing, and for the right sum of money I could cobble the hardware and software together any "workstation" you want, just how much do you want to pay?

I know quite a few people who are doing their 3D graphics not on some expensive Quadro but on a gaming card; it only seems to make a difference for CAD design these days. 

It's good that workstations really no longer exist, but then there's no cachet now either to saying you got some cool hardware; instead you have have to make something with it rather than just admire it.

Skulltrail is the Brillo Box of the PC world!

"Pure 32-bit addressing with large memory and graphics processing, plus 1280x1024 (there was no SXGA term then) colour CRTs"

There were no 5:4 aspect ratio monitors but 4:3 only, so 21" screens used to have 1152x864 mostly since that kept the aspect ratio. That was actually also fixed resolution in some screens that were not multisync capable (Macintosh etc.). So even the 1280x1024 was possible in some systems, it was AFAIK almost never used.

The distinction disappeared years ago. These days, "workstation" is nothing more than a marketing term.

Finally some quality is arising from nation/decades of hard works is about to form Blossom Clusters for PC.Asi ultietoom, maker of graphs over years have stated repeatedly 10 minimum, 25 very good mb/s final data. 10 ghz prosessor bandwidth fills every memory space once persec and memory 10 gh/sec also, so plenty of stuff can interact as it morphs or is stored or outputed by gate setting / instructions, etc, still 10 billion quieries per is enough to hold down 100 mbs or even 1 gb sec of actual data transfer, so why such low numbers as 10 mb/sec out final stream. its simple they need controllers & slot for 64X, thats sweet spot. Oh sure, not tonight, so much tech improvements to utilize that speed is coming on roadmap, MY MIND IS BEING PRICKLED BY CLAWS.
drashek

I remember drooling at those workstations during the early 90's. We considered ourselves lucky to use 4MB 25MHz 386 / 486's but seeing those Sun's and SGI's were mind blowing,

Dual 133MHz 64-bit RISC MIPS or 200MHz Alpha AXP's, 64MB, 24-bit graphics, GPU's proper (expensive) UNIX OS's with SMP and memory protection instead of wanky Windows 3.1, monitors the size of most peoples colour TV's at the time...

These days would need two 30" monitors, dual socket 8-core Nehalem with 64GB RAM and quad nVIDIA G100's to be as sexy!


Standardization drops costs for hardware manufacturers and consumers alike, so the lines becoming blurred for I/O ports and communications protocols makes sense. However, true workstations usually still have some exotic hardware plugged into a common interface, like a PCI-Express slot. Take the "cell accelerator board for example. (http://www.mc.com/products/productdetail.aspx?id=2590&ProductTypeFolder=56)
It has a Cell BE processor at 2.8 GHz
More than 180 GFLOPS PCI Express x16 interface with raw data rate of 4 GB/s in each direction, Gigabit Ethernet interface
1-GB XDR DRAM, 2 channels each, 512 MB
4 GB DDR2, 2 channels each 

How many other 'workstations' have an add in board with a gig of dual-channel XDR and 4 gig of dual channel DDR2?

Want something to boggle your minds. There is an end where you get to a point where you can only put some many processors together before you get a problem with heat. Heat is naturally everyones enemy and transfers into power consumption. One of these days probably before 2015 there is gonna be a point where you cant make the transistors any smaller and you cant put anymore transistors on top of each other. At this point you will probably render anything in the seconds range 3-5 secs. So there is an end of the road. At this point there will be no point in upgrading because why upgrade from 5 seconds of rendering to 3 seconds. Doesnt make sense. But wait a transistor now has a binary function of 0 or 1 and there are millions of them on a die. Now imagine if all those transistors had a binary of 1 or 16,423 that means 16,423 x 16,423 instructions per transistor. Holy shit batman we can now render everything and process everything in 1 sec.... wow... Lmao... its true Motorola has a processor that can do this right now that is the future. in about 7 years all this madness will be over.... upgrading and more upgrading.