Today's PC is yesterday's graphics workstation
THE PERSONAL COMPUTER, led - unfortunately from a technology point of view - since 1982 by X86-based IBM design, has evolved gradually in power and performance. At the same time, another class of desktop or deskside machine evolved in a very different and more demanding user environment - graphics workstations.
Since the mid eighties, graphics workstations clearly set themselves apart from PCs in the use of their 32-bit processors (Motorola 68000 series mostly, followed by the first RISC like SPARC) and large memories. These were often an order of magnitude above the PCs of the time. A select few also had two processors, where the architecture enabled that.
Those with long memories will also remember the very first real GPUs, like NEC's 7220 and Hitachi's HD63484 2-D graphics processors, often coupled with those Motorola 68Ks over buses like VME or VMX, precursors to today's PCI and AGP, for instance - that was over 20 years ago, mind you.
Pure 32-bit addressing with large memory and graphics processing, plus 1280x1024 (there was no SXGA term then) colour CRTs, enabled real-time CAD, CAE and EDA apps handling at the time, even though 3-D graphics cards were still as rare as hen's teeth. And even then, usually expensive two-to-three board custom circuit combos limited to the basic GL polygon, texture and lighting processing offload.
But, those workstations sowed the seeds of today's real-time interactive 3-D graphics, as well as 32-bit OS and application foundations that we see on our PCs today.
In fact, they put up the same base 15 years ago for the, today still fresh, 64-bit desktop computing - does anyone remember 64-bit DEC Alpha, SGI MIPS and and similar HP, Sun and IBM workstation offerings? All their 64-bit UNIX desktops with real time OpenGL graphics, fancy GUI's, large memory support, multimedia authoring or rendering apps and so on? A lot of that is now alive on 64-bit Linux and Windows desktops, diminished a bit by the X86 architecture " imperfections", to put it politely.
So, all those years we had a distict division: PCs with antiquated X86 architecture, so-so configurations, messed up Windows, cheap apps and mediocre graphics on one side; and workstations with high-end RISC processors, humongous server-like memory and I/O, capable and reliable UNIXen, utterly expensive software - sometimes, five-digit amounts per app - and monster graphics. As for the systems alone, if a high-end PC in, say, year 2000 was around US$ 5,000 in a reasonable maxed-out configuration, a top workstation would easily reach 10x that, close to US$ 50,000, when filled up.
What about now?
How different are today's graphics workstations from their high-end PC
contemporaries?
Here's the right example. We recently looked at the Intel Skulltrail and Azus Z7S WS mainboards. One is declared as a very high-end PC, the other as a high-end workstation. Now, let's compare the specs:
Two CPUs - same types and number suported; memory - same ECC FB-DIMM 800 type, slight capacity difference (16GB vs 24GB), with both of these allowing limited overclocking. Chipset: same Intel Seaburg. Graphics (or otherwise) PCIe x16 slots. Four v1 on Intel vs two v2 on Asus - so, in theory, this "high-end PC " has double the graphics card expandability of the 3-D workstation board, even though 99.9 per cent of users may not have more than two graphics cards anyway.
Move to the I/O - all interfaces pretty much same, except Asus forsakes that flimsy on-board sound for an extra PCIe slot to let you add proper sound cards. Power supplies: the same kilowatt-class EPS PSUs. Format: the "PC" Skulltrail is actually using workstation sized EATX format, while the "workstation" Z7S WS fits on the PC sized ATX! And the expected price... the same US$ 600 - 700 bracket.
Operating systems: identical support. Overall performance: pretty much the same, with or without overclocking enabled. If all other components (high end casing, displays, adapters etc...) are the same, the material costs will be the same.
The price
Usually, the workstations would be more expensive due to higher margins
and customised quick-response two- or four-hour support contracts; expensive
engineering or multimedia application certifications, and longer warranties
asked for by the corporate or government users relying on such machines.
What about the high-end PCs of today? Well, dedicated "real human" support lines with quick turnaround; custom high quality components for high reliability or overclocking needs, long or lifetime warranties for many components even if they fail due to (reasonable) overclocking and, due to the "exclusive enthusiast niche" perception, higher margins too.
So, from the physical point of view, the workstation and high end PC are one and same thing now - only the software stack and those little but costly support details may differ. So, the distinction has become irrelevant.
As a good example, you can compare HP's VoodooPC and xw-series workstation configurations - the specs are getting closer than ever. Apple went even closer than that: the 3.2 GHz dual Xeon, FSB1600 Mac Pro is a true combo of high-end PC and a workstation - the final step.
And yes, machines like that Mac Pro, the above mentioned HP's, or your own stuff based on Skulltrail or Z7S WS can also act as nodes in large supercomputng visualisation clusters - imagine 1,000 CPU cores and 500 GPUs on a 128 node real time OpenGL vis cluster rendering together a single 100 megapixel 3-D cave, at guaranteed minimal 60 fps? It is a reality today.
What's next?
Well, the Nehalem generation will complete the physical blending of the two
categories - the single-socket Bloomfield and dual-socket Gainestown will
crosspolinate the high-end PC and workstation boundaries, with each having
offerings on either side of the - disappearing - fence.
The Tylersburg chipset flavours - same. DDR3 memory - same, maybe just with or without ECC. I/O - similar if not same. Power and cooling - same. Graphics, pretty much same, except for those Quadro and FireGL overpriced aberrations - partly due to the expensive GPU software certs for those expensive apps.
So, by the end of this year, it may be truly one and same machine class, with varying usage models. Same will happen with the future AMD-based systems, even though the green camp lost ground in this market - just like most others - last year. µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
Skulltrail is the Brillo Box of the PC world!
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. 
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?
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!

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
The distinction disappeared years ago. These days, "workstation" is nothing more than a marketing term.
"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.
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.
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.
So computers are faster and cheaper now than in the last decades. WOW!!!

Look at what passes for "news" nowadays.
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?