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.
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.
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
Like not crashing, for example
Another of Roscomnadzor's mega-fines is in the offing
Shift will enable 'more flexible handset design', reports WSJ
Because with the good, comes the bad