Finding the Gamer's Edge

Shopping for Workstations

Working closely with Alias on Maya version 5.0.1, 3Dlabs improvedthe rendering speed offered by its Wildcat VP870 graphics acceleratorby taking advantage of its programmable Visual Processing Unittechnology.

We live in a world where computers built to play games can cost moreand run faster than the computers used to create those games. The mostavid gamers are willing to spend large sums to buy the newest hardwareand upgrade it frequently. Professional computer users rarely have thebudget to operate on the bleeding edge of computer technology, and someof gamers' techniques, such as overclocking, are inappropriate for acommercial environment.

However, understanding the options available in professionalhardware for 3D content creation and video production will help you getthe most from your money.

The first step in selecting a computer system is deciding whatsoftware you need to run. This decision determines your range ofchoices among the processors, graphics cards, and other specialtyperipherals on the market. If you're running Windows software, you maybe tempted to run out and buy a mainstream PC desktop. Certainly theprice and performance of these systems are attractive, but don't be sohasty. They might not give you everything you need.

All AMD and Intel processors can generally run the same programs,but this is about to change. The arrival of 64-bit processing isstarting to change the nature of high-end systems, such assupercomputer clusters used as rendering farms, database servers, andcomputer-aided design workstations. The programs are changing, too.There aren't any 64-bit-only programs in the PC market, but there soonwill be.

Nvidia's current top-of-the-line Quadro FX 3000 graphics cardfeatures a 128-bit floating-point graphics pipeline, and vertex andpixel programmability. Zoic Studios used a variety of Quadro FXgraphics cards for its work on the SciFi Channel's BattlestarGalactica miniseries.

AMD and Intel have separate roadmaps for 64-bit computing. Intel'sItanium family represents a sharp break from the company's x86tradition. Itanium is based on a new architecture called EPIC —Explicitly Parallel Instruction-set Computing. EPIC was created toallow the processor to run more instructions at once than conventionalCPUs. Itanium processors are the fastest in the world, but not by awide margin, and not on all code. Itanium's biggest advantage is in bigmultiprocessor servers, but that's not relevant to typical PCbuyers.

In fact, Itanium has weaknesses that may make it irrelevant for you.Though it can run x86 applications, its x86 performance is poor. Ifyour applications are tied to specialty hardware, you might not be ableto use it without new drivers. Even if you can use Itanium, it'sexpensive, and systems aren't widely available.

Intel has an aggressive plan for speeding up Itanium whilesimultaneously reducing the cost of Itanium systems. The companybelieves that Itanium will soon surpass the price/performance ratio ofits x86-based Xeon processor family — but issues of softwarecompatibility will, of course, remain for years.

AMD's new Athlon 64 family may be a better fit for most professionalusers. To get the benefits of 64-bit computing on Athlon 64, you'llneed to run a 64-bit operating system, but your existing 32-bitsoftware will still run at full speed. With 64-bit versions of Windowsand Linux now under development, expect to see a growing selection ofAthlon 64 systems and applications in 2004.

Athlon 64 is the only way to get a new 64-bit computer for less than$1,000, though you'll pay more for reasonable system configs. Many proswill want to step up to AMD's Athlon 64 FX and Opteron processors. Bothare based on the Athlon 64 chip design, but have twice the memoryperformance. Opteron adds multiprocessor support, from two to eightprocessors, depending on the model.

Fujitsu Siemens Computers was one of the first major OEMs to offeran Opteron-based workstation. The company's Celsius V highlights someof the key advantages of the Opteron family, with up to 16GB of memoryand chip-set support for four Serial ATA interfaces, Gigabit Ethernet,AGP 8X Pro graphics, and PCI-X expansion. The Celsius V was introducedwith the 2.2GHz Opteron Model 248 processor, which is faster on manyapplications than Intel's 3.2GHz Xeon, which is sold in FujitsuSiemens' Celsius R series. The Celsius R has similar features, butsupports 4GB of memory and can't run 64-bit operating systems orapplications.

With 64-bit versions of Windows and Linux soon to release, themultiprocessor-capable AMD Athlon 64 Opteron CPU adds chops to gearsuch as Fujitsu Siemens Computers' Celsius V workstation. Featuresinclude up to 16GB of memory, support for four Serial ATA I/Os, GigabitEthernet, AGP 8X Pro graphics, and PCI-X expansion.

Because the Celsius V is sold as a workstation, it's available withNvidia's Quadro graphics cards, and Fujitsu Siemens is pursuingcertifications for popular professional applications, as well asLinux.

Apple is also making a 64-bit play, but from a different angle.Because the PowerPC processor architecture was conceived from thebeginning to support 64-bit operation, there's no conflict between32-bit and 64-bit modes. Apple's Power Mac G5 and Xserve G5 machinessupport 64-bit calculations, even though they're still running 32-bitoperating systems. Software developers have already begun to releasenew versions of key applications that take advantage of the G5's 64-bitcapabilities.

Mac OS X supports OpenGL 3D graphics, and Mac graphics cards havemost of the professional-grade features of their PC counterparts. UntilApple releases a 64-bit operating system, however, Mac applicationscan't address more than 4GB of memory at a time. G5 systems can havemore than 4GB of memory, but this helps only if you're running multipletasks at once.

What's the least expensive 64-bit system with a 64-bit operatingsystem and a wide selection of 64-bit software? Surprisingly, theanswer is Sun. With prices starting at just $1,395, Sun workstationsare worth considering if the applications you need to run are availablefor that platform. Performance at each price point is very weak whencompared to PCs or Macs, but sometimes performance doesn't matter asmuch as capabilities and stability of the operating system — andSun's Solaris operating system is top-notch.

Expansion buses used to be pretty easy to understand. More than 20years ago, the original IBM PC came with ISA card slots. In the 1990s,PCI slots were added, and soon ISA was phased out. PCs aren't so simpleanymore.

PCI was quickly overwhelmed by the needs of new 3D graphics cards,leading Intel to develop AGP, which is an extension of PCI. Othercompanies extended PCI in a different direction to create PCI-X, whichis now widely supported in servers and workstations such as the CelsiusV. PCI-X is currently the only expansion interface that can deliverfull performance for high-performance networking and storage cards, butit isn't available on most desktop PCs.

This summer you'll have another choice: PCI Express, or PCI-E forshort. Though software-compatible with PCI and PCI-X, PCI Express isvery different on a physical level. Instead of the 32-bit or 64-bitparallel interfaces of PCI and PCI-X, PCI-E uses one or more ultra-fastserial channels. Faster cards, like 3D graphics adapters, use morechannels.

Boxx Technologies became one of the first DCC manufacturers toannounce support for AMD's 64-bit Opteron multiprocessor CPU. Boxx's3DBOXX M4.1 encases dual Opterons in an all-aluminum chassis, alongwith a full-length 8X AGP slot, Promise MultiRAID, 2 UltraATA ports, aswell as 2 SATA ports. Current 3DBOXX users include Turner Studios andRainmaker.

Graphics will move to 16-channel PCI-E cards fairly quickly; mostother functions will need just one, two, or four channels. Systemmakers will match the number of PCI-E slots, and their channel counts,to each system's intended market. It'll take a few years for PCIExpress to achieve its full potential, but it will give PC buyersaccess to the same technology being developed for servers andworkstations.

What sort of cards will go into these new slots? Basically the samekinds we can get today. High-performance networking has settled onGigabit Ethernet; it'll be a while before we have to worry about thenext generation, 10G (gigabit) Ethernet. Mass storage is split todaybetween parallel ATA (the lowest-cost solution), Serial ATA (SATA— faster and slightly more expensive), and Ultra320 SCSI (fasteryet, but much more expensive).

The next big thing in storage interfaces is Serial Attached SCSI(SAS). SAS uses the same low-cost physical links as SATA, but the moresophisticated SCSI command structure. SAS promises to reduce the pricepremium for high-end storage systems. You can't buy it yet, but keep aneye on this technology — it's going to be good.

SGI's quad-processor-capable Tezro has sparked new interest in thecompany's Irix-based workstation line with a lower price and improvednetworking capabilities. In February, the Tezro received a processorupgrade to 800MHz via the latest MIPS R16000A processor.

Faster networking and storage are critical for applications such asHD video editing, which involve moving vast amounts of data. You canalready buy disk arrays fast enough to handle multiple channels ofuncompressed high-definition video. But if you try to plug these arraysinto your garden-variety PC desktop, they'll be throttled back to thespeed of your PCI bus. That's one more reason to look for PCI-X.

Nothing in the computer industry has evolved faster than 3Dgraphics. The first 3D cards appeared about 10 years ago. Ever since,every six months new cards have doubled the performance of the previousgeneration. Today's best 3D accelerators are literally a million timesfaster than those first products. This performance improvement isspread across every aspect of 3D-rendering.

Programmability is where the action is. Four years ago, there weretwo options for realtime computer animated objects: the plastic lookand the polished-metal look. Programmable pixel shading changed allthat.

Everyone knows the leaders in the 3D graphics market — ATI andNvidia. ATI's latest FireGL X2-256t card and Nvidia's Quadro FX 3000Gare fighting it out at the top of the market, but they're bothexcellent products. Which you choose will likely depend on the softwareyou want to run. Some applications are written specifically for one orthe other, and each card has certain unique features. The FX 3000G, forexample, can blend the outputs from two projectors on the same screento create a low-cost “powerwall” effect.

HP garnered kudos for the value included in its entry level xw4100workstation, which features a hyper-threaded Pentium 4 at up to 3.2GHz,800MHz frontside bus, 5 PCI slots, convertible mini tower, and room forthree drives. Sporting a fancy paint job, HP offered this particularxw4100 as a prize at the Hands-On Filmmaking Lab held during theSundance Film Festival.

Gamers don't buy these cards, but they're based on the same chipsATI and Nvidia develop for the gaming market. Both companies buildtheir chips with a few extra transistors here and there to support theadditional precision and features that professional users require. InCAD software, you need to know that a line will look the same onscreenwhether you draw it left-to-right or right-to-left. In a fast-moving 3Dgame, on the other hand, such consistency isn't as important as speed.That's why there are different cards and drivers for professionals andconsumers.

There are still reasons to consider the high-end offerings from3Dlabs, a rare survivor from the earliest days of PC 3D. The company'sWildcat VP990 Pro has 512MB of on-board memory, which ATI and Nvidiadon't yet match. If your 3D models require unusually detailed textures,this might be important.

One of the most interesting things about the 3D business is ourasymptotic approach to the goal of photorealism. It's all about tricksthat create better approximations of reality.

We've come a long way, but there's a long way yet to go. It'll takeanother 20 years to produce live computer animation indistinguishablefrom reality, but we'll get there — one step at a time.

The programmable 3D chips we'll see in 2004 will be able to runlonger, more complex programs, even including conditional processing— tests and jumps. You won't want to run your whole applicationon your graphics chip, but you'll be able to.