|
||
|
|
|
              |
|
 |
 |
||||||||||||||||
 |
|
|||||||||||||||
By Chris Angelini |
||||||||||||||||
|
All of your white box machines boast CPUs (central processing units). They probably all wield 3D accelerators, too—graphics processors designed to offload crucial rendering tasks so that the CPU may attend to other duties. Even low-end, mainstream systems generally feature limited graphics processing functionality, if only through a value-added integrated core. The combination of CPU and GPU/VPU (graphics or visual processor, depending if you're NVIDIA or ATI) enables explosive performance in today's most compelling multimedia applications. However, as hardware slowly evolves to incorporate more power, so too does the software landscape change to enable newly developed features.
As a result, other intensive operations, such as audio, physics, networking, and storage are offloaded onto the host CPU. Not that today's high-end Pentium 4 and Athlon 64 processors suffer drastic slow-downs at the hands of your MP3 collection, but games, media encoding applications, home studio recording, and HD time-shifting programs are capable of shaving precious performance from a general purpose processor, leaving it sputtering when another application requires CPU resources. So how might you as a VAR leverage ancillary hardware acceleration to improve the customer's computing experience?
Baby Steps: It Starts With a CPU Without any sort of secondary acceleration, host processors handle most tasks. Whether it's emulation of 3D, software-based audio processing, in-game artificial intelligence, or a software RAID implementation, a CPU's flexibility serves it well as a multipurpose engine. In fact, as many system builders shy away from expensive hardware acceleration in a bid to reduce costs, Intel and AMD dedicate increasing resources to improving performance in environments where many applications call on CPU resources simultaneously. Intel's first concerted effort at augmenting efficiency surfaced in the form of Hyper-Threading. Adoption was initially slow in the face of meager benefits. But as developers began optimizing software for the technology, performance improvements became clearer. Of course, Hyper-Threading works by enabling the parallel execution of two threads on a single physical core. And while it might improve efficiency by a measurable margin, potential is always limited by the resources of that one core.
Adding another physical core proved to be the evolutionary next step, according to both AMD and Intel. The latest dual-core offerings—Athlon 64 X2 and Pentium D—are perhaps your best bets for value-minded systems lacking horsepower from complementary subsystems. To be clear, you won't necessarily realize better gaming performance by integrating a dual-core chip, but the simultaneous execution of a media encoder and scheduled virus scan or any other combination of demanding tasks is certainly handled more gracefully with independent cores addressing each job separately. And as software developers embrace multi-threading universally, customers should notice tangible gains in the most popular apps. Graphics Acceleration: Not Just 3D If there's just one more processor you deploy, make it a graphics accelerator. A vast majority of today's games require some form of 3D adapter, which is why even low-cost motherboards lacking an add-in PCI Express or AGP slot always come with a scaled-down integrated 3D solution. But beside the obvious gaming utility proffered by flagship graphics processors, many also facilitate comprehensive video playback acceleration. The PureVideo technology featured in NVIDIA's GeForce 6- and 7-series processors, for example, is able to decode MPEG-2 (DVD) and Windows Media HD content in hardware, relieving the host processor from a chore it would otherwise shoulder. On top of the raw processing power of PureVideo, advanced scaling and de-interlacing routines maintain a superior visual quality, too. Granted, as host processors have matured, the performance hit attributable to video decoding has waned. The upcoming H.264 digital video codec may change that, though. According to Dave Nalasco, technology marketing manager at ATI, H.264 is a mandatory profile in both next-generation DVD formats, HD-DVD and Blu-Ray. “Once those technologies emerge for desktop PCs, decoding will either have to consume lots of CPU time or be offloaded onto a graphics processor with H.264 support. It'll be similar to the days when DVD decoding necessitated an add-in card for reasonable performance.” NVIDIA's tech guru Nick Stam says the company plans to incorporate support for H.264 in its product line by December of this year.
ATI's adoption of H.264 acceleration is less cryptic. The technology will compress true high-definition resolutions of 1280 x 720 and 1920 x 1080 into less than 10 Mbps compared to roughly 20 Mbps for similar quality in MPEG-2 format. Without hardware assistance, H.264 occupies roughly 95% of a 3.6 GHz Pentium 4's time. Aided by ATI's upcoming R520 core, utilization drops to about 33% give or take. Like NVIDIA, ATI is committed to releasing H.264 support by the end of 2005 with whichever cards center on the R520 processor. If you want to offer your customers such playback capability, a modern add-in card from either manufacturer will be necessary. Audio: One Is the Loneliest Number Audio processor designer Creative Labs is more than just a little bit cocky. But maybe it's deserved. After all, Creative is the only company left offering hardware-accelerated sound cards. The HD audio standard you hear so much about relies on host processor horsepower to drive its feature set, as do the PCI add-in cards with competing controller chips. Many of the latest games make generous use of positional audio and environmental effects. When those features are enabled, frame rates often tank as a result. The same goes for Dolby Digital Live, the real-time encoding technology featured on a few motherboards with integrated HD audio codecs.
In theory, today's CPUs are fast enough to minimize the effects of audio processing. And while it's true that most users probably won't measure processor consumption during periods of light use, heavier loads indeed reveal reduced performance at the hands of a software solution. Because audio is temporally sensitive, pops and clicks are also easily interjected when the CPU's full attention isn't available, most often in games. Integrity and performance are the reseller's friends when it comes to selling accelerated sound. Creative Labs' existing EMU10K2.5 audio processor is able to plow through positional audio calculations, DirectSound 3D streams, and EAX environmental effects. It decodes Dolby Digital content and plays back DVD-Audio, too. The performance benefits are clear. And just as many discrete graphics adapters improve video playback quality, so too does the EMU10K2.5 help enable better fidelity than most integrated HD implementations. The latest news from Creative Labs indicates a stronger commitment to boosting audio processing power. We first discussed the company's plans to design and manufacture a better architecture called X-Fi in last month's issue. Technical details included 51 million transistors, a doubling of clock frequency, and 24 times the processing muscle of today's EMU10K2.5. Unrivaled clarity is of course one of Creative's principle foci. However, the X-Fi will also be incredibly useful to gamers interested in immersive audio without a performance penalty. And while Creative still hasn't announced product availability, a recent press release revealed frame rate increases nearing 20% in Battlefield 2 versus a motherboard equipped with HD Audio. Creative Labs's solid developer relations team has helped ensure immediate availability of optimized games once X-Fi-based cards emerge. Expect a compelling upsell from Creative in the next month or two. Storage: Breaking Down Bottlenecks It's a little more difficult to understand the premise behind hardware-assisted storage, although the technology may wind up making an even more significant impact. As it stands today, a vast majority of RAID solutions for desktop PCs are enabled through software. They require special drivers during a clean Windows installation, and RAID performance is often hampered by the processing overhead associated with striping data on a RAID 0 setup, or more seriously, establishing distributed parity in RAID 5. Netcell is one of the pioneering companies interested in accelerating storage functions in hardware, relaxing the general-purpose processing requirements with a specialized piece of silicon. The first generation incarnation of Netcell's vision centers on improving ease of use for VARs and system builders. Packaged and sold by XFX, the NC3000 and NC5000 storage processors handle RAID 0, 1, 3, and JBOD arrays completely in hardware. Netcell president Andy Mills draws the comparison to graphics acceleration where an entire pipeline is contained with the GPU. Similarly, Netcell's SPU (storage processing unit) addresses all facets of RAID storage. Compared to competing host bus adapters, which exist to enable massive connectivity, Netcell is after the single user market. Thus, the Revo cards are available in 3- and 5-port flavors, each with 64MB of cache. The 3-port variant employs two disks striped together in a unified RAID 0 configuration while a third provides parity data.
Streamlining integration is another focus. The XFX Revo64 cards identify themselves as generic IDE controllers, liberating you from frustrating driver installations. There are no more F6 woes while deploying SBS 2003 (been there) or drive space misrepresentations while recovering an older version of Windows XP (done that). Upgrading to a Revo64 is a matter of connecting your existing boot drive to the card, and that's it. A new software migration tool makes it easy to move data, change array parameters, or swap out old drives. But VARs may have a tough time justifying a $200 expense for several convenience features. The performance landscape is perhaps more compelling though still a tough sell. Netcell's current focus is immensely more exciting. By repurposing some SPU resources and establishing partnerships with key software developers, the company hopes to accelerate anti-virus suites, backup applications, and encryption functions. Instead of calling data over the PCI bus through memory to the host processor and back, an SPU would handle everything, interfacing directly with storage. Have you ever tried to rip a DVD only to have the weekly virus scan pop up underneath? It recently happened to me and performance buckled under sustained 100% CPU utilization on a Pentium 4 3.4 GHz Extreme Edition processor. Netcell's Andy Mills claims that an upcoming revision to the SPU will be able to handle such duties. Optimized applications will start surfacing in the first or second quarter of next year. Physics: It's Getting Real “We're at a point where game graphics are beautiful,” says Andy Keane, VP of marketing for Ageia. “And it's universally accepted that general purpose processors couldn't handle the rendering tasks enabled by today's GPUs. Ageia is looking to take physics processing to that next level with dedicated hardware, as well.” The idea of an add-in board exclusively dedicated to physics calculations may seem somewhat obscure, but it's already generating plenty of excitement in the developer community, with roughly eight games expected in 2005 and 34 more in 2006 fully supporting the physics engine software.
Granted, games today already provide for some level of physics computation. What Ageia is proposing multiplies the complexity of interaction between objects and, in turn, completely alters the player's perception of realism as previously immobile game components can be pushed, shot, or even destroyed. Initial emphasis will be directed at accelerating rigid body dynamics and fluid dynamics. Soft body dynamics, hair, clothing, element analysis, and collision detection will follow sometime later, freeing an increasing amount of CPU resources for AI and other tasks. ASUS is Ageia's initial board partner, with hardware expected in Q4 of this year at a price between $249 and $299. That price is likely inflated somewhat by the 128MB GDDR3 onboard cache. When asked about the immediate value of integrating a physics processor at that price point, Ageia's Andy Keane admits VARs are probably better off waiting for the killer software titles. Mid-way through 2006 is when there'll be enough games available to really pitch physics processing to enthusiasts. In Retrospect The value of a powerful central processor is undeniable and well-understood by resellers. Driving most other sub-systems, CPU alacrity really helps determine overall system performance. Nowadays, graphics processors are also acknowledged for playing an integral role. They're required by a majority of games and heavily utilized in video playback. Fortunately, prominent manufactures offer a whole range of configurations from under $100 to over $600, depending on your customer. The hardware audio processor is a somewhat rarer bird, if only because Creative Labs is the only game in town with one. Hopefully, the X-Fi will deliver on its promises and make a drastic impact on gaming and home audio, re-igniting the market. Storage processing is just starting to emerge and the initial impressions based on XFX's Revo64 board are promising. Should Netcell's vision reach fruition, it'll forever change the PC performance landscape. Look for more on that as we move into 2006. The same goes for physics processing, a yet unproven but hotly anticipated capability near to release. Ageia is banking on heightened interactivity to get your customers excited. |
||||||||||||||||
 Back to top |
||||||||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||||||||||||||
