|
||
|
|
|
              |
|
 |
 |
|||||||||||
|
 |
|
|||||||||
By William Van Winkle |
|||||||||||
Workstation Video At a certain level, the real difference between video on desktops and on workstations is whether or not the user is getting for his efforts. Because if money is on the line, we all know that time is money. The sooner a videographer can finish a project, the sooner he can move to the next job, be able to perform more jobs, and make more green. Thus, while it's nice for a consumer to have fast video editing, for professionals it's imperative. The question is how much money they can afford to pour into the pursuit of saving time. We've spoken with end-user accounts in the graphics and video world, and some stake their livelihood on professional, vendor-certified workstation applications backed by matching certified hardware while others are perfectly content to mix and match with consumer graphics cards and pocket the savings. As a solution provider, you can't be dogmatic about such things. Sometimes, a client's budget gets the last word. But it's clear that there's more value to be found in a workstation-based solution when certain criteria apply. "Resellers are in the business of making good suggestions to customers they can support in their environment while at the same time looking for margin opportunities for themselves," says Jeff Medeiros, business development manager at PNY Technologies. "That value-add opportunity is based on professional products and professional services, not the integration of consumer products, because there are no consumer services to provide behind that. And also because they can't support those consumer products in a professional market. There are lists of graphic boards and corresponding drivers that different software packages support. When you use that combination of elements and you call Autodesk or Maya in the middle of the night when something's crashing and you've got to deliver it tomorrow morning, they're right there for you. Same in the CAD marketplace. When you mismatch consumer products with professional applications, the people that have to support that are at a disadvantage because they can't narrow down why the system is unstable. So they don't support that combination because they can't." Without question, the most prevalent brand of workstation graphics card in the digital content creation (DCC) and non-linear editing (NLE) markets is NVIDIA's Quadro FX. A very common question asks what the difference is between a Quadro FX and its GeForce counterpart, because in most ways they're identical. However, there are some key differences within the GPU. For example, a feature called hardware-accelerated clip planes allow Quadro cards to cut away sections of solid 3D objects in the CAD field. Quadro's support for antialiased lines as standardized through OpenGL is key for preventing the "jaggies" that plague wireframe images. (This isn't needed as much in the mainstream graphics world where rendered polygons, not lines, are the norm. ) In fact, OpenGL is at the heart of why digital video systems benefit from workstation over consumer cards. "Both card types support OpenGL," says Julien Zanchi, NVIDIA's product manager for Quadro FX, "so all the OpenGL-based applications run on GeForce. But in the GPU on the Quadro version, we have some hardware-based acceleration of the OpenGL functions as well as some additional OpenGL functions, so OpenGL will be faster on Quadro. If you just want to experiment with digital video and non-linear editing and your timeline is not realtime, then it's OK to go GeForce. If you need to move quickly, especially when revenue is at stake, then productivity comes into the equation and we're talking about Quadro—especially if you need to show work to a client in realtime." "Specifically discussing Adobe Premiere Pro," adds PNY's Medeiros, "there are some filters that are GPU-accelerated [through OpenGL], so when you browse through your list of transitions in Premiere, there's a category called GPU FX. You can select that and drag it onto your timeline, and then those effects would be GPU-accelerated. The one I recall, a pretty complex effect that takes up a lot of CPU resources, is a ripple effect. You can apply that to a few frames, and then you see like a drop of water on the frame itself. That's one example. There's also a lot of color correction effects. You can change tones on the frame or do color matching calibration between two sequences. GPUs are very efficient for that kind of operation. And all these functions are user-accessible, so they can decide if they want them GPU-accelerated or not." Bottom line: You can't get these OpenGL acceleration benefits on a consumer card. You have to leverage something like a Quadro FX or an ATI FireGL, the number-two workstation card family behind NVIDIA. On the other hand, if your client isn't worried about vendor support and is using applications exclusively based on DirectX, such as Sony Vegas, then by all means save some money and use a consumer card. If you're putting your customer into Adobe Premiere, know that parts of the program now use OpenGL, and, according to Medeiros, Adobe software architects are still two product generations away from having as much OpenGL support as they want. PNY is the single master reseller for all NVIDIA Quadro cards in North America, so while you can research the Quadro product line at NVIDIA's site, you're going to end up buying the cards from PNY through distribution. Two of the SKUs that today present the best value proposition for resellers not specializing in digital video are the Quadro FX 560 (essentially the starting point SKU for OpenGL-based acceleration) and Quadro FX 1500. These are the top-end models from the Quadro entry-level and mid-range respectively.
As with the GeForce cards, there isn't much architectural difference throughout the Quadro FX family. Primarily, they vary by component frequencies and number of pipelines. With a retail price of $299, the FX 560 sports 128MB of GDDR3 memory showing 19.2 GBps of memory bandwidth. For $249, the FX 540 has 128MB of DDR memory but with a bandwidth of 8.8 GBps. However, according to NVIDIA's internal benchmarks in SPECviewperf, the FX 560 shows performance gains of anywhere from single-digits to over 100%. Clearly, a few extra dollars buys a lot of beef in the right application. Interestingly, making the jump to the FX 1500 ($725) does not yield a commensurate spike in SPECviewperf benchmark scores, despite the FX 1500 making the leap to 40 GBps of memory bandwidth. So much for placing too much faith in synthetic benchmarks; an application that leans heavily on GPU memory would show a much greater improvement than SPEC's roughly 10% gain. In addition, the mid-range parts benefit from having a breakout box supporting two dual-link DVI outputs and HD component out on top of conventional standard-def jacks. With dual-link transmitters, the FX 1500 can output up to 3840 x 2400 @ 24 Hz to each panel. Imagine having one 30" display running a Premiere timeline and another monitor running TV-Out for high-def previewing. Match this with a 720p or 1080i camcorder and your customer will have a bona fide 25-megabit HDV FireWire-based high-definition video production solution—something you'd be much harder pressed to create around a consumer card. "To be fair," notes NVIDIA's Zanchi, "most of these applications are very CPU-centric, but you see more and more offloading onto the GPU. Today, it may be 80/20. But in the future, it's going to keep increasing and be more balanced. So having a good CPU is very important. A lot of base memory, as well. But the more effects you use, the more benefit you're going to see from a more powerful graphic card. Red Giant makes plug-ins for Adobe and Avid, and many of those effects are offloaded entirely to the GPU."
Top-shelf workstation builders rarely if ever touch gaming cards. Experience has taught them that the failure rate with consumer cards is too high and that the consistency of the builds across a single model is never the same. For example, PNY's 7900 GTX is different than BFG's 7900 GTX, which is different from MSI's 7900 GTX. These three examples may even all use different drivers. Furthermore, the programmability of the Quadro FX line greatly exceeds the programmability of the GeForce SKUs. SLI frame rendering is available on the FX 3450, 3500, 4500, and 5500, with the 4500 X2 having native SLI onboard because the card sports twin GPUs. Models 4500 and above also support a professional-only feature called genlocking, which is a method of synchronizing multiple video devices to a common cadence. Essentially, genlocking allows users to mix multiple source and destination points without timing errors, which are often visible in the form of a jump during a transition effect, such as a wipe or fade. Also keep in mind that consumer cards will have driver revisions every month if not sooner, and a new driver may not interact the same way with an application as its predecessor. After all, there is no guarantee or certification requiring it. Furthermore, a consumer graphics card usually cycles out of the market in four to five months. You can standardize a client on a bunch of consumer graphics cards, come back in a few months to continue building on that platform, and your common platform is no longer available. When you go to do an RMA, the vendor may send you a different card because the original is out of production, and the new card may well need a different driver than the one you standardized on. Imagine what your video application's tech support staff will say when you call. To be fair, though, there are some who think that workstation software and hardware certification isn't all it's touted to be. "Sometimes I feel like certification is just a marketing image," says Ashley Guy of Guy Graphics. "We'll test boards that we believe work better than the manufacturer's. I think a lot of times it's easier for the manufacturers to provide support for a more limited number of hardware options. You may say, ‘Hey, I've had this problem,' and they'll say, ‘What motherboard are you using? Oh? We've never tested that; it must be the motherboard's fault.' That's how it usually works. If you're not using an approved component, they'll almost always blame problems on that component." The Quadro FX line is not the alpha and omega of workstation cards. As mentioned, ATI does a solid business with the FireGL series, and those who spend much time in digital video inevitably end up at Matrox's door. If you, like many, think of Matrox solely in terms of the dottering Parhelia product line, take another look. Matrox's RT.X2 and Axio lines revolve around adding hardware to Adobe Premiere Pro and Production Studio that software alone wouldn't be able to match. Matrox generally works in tandem with another GPU from ATI or NVIDIA to guarantee realtime performance, realtime output, realtime export to tape, lots of extra effects, on and on. When editing HDV, the Axio LE ($4,495 list) will enable at least two layers of realtime rendering; with DV, you can generally get at least five. In effect, the Axio LE will let you provide customers with a realtime, uncompressed, HD, turnkey video system for under $10,000, which is something of a Holy Grail in today's low-end production market. While Matrox's solution is GPU-agnostic, it's generally held in the video world that Axio favors ATI while Adobe on its own favors NVIDIA. "Interestingly, what Matrox suggests to match with their capture boards today is a lot of ATI consumer," observes PNY's Medeiros. "The reason is that they're simply taking advantage of some of the extensions within the consumer ATI graphic board that allows them to provide DX9 support within that capture board architecture and use host GPU special effects acceleration through their RTX series boards. That's interesting because you have Matrox, clearly in the professional landscape, suggesting the use of consumer boards. But that's the only way they can get that DX9 GPU acceleration. You look at the FireGL line, which ATI would rather have in that space, and the DX9 support isn't there." Regardless of which way you lean with GPUs, the Matrox add-ons can be a serious value for those who need increasingly to work in realtime under Adobe's platform. Moreover, the beauty of it is that you can lead in with a Quadro-based system, then upgrade with Matrox as the client's needs scale upward. There's no need to commit at the outset unless the customer's existing workflow needs the hardware-based performance boost. ...more |
|||||||||||
 Back to top
|
|||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
|||||||||||
