Test Drive: Nvidia Quadro CX and Adobe CS4, Part 2
This month, we're looking at how Nvidia's Quadro CX technology (around $2,000 at B&H Photo Video) can accelerate performance in Adobe Creative Suite 4 (CS4). In the first installment, we learned about why having a powerful graphics card wasn't all that important to previous versions of the Adobe Create Suite, and how CS4's embracing of OpenGL makes graphics performance important again. However, we also saw that Adobe had certified Nvidia's GeForce GTX 260 for use with After Effects and Photoshop, and that this $300 card costs a fraction of the Quadro CX's retail price tag. So is the Quadro CX worth the significant investment and why?
| Related Links |
||||||
|
I looked at three applicationsPhotoshop, After Effects, and Premiere Proto answer that question, running comparative tests that I'll detail below. Basically, when considering this card, it comes to two aspects: capacity and performance.
Figure 1. This Nvidia-supplied Photoshop test convinced me that I needed OpenGL, but not necessarily the Quadro CX.
As with all performance-related reviews, I looked to Nvidia for help in testing the CX. After all, the company knows its product better than I do and can make my efforts much more efficient.
In this regard, I was very happy to see a longish reviewer's guide with a DVD full of content. Unfortunately, at least in regards to Photoshop, the projects were designed to convince me that I needed a card with OpenGL, which I already knew, not to distinguish the performance of OpenGL cards such as the GeForce GTX 260 ($300) and the Quadro CX (around $2,000). I discussed this with my Nvidia rep, who focused on the benefits of the CX's capacity, which I detail below, not necessarily performance.
To chart a course for Photoshop performance testing, I looked back at tests I had performed in 2007 and 2008 to benchmark the performance of a range of multiple-core workstations. Specifically, to test the performance of the Quadro CX, I ran several of the same tests on an HP xw8600 workstation running 64-bit Windows Vista. First, I tested with the Quadro FX 1700 that came with the workstation, and then I installed the Quadro CX and tested again, finding that the performance difference was negligible.
This makes perfect sense, since there are two tiers of performance enhancement that relate to a product such as the Quadro CX. The first, and most significant, relates to the boost provided by OpenGL support as compared to a card without Open GL support. Since the Quadro CX 1700 comparison card also supports OpenGL, there was little difference in performance here.
The second tier relates to specific hardware-supported features relevant to Photoshop performance on the Quadro CX that aren't available on 1700. Again, when I asked Nvidia to suggest tests that would highlight these differences, the company didn't provide any; but it did provide several tests that highlighted the advantages of the CX over the 1700 when used with After Effects.
If you're running CS4 on a graphics card without OpenGL graphics support, the CX will deliver a very significant increase in performancethough it's likely that any card that supports OpenGL, such as the $300 GeForce GTX 260, could deliver similar benefits. On the other hand, if your current graphics card does support OpenGL and is reasonably current and powerful, installing the CX won't significantly change your editing experience.
If you read any other reviews of the CX that discuss Photoshop performance, be sure to identify the baseline comparison platform. If it's a computer without OpenGL graphics, expect the praise to be flowing. That's fine; just remember that if your current card supports OpenGL, as I'm guessing it does, you won't see the same improvement.
Otherwise, the Nvidia rep did point out that the 1.5GB of onboard memory afforded the Quadro CX a greater capacity than the 1700. Specifically, understand that Photoshop can open a maximum of 16 OpenGL-accelerated images at one time. When displaying on a 30in. panel at 2560x1600, the Quadro CX can support all 16 images, while the Quadro FX 1700with only 512MB of memorycould accelerate only nine. Similarly, the Quadro CX could support 16 accelerated images on a 4096x1536 desktop, while the 1700 could support only six.
If you're a Photoshop power user who edits multiple images all day long, you'll find this capacity highly useful. On the other hand, if you use Photoshop primarily as an equal component of the overall CS4 suite, say to design Encore menus or Premiere Pro titles, the CX is probably more than you need.
As mentioned, the Nvidia rep did identify several scenarios where the Quadro CX supported specific After Effects operations that the 1700 didn't, including depth of field and cartoon effects, and he supplied tests that confirmed the advantage. In these tests, I timed how long it took to preview each project using the 1700 and then the CX. Table 1 shows the details.
As you can see, the benefit when previewing the popular cartoon effect was quite profound, with a noticeable boost in the preview speed of the bilateral blur. I'll note that in the reviewer's guide, Nvidia cited a 38X performance boost for the cartoon operation, but again, that was the difference between running the test with the CX and rendering without OpenGL and relying solely upon the CPU, not the CX vs. a different graphics card with OpenGL support.
With Premiere Pro, OpenGL support is limited, and it doesn't appear to leverage any hardware-specific features of the CX over those of other OpenGL cards. For this reason, I didn't compare the 1700 to the CX with any OpenGL-specific tests. Rather, I focused on H.264 encoding, which the Quadro CX accelerates with a technology called Elemental Technologies (ETI) RapiHD.
From a workflow perspective, Adobe Media Encoder simply has another format option called ETI RapiHD. When you select that option, you encode using the CX; when you encode using any other format, such as H.264-Blu-ray, you encode using Adobe's own codecs driven solely by the computer's CPU. Nvidia supplied a sample 4-minute project, consisting of two 2-minute trailers for the latest James Bond movie, Quantum of Solace.
I started out on with the full eight cores on the HP xw8600, and I found a reduction of encoding time of 29 percent, from 5:14 to 3:43. I then removed one processor and tested again. Encoding time for the MainConcept encode jumped to 6:33, while RapiHD took about the same time, reducing encoding time by 44 percent. I produced at 25Mbps, and all encoded streams looked great.
In my discussions with the Nvidia rep, he had mentioned that performance times would, to some degree, depend upon the type of effects being rendered on the timeline. That's because if significant effects were involved, Premiere Pro would have to render the effects, then hand off the frames to either encoder for transcoding to H.264. If Premiere Pro's rendering was a bottleneck, it would slow both encoders and erode some of RapiHD's competitive advantage. In effect, since no rendering was involved, my initial tests were a best-case presentation.
Accordingly, I added three effects to the Bond previewscolor correction, dust and scratches, and camera viewand I used keyframes to shift the camera view over the course of the video. This boosted overall rendering time significantly, and as anticipated, it abridged the CX's performance advantage down to 24 percent. Running all eight cores on the xw8600 speeded both operations, and the performance advantage afforded by the CX rose to 35 percent.
The Nvidia press materials didn't focus on encoding H.264 for streaming. However, since many producers care more about H.264 for streaming than for Blu-ray, I ran some trials, again encoding with RapiHD and MainConcept and timing the results, which are shown in Table 3.
I tested by encoding files directly in Adobe Media Encoder, which means no effects on the Premiere Pro timeline, again the best-case test scenario for the CX. With 640x480 files, RapiHD shaved almost 49 percent off the encoding time when using four cores, and 32 percent with eight cores.
However, the results were anomalous with a 720p file encoded to the same 1Mbps, Here, when running eight cores, RapiHD actually took longer than MainConcept, which made no sense, though I tested twice. Bumping down to four cores proved much more efficient, and the CX produced time savings in the 51-percent range. I compared the quality of the RapiHD and MainConcept 720p files and found them very similara good result for RapiHD since Premiere Pro's Main Concept encoder is a very high-quality producer.
Where does this leave us? Overall, Nvidia has done a very good job establishing that an OpenGL graphics card is a must for even casual CS4 users. However, the company hasn't shown enough hard data, or provided me with enough direction, to prove that the CX should be that card except in a few, very well-defined instances.
Photoshop users with multiple 30in. monitors and the need to juggle the maximum number of OpenGL-accelerated images will see a return on investment; otherwise, you probably can get away with a less expensive card, if not your current card. With After Effects, cartoon-heavy effects will respond dramatically; otherwise, it takes a lot of data to justify the CX.
For Premiere Pro users, several H.264-specific use cases come to mind. If you have an older computer and just need to transcode unedited video from one format to H.264, the CX could be an inexpensive way to quickly boost the performance of that system. At the other end of the spectrum, if price is no object, and you want the biggest, baddest, fastest editing and encoding H.264 for Blu-ray system on the planet, get an eight-core system such as the HP xw8600 and spring for the CX. On the other hand, if producing H.264 for Blu-ray isn't a big chunk of your job description, you almost certainly can find a much less expensive OpenGL card that performs similarly in day-to-day operation.
I hope (and challenge) Nvidia will do more to objectively show when and where the CX truly lives up to the claim that it is "the accelerator for Adobe Creative Suite 4"and more cost-effective than cards costing a fraction of the price, even after a new bundle deal announced by Nvidia.
On the other hand, don't throw out the baby with the bathwater. Nvidia became my graphics vendor of choice back in the late 1990s, and in my view, no graphics-card vendor can match its performance, affordability, and most importantly, stability. I have too many computers to count in my office and home, and Nvidia happily lives in at least 95 percent of them, and 100 percent of those acquired in the last four years. There's no doubt that if I were buying a graphics card for CS4 production today, it would be an Nvidia card. It just wouldn't be the Quadro CX.








