Original Link: http://www.anandtech.com/show/1844
With a focus on the low-cost and integrated video setups, it is not too often that we hear from S3 Graphics, VIA's video-based partner in crime. As their last major low-cost product (GammaChrome) was introduced back in mid-2004, S3 has been overdue for some time now for a refresh to keep pace with ATI and NVIDIA's latest products, which have since then eaten in to what market and mind share S3 has previously held. Today, that new product comes to life in the Chrome S20 series, and while S3 may not be attempting to re-enter markets they have since long left, they still are nonetheless very interested in keeping themselves in the low-cost markets.
To that extent, it should come as no surprise that today's launch is, in fact, not taking place in North America, but rather in China where there is a larger and more diverse market for S3 to tap. Unfortunately, this is a pure paper launch at this point, as S3 does not yet have any final boards, nor were they able to supply us with a pre-release board to review. While we will have a pre-release board to look at soon, S3 has in the meantime let us in on all the details of their new products, which we bring you today.
The ChromeS20 GPU
As has become a trend lately in the GPU industry, there is often a feeling of "if it ain't broke, don't fix it" occurring within the major players industry, and this is occurring with S3 as well. As with NVIDIA and the G70 core, S3 already has a strong design in their previous GammaChrome parts, so much of the design principles of the Chrome S20 are not surprisingly simply refinements of what GammaChrome could do. The advantage of this, of course, is that S3 can start with proven technologies at a cost of not having much flexibility to introduce many new features.
Before we break down the feature set of the Chrome S20, we should note that the GPU powers 2 products under the series of the same name: the full-powered Chrome S27 and the lower-cost Chrome S25. Both parts use the same GPU and have the same base features, so the difference in the two cards will come down to how they're configured (more on that later). So, other than a performance difference, both parts should behave identically.
To that extent, for readers familiar with the GammaChrome feature set, there should be few surprises in what the S20 can offer. At its heart, the S20 is a Shader Model 2.0+ chip employing an 8/4 setup - that is 8 pipelines each with a full pixel shader, and 4 vertex shaders. Built on a 90nm process, the S20 follows S3's traditional strengths in keeping chip costs and heat output down, so the S20 is a nearly-anemic chip compared to the bulkier parts offered by ATI and NVIDIA, with the S20 only being composed of under 70 million transistors. Because S3 is partnering with Fujitsu for their 90nm process, we're not sure what to expect as far as initial yields on the chip. With a smaller chip than most, S3 should have an easier time attaining the yields that they want (compared to other 90nm parts), and perhaps hit some higher clockspeeds too.
With such a small transistor count, S3 is not hesitating to point out the benefits of such a small device. Between the chip size and the ability to reduce power usage by the S20 when it's not being fully used, S3 claims that the highest average power usage of the fastest S27 board is only at 30W, with it falling to 20W for a 64bit S25 board, and lower still from there. As a result, S20 series parts can be built in a fanless configuration, putting S3 in a good position to attack the mobile and HTPC markets. We will still see many designs with fans because it is often cheaper to attach a very small fan rather than a good heatsink. But for those who need a cool and quiet graphics solution, S3 has opened the door for manufacturers.
As for the RAM and boards with which the S20 will be coupled, it follows the GammaChrome tradition of being a PCI-express-only part, with support for anywhere between an x1 and x16 slot. The S20 will support boards featuring anywhere between a 32bit and 128bit memory bus, which (when coupled with GDDR3 RAM) should be plenty of bandwidth to keep the S27 well fed. The S20 also supports GDDR1 and GDDR2, which will be used on the S25 boards; and interestingly enough, while the S20 GPU supports a 512MB configuration, it only does so with GDDR2 and not GDDR1 or GDDR3. However, as we noted in Tour de Taiwan, GDDR2 prices have fallen below GDDR1, so GDDR1-based cards are unlikely at this point anyhow.
As a further cost-cutting measure, S3 has also implemented a way for the S20 to utilize system memory via the PCI-Express bus (á la TurboCache) dubbed AcceleRAM. This is an enhancement of the GammaChrome's ChroMAT ability, which allowed utilizing system memory for textures, while AcceleRAM will allow full utilization for textures, pixel shader operations, and even the frame buffer itself. Rendering directly to system memory (rather than rendering local and copying out) is important in maintaining performance in low end parts. Like their competition, S3 is gearing this feature towards their cheapest cards, and will only be utilized on the cheapest S25 parts. At this time, S3 will only support 32MB configurations when using AcceleRAM, which should cut down on some of the confusion that we've seen in the past with other implementations of this kind of technology.
Also being brought over from GammaChrome is S3's video processing capabilities, dubbed Chromotion. Now in its 3 rd iteration, Chromotion typically represents S3's strongest hand in features, so S3 had to change little to stay competitive with video processing. The S20 supports both WMV9 and MPEG2/4 decode acceleration, complete with deblocking and adaptive deinterlacing, but H.264 acceleration is noticeably absent - this is something that we would have expected given S3's previous focus on video processing. New to the S20 is non-linear scaling, which allows a 4:3 image to be scaled to 16:9 in an unequal method that results in the sides of the images being scaled more than the center, in order to improve the preservation of the proportionality of the objects at the center. Also new is a technique called PanelDrive Response Time Enhancement, which according to S3 improves the response rate of an LCD by actually having the panel attempt to change to another color beyond the desired color, and then stopping short to reach the right color. We've seen others implement features like this, but without a sample on hand, we can't confirm how well it works.
As of press time, S3 has not decided on the final specifications for their cards, so take all of this with a proper grain of salt. Starting first with the Chrome S27, S3 is shooting for a core speed of 700MHz, and pairing that with some configuration of 700MHz GDDR3, likely 256MB. As their fastest product, S3 intends to position it against the GeForce 6600, which means that such a card would settle below $125 - the high end of the current prices for the 6600. As far as what will actually ship, we're taking a wait-and-see attitude, largely dependent on the manufacturing issues mentioned above. There's no way to tell at this point if S3 will be able to pull off 700MHz on such an unwieldy process.
As for the S25, S3 is shooting for a core speed of at least 600MHz, and a memory clock not in excess of 400MHz. Like with the S27, we'd expect the S25's core speed to still be up in the air, though in this case it is largely dependent on what the S27 premiers at, due to the need to keep the products reasonably different. Because S25 parts can ship at up to a 128bit bus like the S27, it's a reasonable assumption that there will be a mix of S25 parts with and without AcceleRAM, covering the sub-$100 market for which the S27 would be too expensive. What configurations the AcceleRAM cards will use is also unknown at this time, as S3 gave us a theoretical example of a 64bit card using 400MHz GDDR2 RAM, but given current memory prices, we'd be surprised if this configuration didn't appear. It should be noted, however, that such a configuration puts S3 in an odd place as far as power consumption goes. GDDR2 is well-known for its high power requirements compared to other versions of GDDR, so we're not sure how S3 intends to pull off a 20W total unless they used some other version (possibly including standard DDR1 or DDR2).
MultiChrome - S3 Gets SLI
Although it's not ready yet, S3 also has one last feature in the S20 that we haven't covered yet: MultiChrome, their implementation of multi-card rendering. At this point, S3 hasn't announced any specific details, let alone shown us a working sample, so we only have a small amount of confirmed information.
MultiChrome is intended for the S27 cards only, and while the GPU on the S25 should be capable of supporting it, this is another form of product segregation between the two cards and is unlikely to change any time soon. For MultiChrome, S3 is going the connector-less route, which we have seen other GPU makers pull off in their similarly low-end products too. Lastly, S3 is shooting for allowing MultiChrome on multiple platforms, and while they haven't been specific yet on what platforms these will be, we'd expect no less than proper support under NVIDIA, VIA, and Intel chipsets. We were told that anything supporting more than one physical x16 PCIe connector would work with MultiChrome, but that does remain to be seen.
What we don't know at this point are the how's and when's. S3 has no solid date on when they'd expect to actually enable MultiChrome, and they haven't given us any further information on how it will work. Because the S20 is based upon the GammaChrome and not a completely original design, it's entirely possible that they are in a similar situation as what ATI was with the x800 series and are unable to make significant GPU changes, having to rely on drivers and external hardware instead.
With their launch today in China, S3 has showcased their intentions on remaining relevant in a market of which they have been increasingly pushed out. We are disappointed at this time, however, that this is a complete paper launch, and that S3 isn't even capable of seeding reviewers with hardware (let alone customers), and we hope that they'll realize the error soon enough and start following NVIDIA's example of on-time delivery. In the meantime, we will bring you more details on the hardware and the impending North American launch as we get them.