S3 Graphics has just pulled back the curtains on its newest series of graphics chips, the Chrome S20 family. S3 turned to a new foundry partner, Fujitsu, to produce these chips, as we reported yesterday. Fujitsu manufactures the GPUs using a low-K, 90nm fab process, and S3 claims this advanced fabrication process, along with S3’s own Power Wise power management techniques brought over from its mobile products, has helped the Chrome S20 series achieve killer performance per watt.
The S20 series is made of up two variants, Chrome S25 and Chrome S27. From the looks of it, the S25 and S27 are probably the same basic silicon spun out into separate products for different markets.
Although these are new GPUs, the Chrome heritage of the S20 series is obvious. (Those of you who are curious about the architecture can read my review of one of the first DeltaChromes for more info.) The chips come with native PCI Express interfaces, and like the rest of the Chrome line, the graphics core supports DirectX 9’s Shader Model 2.0. The S20’s pixel shaders support a “full” 96 bits of floating-point precision per pixel, and they sport more registers than past Chrome GPUs, for faster, more efficient execution of code. Most notably for performance, the S20 series GPUs combine four vertex shaders, eight pixel shaders, and four texture units per GPU. That should give them a mix of attributes fairly similar to NVIDIA’s GeForce 6600 series, although real-world per-clock performance comparisons between different GPU architectures are tough to handicap based on specs alone.
The sexier of the two parts is the S27, intended as a competitor to the GeForce 6600 and the Radeon X1300 Pro. This chip will run its eight pixel shader pipes at a staggering 700MHz, and the fastest cards based on the S27 will use a 128-bit interface to talk to 128MB of GDDR3 memory, also clocked at 700MHz. S3 plans to sell these 700MHz/700MHz cards for about $100, which could make the S27 a reasonably appealing option in the so-called mainstream graphics market. NVIDIA’s recently announced GeForce 6600 with 256MB of DDR2 memory, for comparison, has reference clocks of 350MHz core and 400MHz memory—much less than the S27. The $99 version of the Radeon X1300 has 450/500MHz clocks and comes with 128MB of RAM. S3 believes the S27’s performance should be more than competitive in this segment of the market.
The S27 has learned a new trick, too: a multiple graphics card technology that S3 calls… MultiChrome, of course. Like the current low-end SLI options from NVIDIA—and like ATI’s plans for Radeon X1300 CrossFire—MultiChrome will need no external connectors to operate, instead passing data over PCI Express lanes. To their great credit, S3 and VIA have decided to make MultiChrome a platform-independent affair; any capable PCI Express implementation should be able to work with MultiChrome, with no bogus driver-based lock-outs like NVIDIA and ATI impose with SLI and CrossFire.
I somehow failed to ask how S3 handles load sharing between two cards and whether they’re planning to use a feature like application profiles to help determine which load-balancing method is best for a given application. They did say, though, that performance scales up to roughly 75% higher than a single S27 card, which isn’t bad given the connector-less implementation. I expect to get my hands on a dual S27 configuration before too long, so we’ll get those questions answered soon.
The S25 is the lower end part, and with peak clock speeds of 600/400MHz for GPU core and memory, it will do battle with the Radeon X1300 HyperMemory and GeForce 6200 TurboCache cards of the world. Indeed, S3 has endowed the S25 with its own AcceleRAM tech, which is the same basic thing as TurboCache and HyperMemory: the S25 can read from and write to main system memory over the PCI Express interface, just as it can to local memory. S25 AcceleRAM cards will come with anywhere from 32 to 128MB of local memory, and they’ll allocate more space in system memory to expand their effective storage space.
Both S20 series GPUs come with the latest version of S3’s Chromotion video engine, massaged and revamped from past Chrome products. Chromotion 3.0 will support WMV HD acceleration, but stops short of support for H.264 decode acceleration—a compute-intensive task that will likely challenge the ATI and NVIDIA chips in this segment. The S20’s display engine can drive an HDTV in modes up to 1080p or a couple of LCD displays via a pair of DVI channels. Like most DX9 GPUs, S3 expects the S20 to be capable of handling Windows Vista’s AeroGlass interface, and the S20 can even run in 10-bit-per-color-channel video modes for high-color photo editing and the like.
As one might expect given S3’s focus on performance per watt, mobile versions of the S20 series are in the works. Also, S3 has big plans for future GPU generations. Rather than aim for Shader Model 3.0, where the S20’s primary competitors are now, S3 is planning a jump to Shader Model 4.0 in its next-gen GPUs.
As ever, the Chrome S20’s success in North America will hinge on actual availability of the product. (We can talk about driver quality and game compatibility, as well, but not if the cards aren’t selling.) S3 says cards based on the Chrome S20 chips will be available today in China, and that it’s working on getting major, respected U.S. e-tailers to make cards available on these shores. We’ll be watching to see what develops.
Share
