Informatika és tudomány

  • mainruler
    #87
    Na, egy szép zsíros előzetes az S3-tól:


    Introduction




    S3: From Virge to Savage 2000
    Since its inception, the landscape of the 3D graphics market on the desktop has changed significantly. Where there were once dozens of players competing for market share (including the grand daddy of them all, Intel), the industry has now whittled down to a handful of players. The companies that are largely credited for being at the forefront of 3D graphics’ inception, 3dfx and Rendition, are no longer even in business!

    The company we’re discussing today, S3, was one of the larger companies in 3D graphics, although early on many enthusiasts jokingly referred to their Virge line of products as 3D graphics decelerators.

    Realizing this, S3 sought to reinvent themselves by producing an all new graphics core, Savage3D, which was introduced in mid ‘98. With Savage3D, S3 wasn’t shooting for the high end of the market, instead they were going for the mainstream consumer. Not everyone had the cash to shell out for a Voodoo2 graphics accelerator, and S3 was glad to service these consumers. In addition, S3 did not forget its OEM roots it had depended on for so long with Savage 3D. S3 made sure that Savage3D had the right combination of features and price to make these customers happy as well. Overall Savage3D was a limited success in that it reestablished the company as more than just a 2D player with a few 3D buzzwords mixed in for OEMs, but Savage3D came to market later than its competitors and as a result a lot of S3’s previous customers were drawn to NVIDIA’s RIVA 128/TNT lines.

    Following Savage3D (and its more performance-oriented “Supercharged” variant), S3 came back with Savage4. Savage4 was essentially a Savage3D chip with an extra texture unit, and ultimately achieved higher clock frequencies (and up to 32MB memory) with the Savage4 Pro. The real groundbreaking technology in Savage4 however, was its support for S3’s texture compression technology, S3TC. S3TC allowed for high resolution textures that normally would have been hundreds of megabytes down to a few dozen megabytes.

    As a result, Savage4 cards actually ran games such as Unreal Tournament faster with S3TC enabled than with default textures being used. Unfortunately for S3, Savage4 and its follow up variants just didn’t catch on with consumers or OEMs, who favored 3dfx’s Voodoo3 and NVIDIA’s TNT2/TNT2 Ultra, even though it was priced lower than the competition. S3 continued to lose share to its competitors.

    Then, in late 1999 S3 launched its Savage 2000 core. The company was shooting for fill rate figures that were twice that of anything else that was currently available on the market, and Savage 2000 offered hardware transformation and lighting (T&L). Finally S3 would be a player at the high end of the market.

    Savage 2000 to DeltaChrome




    Too little, too late
    Of course, we all know by now that history didn’t end up the way S3 had envisioned it. Poor yields at the 175MHz clock frequency S3 was shooting for resulted in a chip that ultimately shipped at 125MHz (nearly 30% lower than the product that was initially announced) and shipped later than it’s biggest rival: NVIDIA’s GeForce 256, which supported many of the same features. But what really killed Savage 2000 was its poor drivers and lack of hardware T&L.

    S3 promised to correct this with a driver release that ultimately never came, instead S3 became SONICblue; while the graphics division of the company (S3 Graphics) is jointly owned by both S3 and VIA Technologies. S3 Graphics has stayed out of the desktop graphics segment, focusing instead on mobile and powering the graphics behind VIA’s integrated chipsets. To date S3 Graphics has released the SuperSavage, ProSavage, and SavageXP for the mobile segment.



    DeltaChrome is born
    Now S3 Graphics is set to release its first desktop graphics chip, DeltaChrome, along with its companion chip for the mobile segment, AlphaChrome. Both of these graphics cores list 2.0+ pixel and vertex shaders among their long list of features, making them bona fide DirectX 9.0 cards.

    Internally codenamed “Columbia”, DeltaChrome and AlphaChrome have been in development at S3 Graphics for the past two years. Actually, we’re not quite sure if that directly applies to DeltaChrome, as S3 executives confirmed that the mobile part actually came first. S3 Graphics has always wanted to get back into the desktop space, and they’ve decided that now would be the best time to do so.

    With NVIDIA’s current weakness at the high end, this certainly makes sense from a strategic point of view. Like us, S3 probably didn’t envision ATI coming out of nowhere to claim the graphics throne from NVIDIA.

    In any case, this means that DeltaChrome’s roots lie in mobile, unlike most contemporary graphics cores which start life in the desktop segment before that technology is scaled down to mobile. As a result, DeltaChrome sports a die size that is considerably smaller than other DirectX 9 cores (S3 wouldn’t give us an exact figure, but they claimed it will be nearly twice as small as RADEON 9700) as well as boasting lower power consumption.

    It remains to be seen if DeltaChrome will ship with the frequency scaling found on the mobile chip (which will also be present in GeForce FX), but it’s a bit refreshing to see a DirectX 9 graphics card that doesn’t need an external power source. This is due in part, to DeltaChrome’s 0.13-micron manufacturing process and low transistor count.

    Architecture Overview


    DeltaChrome will be built using TSMC’s 0.13-micron manufacturing process. This makes it the second DirectX 9 part to be announced at 0.13-micron, the other of course being NVIDIA’s GeForce FX. In terms of core complexity however, DeltaChrome is dramatically different than the other DirectX 9 accelerators that have been announced.

    While both RADEON 9700 and GeForce FX contain over 100 million transistors, S3 executives state that DeltaChrome will contain just over half the 125 million transistors contained within NVIDIA’s GeForce FX. S3 wants to keep the exact number confidential, but if we’re looking at 80-90 million transistors, that figure would be significantly less than the other 8-pixel pipeline architectures that have been announced. S3 also wanted to make it clear that chip complexity doesn’t revolve completely around transistor count. The number of gates is also very important, although S3 didn’t want to disclose the gate count within DeltaChrome.



    Clock frequencies
    Naturally, with so few transistors within a 0.13-micron core, you’d expect the DeltaChrome to boast breakthrough clock frequencies. However, S3 is currently only shooting for 300MHz for DeltaChrome. In comparison, RADEON 9700 PRO is shipping today at 325MHz (and with a 110 million, 0.15-micron manufacturing process), while NVIDIA is targeting 500MHz for its high-end variant of the GeForce FX.

    With GeForce FX also built off TSMC’s 0.13-micron process and a higher transistor count, we assumed beforehand that DeltaChrome would posses even higher clock frequencies. While we’re sure that there is some variation in TSMC’s manufacturing process (Matrox Parhelia being a good real world example of this), we still assumed DeltaChrome’s clock frequency would be a little higher than 300MHz, after all ATI is already at 325MHz, and they’re at 0.15-micron.

    When we questioned S3 about this they stated that they’ve only tested DeltaChrome at 300MHz, but they’re hoping that the final part will be clocked higher. They’re hoping to get good yields on chips as much as 30% faster than the 300MHz goal, with a lower end part(s) shipping somewhere in the 200-240MHz clock frequency range. They’re ultimately planning to release multiple DeltaChrome products in much the same way they launched Savage3D and Savage4 (which eventually ended up with three variants) at multiple clock frequencies.

    Again, the specifics of everything aren’t written in stone, as final silicon doesn’t exist yet. But with the 300MHz core clock frequency that is currently planned, DeltaChrome will boast a fill rate of 2.4Gigapixels/second, just shy of RADEON 9700 PRO’s 2.6Gigapixels/sec.

    In terms of the memory subsystem, S3 confirmed that DeltaChrome would utilize conventional DDR SDRAM, just like RADEON 9700. The memory interface itself will be 128-bits wide, with peak memory bandwidth topping out at 11GB/sec. Like GeForce3/4, S3 will be utilizing four 64-bit memory controllers to access data, for maximum efficiency. At 11.2GB/sec, this puts DeltaChrome just between the RADEON 9700 (17.6-19.2GB/sec) and RADEON 9500 (8.8GB/sec) series of graphics cards.

    Pixel/Vertex Engines and Shaders




    8-Pixel Pipelines
    S3 has chosen the “V8” moniker to describe the 8-pixel pipeline architecture of DeltaChrome. Like RADEON 9700 PRO and GeForce FX, S3 has implemented eight pixel pipelines in DeltaChrome, with one texture unit per pipe. Like the other DirectX 9 accelerators on the market, DeltaChrome will be capable of processing up to 16 textures per pass.

    The geometry engine is equally capable of pumping out triangles, like RADEON 9700, DeltaChrome features four vertex shaders, although we weren’t given a figure for the chips’ transform rate.

    As we’ve also come to expect from a DX9 graphics accelerator, DeltaChrome supports floating point data formats, allowing for increased levels of precision. We’ve spoken many times before on the benefits of going from integer-based to floating point data formats, so we won’t go into detail on that here, but essentially this will allow DeltaChrome to perform all kinds of complicated math calculations that wouldn’t have been previously possible with older graphics accelerators.

    Breaking down the pixel and vertex engines on DeltaChrome a bit further, we see that the vertex precision is 128-bit (32-bit floating point format) while DeltaChrome supports 96-bit pixel precision (24-bit floating point format), just like RADEON 9700 and unlike GeForce FX, which offers 32-bit floating point format support in both. NVIDIA has claimed that they believe this should theoretically give them a performance advantage in both 64-bit and 128-bit color modes, although we’ll have to wait and see how this plays out when software applications that boast these features begin to appear.



    2.0+ vertex shaders
    Like NVIDIA with GeForce FX, S3 is proud to boast that DeltaChrome’s pixel and vertex shaders go beyond Microsoft’s specifications called for in DirectX 9. While DX9 2.0 vertex shaders are limited to a maximum of 1,024 instructions, both GeForce FX and DeltaChrome support up to 65,536. The number of temporary registers and max loops also goes beyond the specifications set forth by Microsoft.

    DirectX 9 R300 NV30 Columbia
    Vertex Shaders 2.0 2.0 2.0+ 2.0+
    Max Instructions 1024 1024 65536 65536
    Temporary Registers 12 12 16 16
    Max Loops 4 4 256 256



    2.0+ pixel shaders
    S3’s pixel shader instructions also go beyond DX9, although the specs don’t quite match GeForce FX. S3 doubles the maximum number of color instructions supported to 128 and supports additional features such as programmable per pixel gamma correction, a programmable depth shader, programmable table filtering, and a programmable video shader. S3 also adds a few proprietary pixel shading instructions for post-processing capabilities like iDCT, adaptive filtering, deblocking, and deinterlacing.

    DirectX 9 R300 NV30 Columbia
    Pixel Shaders 2.0 2.0 2.0+ 2.0+
    Max Color Instructions 64 64 1024 128
    Max Temp Storage 12 12 64 N/A

    Finally, DeltaChrome’s pixel shaders feature bi-directional Z and color buffers. Standard DirectX 9 pixel shaders utilize unidirectional color and Z buffers. This allows the color and Z buffers to send or receive data from the pixel shaders.

    Occlusion Culling/Dual Display, HDTV, etc.




    Deferred Rendering
    Of course, having all the memory bandwidth in the world does nothing if that bandwidth is wasted on unnecessary data. Matrox’s Parhelia 512 is a perfect example. On paper Parhelia offers an astounding 20.8GB/sec of bandwidth to the graphics processor, that’s more than any other graphics accelerator on the market. Yet, in high-resolution scenarios where this bandwidth is needed the most, Parhelia falls short of NVIDIA’s GeForce4 Ti 4600, which is limited to 10.4GB/sec (half of what Parhelia offers).

    This is due in part, because the Matrox Parhelia lacks occlusion-culling technology. While ATI has depended on its HYPERZ and NVIDIA has had its own culling technology since GeForce3, Parhelia has nothing, essentially wasting much of its memory bandwidth.

    For DeltaChrome, S3 has implemented its own occlusion technology, first and foremost would be its Hierarchical Z. Hierarchical Z essentially splits the Z-buffer into blocks of pixels, then determines if those blocks will be visible on the final display or not. If the block of pixels is hidden, the block is discarded and moves on to the next block, increasing efficiency.

    While this will certainly improve DeltaChrome’s performance in high-resolution and/or anti-aliasing environments, it appears that DeltaChrome lacks some of the compression technologies that have found their way into NVIDIA and ATI products such as color compression and Z compression. Fortunately DeltaChrome does support technologies such as Fast Z-Clear, Alpha, and RGB Clear. This saves time and conserves memory bandwidth.



    PicturePerfect
    PicturePerfect technology refers to the anti-aliasing techniques used by DeltaChrome in both 2D and 3D modes. There really isn’t much new to say as far as 2D text is concerned, as far as we can tell from S3 documents, RADEON 9700 already accomplishes much of the same tasks today, and S3 is a little vague regarding its 3D anti-aliasing implementation, but here’s what we can tell you so far.

    Like other RADEON 9700 and GeForce series, DeltaChrome utilizes multi-sampling. We’re not certain if S3 goes so far as to offer programmable jittered multi-sampling like ATI, but we do know that modes of 2, 4, and 6 samples per pixel are supported (matching the basic specs of 9700). For anisotropic filtering, we don’t know if bilinear or trilinear samples are used. But we do know that modes as high as 16x will be supported, again matching RADEON 9700’s specs.



    DuoView
    As it sounds, DeltaChrome supports dual displays via its DuoView technology. But what really separates it from all other graphics architectures is that it natively supports outputting to your HDTV thanks to its Hi-Def HDTV encoder. This gives S3 quite an edge over the competition: could you imagine outputting straight from your PC or laptop straight to your HDTV? With DeltaChrome, you can!

    Resolutions of 480p, 720p, 1080i, and 1080p are supported via a component (YPbPr) output, while 480i is supported via standard S-Video/Composite output. DeltaChrome will offer full range RGB to YUV color space conversion with hue, saturation, and contrast adjustment.



    Chromotion Programmable Video engine
    Like the other DirectX 9 accelerators that have been announced, DeltaChrome uses its pixel shaders to clean up streaming Internet video signals (such as webcasts) as well as processing MPEG-2/4, Windows Media 8/9, and other video standards. Nothing really new here, but it’s a cool feature for those of you who do a lot of video editing, or those of you who would like to clean up some of your vid caps.

    Pricing and Availability




    Availability
    By now you’re probably wondering how much DeltaChrome will cost and when it will be available. To answer the latter question, S3 stated that they’re shooting for retail availability at the end of Q2. That’s right, the end of Q2. That means you may not see DeltaChrome on store shelves for as long as five months from now.

    In the graphics world, five months is an eternity, so we’re hoping that S3 can push up DeltaChrome’s scheduled release a bit more. RADEON 9700 essentially supports everything DeltaChrome does, in some cases falling short while in others going a step beyond DeltaChrome, and is available today. NVIDIA’s GeForce FX will also be out on the market well before DeltaChrome.

    If S3 sticks to its current schedule, not only will RADEON 9700 and GeForce FX be obtainable on the market, but ATI’s successor will probably be available as well, and it’s very likely that NVIDIA’s RADEON 9500 competitor, NV31, will also be out on store shelves. This will make life very tough on DeltaChrome, as video card manufacturers here in the US will likely be hesitant to adopt it, and even if they do there’s no guarantee that it will be a sales success. Many consumers still remember the Savage 2000 debacle three years ago and S3’s inconsistent driver support. We’re hoping that S3 is taking the extra time to polish DeltaChrome as much as possible, the last thing they should do is release another product with poor drivers.



    Pricing/Board variants
    In terms of pricing, we can’t get too specific, as DeltaChrome is many months away from release. But S3 has confirmed that they will undercut prices on ATI and NVIDIA’s equivalent product and that there will be multiple DeltaChrome variants upon release. In particular, S3 currently plans to release two 8-pixel pipeline products, with the only difference between the two cards being clock speeds. As we stated previously they’re currently planning to produce a 200-240MHz variant, as well as a 300MHz model on the higher end. But ultimately they’re shooting for a 300MHz+ product if the yields are there. On the very low end S3 is planning a 4-pixel pipeline product that will also ship in the 200-240MHz clock speed range. This is essentially the same strategy ATI has employed with their RADEON 9500/9700 series.

    Based on DeltaChrome’s fundamental architecture, we don’t see any reason why S3 won’t be able to live up to these claims, as their product contains significantly fewer transistors than the ATI/NVIDIA equivalent, and boasts a smaller die size. When you factor in the 0.13-micron manufacturing process (on 12” wafers we’re told), DeltaChrome should certainly be cheaper for S3 to produce.

    Conclusion


    As interesting as DeltaChrome is, we can’t help but wonder how well it will do in the American market when it’s released. With NVIDIA dropping the ball last fall, the window of opportunity for S3 is now, not four or five months from now. ATI cashed in on this with RADEON 9500 and RADEON 9700, and S3 would have an easier shot at doing the same if DeltaChrome were ready for mass production today, even if the final clock speed is 300MHz.

    Due to architecture differences S3 will still be able to maintain a price advantage over ATI and NVIDIA, but by the time DeltaChrome ships that advantage may not be as significant. Just as it makes no sense to pay $180 for RADEON 9500 when RADEON 9500 PRO is currently priced at $200 (we’re quoting list prices), we wouldn’t be surprised to see consumers willing to pay an extra $30 or $40 to purchase a vanilla RADEON 9700 over DeltaChrome. Of course, we’re assuming that RADEON 9700’s successor, R350, will be out on the market by the time DeltaChrome is released (causing RADEON 9700 prices to plummet), which may or may not end up holding true.

    We can also tell you that S3 has plans for two additional products following DeltaChrome. The first, DeltaChrome 2, will be based on essentially the same architecture, the only differences being that it will run at even higher clock frequencies and support DDR-II memory.

    Even further down the pipe is S3’s next generation product, internally codenamed “Destination Films”. This product will support next generation pixel and vertex shaders, and offer considerably more performance than DeltaChrome. We’re really looking forward to this product, as it has been built from the ground up for the desktop market. Lets face it, DeltaChrome was initially conceived for the mobile market, and it many places, it shows. How many desktop products do you know of that support hardware-assisted display rotation after all? While this is a great feature for Tablet PCs, you won’t see this taken advantage of in any conventional PC applications.

    In any case, it’s always good to see another player entering the desktop graphics segment, as increased competition is always a good thing for consumers. STMicro shocked everyone a few years back with Kyro II, and NVIDIA responded with price cuts and multiple GeForce2 MX variants to compete. If S3 is able to deliver an NV31/RADEON 9500 killer with DeltaChrome, S3 could make a very dramatic comeback later this year. It’s too bad we’ll have to wait a few months to find out.
    \"The problem of leadership is inevitably: Who will play God?\"