3DVelocity, your online hardware resource

Contact The Author
Wayne

Trident XP4 Technology Overview
22nd August 2002

Introduction : Many people in the industry, myself included, expressed some surprise when news broke on the 5th of August that Trident were to dip a toe back into the hostile waters that are today's graphics card market. With a little more thought though perhaps this move isn't so surprising. With a history in desktop graphics stretching way back to 1987 Trident were very much part of the primordial soup that gave rise to the modern graphics accelerator and though they decided to gracefully bow out and focus their attention elsewhere when the market began to expand that doesn't necessarily mean they don't have what it takes to rub shoulders with the best........ it's no guarantee that they do either.

I've tried hard to find the technical details that might help us decide what the XP4 is potentially capable of, and I have to admit I failed rather miserably. Hard facts and figures seem a little hard to come by but we do have some preliminary specifications to glance over.

What We Know : What we do know is that Trident's XP4 will come in four flavours, a discrete notebook solution and three desktop solutions known as the "T1", "T2" and "T3". The GPU will be manufactured on a 0.13um process using copper interconnects and features an almost unbelievably low transistor count at 30 million.

We also know that Trident's maths don't quite add up. They're claiming 70% of the performance of the new ATi Radeon 9700 which based on current performance figures puts it on a par with NVIDIA's Ti4600 yet they're also claiming 80% of the performance of the Ti4600? Clearly we'll have to wait and see the actual hardware before we know for sure but if I had to make an educated guess I'd say we can expect Radeon8500 level performance which is still a good showing for a $100 DirectX 8.1 graphics card. I now simply hope Trident prove me wrong and exceed this with clever drivers and optimal use of what it now seems will be a tile based rendering engine.

Let's get the boring bit over and take a browse over the specs as quoted by Trident ~

Mobile XP4 Specifications

Overview
• Performance 3D graphics solution for low-power notebooks
• Scalable to mainstream and thin/light portables
• CoolPower™ management (3.0 watts maximum dissipation)
• Power supply: 1.2v, 1.5v, 2.5v, and 3.3v
• 30 million transistors in 0.13um CMOS process
• Standard 612 BGA package (31mm x 31mm)

Performance
• 250 MHz graphics engine clock
• Pixel fill rate: 1 billion pixels/sec (peak)
• Texel processing rate: 8 billion texels/sec
• Up to 666 MHz DDR memory clock
• Up to 10.4 GBytes/sec memory bandwidth (peak)

SmartTileTM Memory Bus Architecture
• 128/64-bit Double Data Rate (DDR) memory at up to 666 MHz
• Up to 256 MBytes of local frame buffer with 16Mx16 memory
• Supports 8Mx32, 4Mx32, 2Mx32, 16Mx16, 8Mx16, 4Mx16
• Supports hierachical pixel tiling

Output Displays
• Four simultaneous outputs: CRT, TFT, DVI and, TV out
• Dual-channel LVDS driver for interface to TFT
• Single-channel TMDS driver for interface to DVI with HDCP
• 420 MHz RAMDAC for interface to standard CRT
• Digital interface to Trident's TVX2 NTSC/PAL encoder

CoolPowerTM Management
• Unified software and hardware architecture for power management
• Dynamic clock gating, frequency scaling, work load balancing
• Supports Intel® defined Device Performance States (DPS)
• Optional battery optimizer under user control

DirectX 8.1
• MICROSOFT® graphics standard for Windows® XP
• Fully programmable Vertex Shader and Pixel Shader hardware
• All DirectX 8.1 texture types (volume, projective, bump maps, etc.)
• Special low resolution, edge correction for better image quality

BrightPixelTM Graphics Engine
• Up to 250MHz engine clock
• Based on hierachical pixel tiling for rasterization
• Four independent pixel pipelines (512-bit wide)
• 4 pixels/clock with 8 textures/pixel
• Texture size up to 4K x 4K and non-power of 2
• Bi-linear, tri-linear and anisotropic texture filtering
• Full anti-aliasing support for texts, lines, scenes
• Special hardware acceleration for silhouette anti-aliasing
• Special multi-resolution depth buffer
• Special bandwidth reduction hardware via compression
• Multi-level caches
• Fully OpenGL-compliant blending including fog & depth cueing

Video Engine
• State-of-the-art video de-interlacing (beyond Bob-and-Weave)
• Supports HDTV resolution up to 1920 x 1080
• TrueVideo® provides bilinear interpolation with proprietary edge recovery scaling
• Dual apertures for simultaneous access to graphics and video display memory areas

DVD Support
• Microsoft's DirectX Video Acceleration (DXVA)
• Includes both Motion Compensation and IDCT hardware
• Real-time playback (30 fps) of 9.8 Mbps and MPEG-2 video bitstream,
with 85% CPU headroom for other applications

Software
• Windows® XP, 98, ME, 2000, NT 4.0, and 5.0
• DirectX 7.0 and 8.1
• DirectX Video Acceleration (DXVA)
• DirectShow 3.0 and 4.0
• OpenGL ICD 1.2 and 1.3
• Linux

XP4 For Desktop

	Memory Clock	CPU Clock	Fill Rate	Memory Bandwidth
XP4 - T1	Up to 500 MHz (64 bit memory bus)	250 MHz	1 billion pixels/sec	Up to 4.0 GBytes/sec
XP4 - T2	Up to 500 MHz	250 MHz	1 billion pixels/sec (peak)	Up to 8.0 GBytes/sec
XP4 - T3	Up to 700 MHz	Up to 300 MHz	1.2 billion pixels/sec (peak)	Up to 11.2 GBytes/sec

XP4 T1 Specifications

Overview
• Performance 3D graphics solution for mainstream value desktop
• 64-bit DDR memory bus with up to 64 Mbytes frame buffer
• CoolPowerTM management (3.0 watts maximum dissipation)
• Power supply: 1.2v, 1.5v, 2.5v, and 3.3v
• 30 million transistors in 0.13um CMOS process
• Standard 612 BGA package (31mm x 31mm)

Performance
• 250 MHz graphics engine clock
• Pixel fill rate: 1 billion pixels/sec (peak)
• Texel processing rate: 8 billion texels/sec
• Up to 500 MHz DDR memory clock
• Up to 4.0 GBytes/sec memory bandwidth (peak)

SmartTileTM Memory Bus Architecture
• 64-bit Double Data Rate (DDR) memory at up to 500 MHz
• Up to 64 Mbytes of local frame buffer
• Supports 8Mx32, 4Mx32, 2Mx32, 8Mx16, 4Mx16
• Supports hierachical pixel tiling

Output Displays
• Three simultaneous outputs: CRT, DVI and TV out
• Single-channel TMDS driver for interface to DVI with HDCP
• 420 MHz RAMDAC for interface to standard CRT
• Digital interface to Trident's TVX2 NTSC/PAL encoder

CoolPowerTM Management
• Unified software & hardware architecture for power management
• Dynamic clock gating, frequency scaling, work load balancing
• Supports INTEL® defined Device Performance States (DPS)

DirectX 8.1/9.0
• MICROSOFT® graphics standard for Windows® XP
• Fully programmable Vertex and Pixel Shader hardware
• All DirectX 8.1 texture types (volume, projective, bump maps ...)
• Special low-resolution edge correction for better image quality

BrightPixelTM Graphics Engine
• Up to 250 MHz engine clock
• Based on hierachical pixel tiling for rasterization
• Four independent pixel pipelines (512-bit wide)
• 4 pixels/clock with 8 textures/pixel
• Texture size up to 4Kx4K and non-power of 2
• Bilinear, tri-linear and anisotropic texture filtering
• Full anti-aliasing support for texts, lines, scenes
• Special hardware acceleration for silhouette anti-aliasing
• Special multi-resolution depth buffer
• Special bandwidth reduction hardware via compression
• Multilevel caches
• Fully OpenGL-compliant blending including fog & depth cueing

Video Engine
• State-of-the-art video de-interlacing (beyond Bob-and-Weave)
• Supports HDTV resolution up to 1920 x 1080
• TrueVideo® provides bilinear interpolation with proprietary
edge recovery scaling
• Dual apertures for simultaneous access to graphics and video
display memory areas

DVD Support
• Microsoft's DirectX Video Acceleration (DXVA)
• Includes both Motion Compensation and IDCT hardware
• Real-time playback (30 fps) of 9.8 Mbps MPEG-2 video bitstream
with 75% CPU headroom for other applications

Software
• Windows® XP, 98, ME, 2000, NT 4.0 and 5.0
• DirectX 9.0, 8.1 and 7.0
• DirectX Video Acceleration (DXVA)
• DirectShow 3.0 and 4.0
• OpenGL ICD 1.2 and 1.3
• Linux

XP4 T2 Specifications

Overview
• Performance 3D graphics solution for mainstream desktop
• 128-bit DDR memory bus with up to 64 Mbytes frame buffer
• CoolPowerTM management (3.0 watts maximum dissipation)
• Power supply: 1.2v, 1.5v, 2.5v, and 3.3v
• 30 million transistors in 0.13um CMOS process
• Standard 612 BGA package (31mm x 31mm)

Performance
• 250 MHz graphics engine clock
• Pixel fill rate: 1 billion pixels/sec (peak)
• Texel processing rate: 8 billion texels/sec
• Up to 500 MHz DDR memory clock
• Up to 8.0 GBytes/sec memory bandwidth (peak)

SmartTileTM Memory Bus Architecture
• 128/64-bit Double Data Rate (DDR) memory at up to 500 MHz
• Up to 64 Mbytes of local frame buffer
• Supports 8Mx32, 4Mx32, 2Mx32, 8Mx16, 4Mx16
• Supports hierachical pixel tiling

BrightPixelTM Graphics Engine
• Up to 250 MHz engine clock
• Based on hierachical pixel tiling for rasterization
• Four independent pixel pipelines (512-bit wide)
• 4 pixels/clock with 8 textures/pixel
• Texture size up to 4Kx4K and non-power of 2
• Bilinear, tri-linear and anisotropic texture filtering
• Full anti-aliasing support for texts, lines, scenes
• Special hardware acceleration for silhouette anti-aliasing
• Special multi-resolution depth buffer
• Special bandwidth reduction hardware via compression
• Multilevel caches
• Fully OpenGL-compliant blending including fog & depth cueing

Video Engine
• State-of-the-art video de-interlacing (beyond Bob-and-Weave)
• Supports HDTV resolution up to 1920 x 1080
• TrueVideo® provides bilinear interpolation with proprietary
edge recovery scaling
• Dual apertures for simultaneous access to graphics and video
display memory areas

Software
• Windows® XP, 98, ME, 2000, NT 4.0 and 5.0
• DirectX 9.0, 8.1 and 7.0
• DirectX Video Acceleration (DXVA)
• DirectShow 3.0 and 4.0
• OpenGL ICD 1.2 and 1.3
• Linux

XP4 T3 Specifications

Overview
• Performance 3D graphics solution for mainstream high-end desktop
• 128-bit DDR memory bus with up to 256 Mbytes frame buffer
• CoolPowerTM management (4.0 watts maximum dissipation)
• Power supply: 1.2v, 1.5v, 2.5v, and 3.3v
• 30 million transistors in 0.13um CMOS process
• Standard 612 BGA package (31mm x 31mm)

Performance
• Up to 300 MHz graphics engine clock
• Pixel fill rate: 1.2 billion pixels/sec (peak)
• Texel processing rate: 9.6 billion texels/sec
• Up to 700 MHz DDR memory clock
• Up to 11.2 GBytes/sec memory bandwidth (peak)

SmartTileTM Memory Bus Architecture
• 128/64-bit Double Data Rate (DDR) memory at up to 700 MHz
• Up to 256 Mbytes of local frame buffer
• Supports 8Mx32, 4Mx32, 2Mx32, 16Mx16, 8Mx16, 4Mx16
• Supports hierachical pixel tiling

BrightPixelTM Graphics Engine
• Up to 300 MHz engine clock
• Based on hierachical pixel tiling for rasterization
• Four independent pixel pipelines (512-bit wide)
• 4 pixels/clock with 8 textures/pixel
• Texture size up to 4Kx4K and non-power of 2
• Bilinear, tri-linear and anisotropic texture filtering
• Full anti-aliasing support for texts, lines, scenes
• Special hardware acceleration for silhouette anti-aliasing
• Special multi-resolution depth buffer
• Special bandwidth reduction hardware via compression
• Multilevel caches
• Fully OpenGL-compliant blending including fog & depth cueing

Software
• Windows® XP, 98, ME, 2000, NT 4.0 and 5.0
• DirectX 9.0, 8.1 and 7.0
• DirectX Video Acceleration (DXVA)
• DirectShow 3.0 and 4.0
• OpenGL ICD 1.2 and 1.3
• Linux

Unlike SiS it seems Trident have decided not to opt for AGP 8x implementation preferring to stick with the tried and tested 4x interface. With Trident claiming they're now on a 6 month refresh cycle I really don't see this as being a problem provided they can implement AGP 8x next year without too much of a redesign. They've also decided to use a rather limiting four rendering pipelines each capable of two textures in a single pass and this fact alone is proof that Trident aren't gunning to nudge off any performance crowns, not yet at least. It does however give it a distinct theoretical advantage over ATi's Radeon 9000. An interesting point is the quoted 11.2 GB/sec memory bandwidth (peak) for the XP4 T3. This stacks up very well against the NVIDIA Ti4600's 10.4GB/sec, on paper at least.

Maybe the biggest question is just how Trident have managed to build a DirectX 8.1 part on a 30 million transistor die and Trident's answer is that they've used some clever maths and optimised algorithms to drastically boost the efficiency of the four pipelines which by the way are non-uniform, each being around half the size of the adjacent one running 15 million transistors for the first, 7.5 million for the second, 3.8 million for the third and 2 million for the fourth pipeline. Is this marketing speak that translates as "we do in software what others do in hardware"? No not necessarily but I'd be very interested to know just what features are reliant solely on drivers and which are actually hardware accelerated.

I also don't believe the XP4 features an independent hardwired TCL unit, relying instead on its hardware Vertex and Pixel Shaders to do the donkey work. If these are efficient enough we shouldn't see too much of a problem with this assuming the bandwidth is cleverly managed.

So what about the decision to use a tile based rendering approach? Well, Kyro proved that it can be done and it can be done well but those who may remember the PowerVR chipset may also remember just how bad an idea this can be without very specifically coded drivers. Tile based rendering is certainly a great bandwidth saver and I don't think it's any secret that both NVIDIA and ATi have heavily researched the area. Combine this with other bandwidth trimming techniques such as vertex caches, texture compression, Z-compression and occlusion culling, all of which Trident say are implemented in their silicon, and you can perhaps begin to see why they felt they could live with using a rather conservative 128-bit memory bus (64 bit on the T1).

According to Trident their FSAA will comprise a variety of methods including super-sampling and multi-sampling though precisely which is used when remains a mystery. There's no mention of Anistropic filtering techniques so that's one to save for the hardware reviews. We also understand that displacement maps and shadows are supported in hardware along with hardware acceleration for silhouette anti-aliasing.

Surely one of the big pluses for the XP4 will be its frugal power requirements which stem directly from its low transistor count. The T3 is quoted at only 4watts maximum dissipation which is very impressive and presumably means the card will likely ship with only passive cooling (heat sink and no fan). Trident may have a very strong product for notebook users if nothing else, but I'm sure SiS are saying the very same thing about their Xabre GPU.

Here's a very rough and ready look at how the XP4 compares to some of its combatants ~

	Trident XP4 T3	ATI Radeon 8500	SiS Xabre 400	GeForce4 Ti4600
Process	0.13 micron	0.15 micron	0.15 micron	0.15 micron
Core clock	Up to 300MHz	250MHz	250MHz	300MHz
Memory clock speed.	Up to 350MHz	250MHz	250MHz	325MHz
Memory Bus width	128-bit - DDR	128-bit -DDR	64 / 128-bit -DDR	128-bit -DDR
Memory Type	Up to 256MB DDR-SDRAM	128MB DDR-SDRAM	Up to 128MB DDR-SDRAM	128MB DDR-SDRAM
Transistors	30 million	60 million	?	63 million
Memory Bandwidth.	11.2GB/sec (peak)	8.8GB/sec	8.0GB/sec	10.4GB/sec
Rendering Pipelines.	4	4	4	4
Texture Units Per Pipeline.	2	2	2	2
Theoretical Fill Rate.	1,200Mpixels/sec	1,100Mpixels/sec	.	1,200Mpixels/sec
AGP	4x	4x	4x/8x	4x

Conclusion :

So what can we expect for the XP4? Well with so little information available it's hard to say but there's certainly an opening at the budget end of the graphics market of a competent DirectX 8.1 accelerator. Provided the XP4 lives up to its claims I don't see either ATi's 9000 or NVIDIA's MX series offering a serious threat in terms of the technology on offer. Where the XP4 may loose out is to NVIDIA's Ti4200, SiS' Xabre400 and ATi's Radeon 8500 (while it remains available). Not necessarily because they're either faster or better, though they could very well be both, but because they're known quantities with a head start, mature or maturing drivers and a good marketing team. Trident don't just need to demonstrate good performance, they also need to provide solid drivers and quality support if they're to gain the trust of a very fickle and increasingly knowledgeable public. Can they do it? Come the end of October we can perhaps begin to answer that question but until then you'll have to watch for the usual assortment of leaked benchmarks and draw your own conclusions.

Home