ATi Releases Mobility RADEON
The mobile market for notebooks is on the rise. 25% of corporate computers are notebooks and the numbers continue to rise. While NVIDIA gained a giant market share in the desktop arena and created a very strong brand name in this market, the #1 provider for mobile graphic solutions remains ATi.
ATi entered the mobile business in 1998 with an AGP graphics solution. A year later, the first ATi chip with embedded SDRAM followed. The OEM strength in the chip business provided ATi the inside look for the needs of the notebook makers, which are good driver support, same footprint for different chips to serve different segments of the market, low power consumption and no need for external memory.
NVIDIA is expanding to every segment of the graphics business. It covers professional graphics with the Quadro2 Pro, hardcore gaming with the Geforce2 GTS/Ultra series and the value segment with the Geforce2 MX and the Vanta. The only two areas that are growing faster than the desktop market are the console market e.g. the Playstation2 and the mobile computer market. Those two areas are now the targets of NVIDIA’s expansion plans. Is NVIDIA strong enough for all those businesses? Twice a year NVIDIA releases a further advanced desktop chip. Microsoft is pushing hard for the X-Box. The professional graphics sector requires plenty of R&D resources and now there is the mobile market with completely different requirements in driver and ODM/OEM support. The drivers need more power management features and stability. The support for the chip customers must change too. You can’t just take a reference board and plug it into a notebook. Only time will tell if NVIDIA is able to handle all those many things at the same time.
The current market situation looks like this.
Mercury Research – Portable Graphics Chip Market Share (Q3 00)
NVIDIA threatens ATi’s position in the high-end market, while the value segment with the volume and cost pressure is moving to the integrated graphics solutions. VIA, ALi, SiS and Intel offer chipsets with embedded graphics. From the desktop graphics performance point of view, the speed is average at best. Some notebook users would kill for desktop 3D-performance. The only place where you can’t upgrade your VGA solution offers the lowest 3D performance. Powerful 3D eats up wattage and generates a lot of heat. Different requirements need different approaches. NVIDIA and ATi found a way to get more 3D power into a notebook.
Waiting For Arrival …
When a new chip is launched, you’d expect a product review on the day of its release and general availability a few weeks later. We usually receive test samples a considerable time before the release date of the actual product. NVIDIA announced its Geforce2 Go on November 13th, 2000. Still nobody has been able to get a notebook with that chip. Now ATI is striking back with its Mobility RADEON and there isn’t any notebook around for testing either.
Unlike desktop components like processors, motherboards or graphics cards, it takes a lot longer to release a new notebook. A motherboard design for a new chipset takes 3 to 6 months. If the bus structure changes, as it did with the Pentium4, it can take up to 9 months. In the desktop graphics card business, the card manufacturer uses the reference design of the chipmaker to hit the time to market momentum. A few months later the graphics card maker will introduce its own design with some value-added features or a cost-down solution.
The design of a notebook can take 12 months and is much more complex than a motherboard or a graphics card design. In fact, it is almost like composing an opera. Long battery time, small size, lowest possible weight and a cooling solution that removes the heat produced by CPU, Chipset, Graphics and Memory must fit into a new and nice customized case.
Both the Geforce2 Go and the Mobility RADEON are supposed to be available within the next month. Why not now? The answer is Intel’s roadmap. Soon there will be new notebook chipsets available with new CPUs. As it takes such a long time to design a new notebook, the manufacturer can wait a few months with the presentations of its next-generation product that includes both the new graphics chip as well as the new chipset. Furthermore, notebook ads shows the graphics chip listed below CPU, memory size, display size and hard disk drive (HDD), as it’s one of the few things that can’t be changed easily.
The two new mobile graphics chips are based on desktop graphics chip solutions. Some features were added while the clock and voltage were reduced. Lets take a look at the performance we expect from the upcoming 3D chips by simulating the new chips.
RADEON Gets Mobile
ATi’s last mobile graphics chip, the ‘Rage Mobility’ was based on the desktop chip ‘Rage 128’, so it’s not surprising that the new ‘Mobility RADEON’ has its roots in ATi’s latest desktop chip, the RADEON. However to reduce silicon real estate and power requirements, Mobility RADEON represents a significantly stripped-down version of the RADEON chip we know. While RADEON comes with two rendering pipelines, each including three texturing units, Mobility RADEON has only got one of them, cutting the possible fill rate in half. The transform and lighting unit of RADEON, ATI calls it ‘Charisma Engine’, is not present in Mobility RADEON either, thus leaving NVIDIA’s upcoming GeForce2Go as the only mobile graphics chip with integrated T&L. The core and memory clock of Mobility RADEON can go up to 200 MHz, thus reaching the same clock rates at RADEON. However, the notebook maker will decide which core/memory clock will meet the power consumption and thermal requirements of its notebook and it is very unlikely that any of the upcoming notebooks with Mobility RADEON will run that chip at a clock as fast as 200 MHz.
Other important features of RADEON remained in Mobility RADEON. ATi’s new mobile chip will include the ‘HyperZ’ technology, which is supposed to reduce Z-buffer bandwidth requirements, and the iDCT and motion compensation features for MPEG2/DVD playback will be found in Mobility Radeon as well. ATi added a dual monitor solution called ‘HydraVision’, competing with GeForce2Go’s ‘TwinView’. Since Intel’s introduction of SpeedStep, a power saving solution that reduces clock and voltage of Intel’s mobile processors once the notebook is battery-operated, a similar technology made its way into the design of mobile graphics chips too. Mobility RADEON will come with a power saving feature that will reduce clock and voltage when the notebook is not hooked to an AC-power outlet.
Mobility Radeon will come with integrated graphics memory of 8 MB or 16 MB in size, which is connected to the core via a 128-bit wide memory interface. The external memory that can be added to Mobility RADEON to sum up to up the huge size of 64 MB of video memory, but it interacts with the graphics core over an only 64-bit wide interface.
Comparing Mobility RADEON to NVIDIA’s GeForce2Go shows its strengths and weaknesses. GeForce2Go comes with integrated T&L and offers a higher pixel fill rate as well as texel fill rate. However, Mobility RADEON is able to do without external video memory, it improves MPEG2/DVD decoding with its integrated iDCT and it is supposed to have significantly lower power requirements than GeForce2Go, although ATi wouldn’t supply us with any actual numbers.
RADEON Gets Mobile, Continued
Here is a comparison between the two new competitors, listing their most important features:
Features | Mobility RADEON | Geforce2 Go |
Core clock | Up to 200 MHz | Up to 143 MHz |
Rendering Pipeline | 1 | 2 |
Texturing Units Per Pipeline | 3 | 2 |
Maximal Possible Pixel Fill Rate (highly depending on core clock) |
200 Mpixel/s | 386 Mpixel/s |
Maximal Possible Texel Fill Rate (highly depending on core clock) |
600 Mtexel/s | 770 Mtexel/s |
Integrated T&L Unit | no | yes |
Video Memory | 8/16 MB internal DDR SDRAM, up to 48 MB external SDR/DDR SDRAM |
No internal memory, Up to 32 MB external SDR/DDR SDRAM |
Width Of Internal Memory Interface | 128-bit | N/a |
Width Of External Memory Interface | 32/64-bit | 64/128-bit |
Memory Types Supported | SDR-SDRAM / DDR-SDRAM | SDR-SDRAM / DDR-SDRAM |
Maximal Possible Memory Bandwidth | 6.4 GB/s internal / 3.2 GB/s external | 5.4 GB/s |
Memory clock | Up to 200MHz | Up to 166MHz |
MPEG2/DVD Decoding Acceleration | Integrated iDCT, Motion Compensation | Motion Compensation |
external TV Tuner | yes | no |
Multiple Displays | HydraVision | Twinview |
How would I implement Mobility RADEON in my theoretical notebook? How fast is fast enough for my notebook?
- Target System is a 2 lbs mobile computer. This weight includes the battery, so the battery will be small. The battery run down time must be acceptable too. In such a case I would use a Mobility RADEON with integrated 8 MB at 66 MHz/1.5V. There is space for external memory and no expensive cooling for the graphic chip.
- This time I want to build a desktop replacement system. The large flat panel screen found in this kind of notebook has the benefit of providing a large notebook base with plenty of space. It will contain a high speed CPU and a nice, big, fast and fat graphics solution e.g. 32 MB video memory. Depending on the space, I would implement the chip with 16 MB of integrated memory at 200MHz/1.8V and two 2 Mbit/32-bit DDR chips next to it. If space is no limit, I might add four additional 2Mbit/32-bit DDR chips on the other side of the notebook motherboard and have a real 64 MB racing machine.
- For the value segment I’d suggest the Mobility RADEON chip with 16 MB of integrated memory at 166MHz/1.7V. An 8 MB version with optional 8 MB of external memory would be nice too, depending on the chip footprint.
- If thermal problems arise, just lower the voltage or the clock speed.
Simulating Mobility RADEON
We already mentioned it above that ATi is currently not supplying any test samples of Mobility RADEON. The Canadian company is even making a huge secret about the notebook OEM that will introduce the new chip. However, we are in the lucky situation to know that Mobility RADEON is actually derived from ATi’s new low-cost desktop chip RADEON VE, giving us an excellent chance to simulate it.
ATi’s Mobility RADEON chip is based on RADEON VE. For our simulation of the Mobility RADEON we used an ATi engineering sample RADEON VE card with 32MB DDR memory and altered the memory as well as core clock. Our testing is most likely going to show close to best-case results because of the 32 MB video memory on the test card. However, the chip and memory were not clocked at the theoretical maximum of 200 MHz, which on the other hand might not be found in any future notebook using Mobility RADEON anyway.
The Family Bands Of GeForce2Go
NVIDIA’s low-cost Geforce2 MX is the ‘big brother’ of the Geforce2 Go chip. The memory interface of GeForce2 Go can be 32/64/128-bit wide and support SDRAM or DDR-SDRAM memory. There will be a big range from 8MB/32-bit SDRAM to 32MB/128-bit SDRAM solutions depending on the notebook maker. We already mentioned the strengths of GeForce2Go, which are fill rate and integrated T&L. This will most likely make this chip faster than Mobility RADEON in 3D-applications.
With the information I have at this time I would design my Geforce2 Go solution like this:
- Most graphics cards use 2Mbit/32-bit memory chips. This enables the following memory configurations: 8MB/32-bit, 16MB/64-bit and 32MB/128-bit. Other designs might be possible too. The 32MB version will be a challenge to implement with 128-bit. The preferred solution will be a 32MB/64bit DDR SDRAM.
- Both the core and the memory clock should be adjustable depending on the power status. A high clock setting would make sense when the notebook is using AC-power and a low clock setting when it’s battery powered.
- The thermal problem can be tackled best by reducing the voltage. Every overclocker knows however, that higher clocks require higher voltages. Depending on the thermal envelope available in the notebook design the chip might have to run at a lower clock than theoretically possible.
- This graphics solution will require a lot of real estate inside the notebook and consume more power then NVIDIA is willing to admit. You need space not only for the graphics chip, but also for the memory nearby, and believe it or not, both will consume power. The power consumption of the memory chip(s) has to be added to the power consumption of GeForce2 Go in order to make it comparable to an ATi chip with integrated memory. 8MB will add up to 0.25W to NVIDIA’s power calculation.
Toshiba is NVIDIA’s launch partner, using GeForce2 Go with a 16MB 64bit DDR-SDRAM solution, running at 143MHz core clock and 166MHz memory clock. These Toshiba notebooks are still unavailable, so we were forced to simulate a GeForce2 Go as well. We used a 32 MB ASUS V7100 Geforce2 MX card and clocked it down to the above-mentioned frequencies of GeForce2 Go. This simulation is somewhat of a best-case scenario for GeForce2 Go, because it is using 32 MB of graphics memory. The peak memory bandwidth of the simulation is identical to the upcoming Toshiba solution. While the Asus V7100 card is using SDR-SDRAM at a 128-bit wide interface, the Toshiba notebook is supposed to run DDR-SDRAM on a 64-bit interface.
An Overview Of Current Mobile Graphics Chips
The selection of mobile ATI solutions is huge. I already cut out the old ATI chips.
Silicon Motion | S3 | Nvidia | |||
Features | LynxEM4 (SM710) | LynxEM+ (SM712) | Lynx3DM (SM721) | Savage/MX | Geforce2 Go |
Power / worst case | 0.55W | 0.55W | 1.1W | 1.9W | 2.8 W |
max. memory | 4MB | 4MB | 16MB | 16MB | 32MB DDR/SDR |
built-in memory | 4MB | 4MB | 8MB | – | – |
Dual View | X | X | X | X | X |
DVD support | – | – | X | X | X |
3D support | – | – | X | X | X |
Trident | ||||||
Features | Cyber 9525 DVD | CyberBlade XP | Blade T16 | Blade T64 | Blade XP | Blade XP Turbo |
Power / worst case | 2.3W | 1.8W | 2.0W | 2.3W | 2.6W | |
max. memory | 2.5MB | 32MB | 16MB | 32MB | 32MB | 32MB |
built-in memory | 2.5MB | – | – | – | – | – |
Dual View | X | X | X | X | X | X |
DVD support | X | X | X | X | X | X |
3D support | X | X | X | X | X | X |
ATI | |||||||
Features | Rage Mobility M | Rage Mobility M1 | Rage Mobility 128 | Rage Mobility M4 (AGP 4x) | Rage Mobility M4 (AGP 4x) | Mobility RADEON | Mobility RADEON |
Power / worst case | 1.3W (w/8MB) | 1.3W (w/8MB) | 2.2W (w/8MB) | 2.2W (w/8MB) | 2.2W (w/8MB) | 2W (w/8MB DDR) | 2.4W (w/16MB DDR) |
max. memory | 8MB | 8MB | 16MB | 16MB | 32MB | 64MB DDR/SDR | 64MB DDR/SDR |
built-in memory | 4MB | 8MB | 8MB | 8MB | 16MB | 8MB | 16MB |
Dual View | X | X | X | X | X | X | X |
DVD support | X | X | X | X | X | X | X |
3D support | X | X | X | X | X | X | X |
Test Setup
Chip | Core clock MHz |
Memory clock MHz | Memory Size | Memory Type |
Simulated Mobility RADEON | 183 | 183 | 32MB | DDR SDRAM |
Geforce2 MX | 175 | 166 | 32MB | SDRAM |
Simulated Geforce2 GO | 143 | 166 | 32MB | SDRAM |
Savage/MX | 100 | 100 | 8MB | SDRAM |
The S3 Savage/MX chip is one of the fastest graphic chips available and represents a shipping notebook solution. This chip is about 1 year old and will soon be replaced by ‘Twister’.
This is the graphics solution found inside the ASUS L8400B Notebook.
The complete Setup looks like this.
Desktop-System | |
CPU | Intel Pentium III 750MHz |
Motherboard | ASUS CUSL2 |
Memory | Workx 128MB CAS2 SDRAM |
HDD | IBM DTLA 307030 |
VGA ATI | ATI Radeon VE, 32MB DDR |
VGA Geforce2MX | ASUS AGP-V7100, 32MB SDRAM |
Notebook-System | |
CPU | Intel Pentium III 750MHz Speedstep |
Notebook | ASUS L8400B |
Memory | 192MB SDRAM |
HDD | IBM Travelstar DJSA-220 |
VGA S3 | Savage/MX, 8MB SDRAM |
Software | |
Operating System | Windows 98 SE 4.10.2222A |
DirectX Version | 7.0 |
Nvidia Driver | 6.47 |
ATI Driver | D7.20-01176-129C-ATI |
S3 Driver | 4.12.01-7009-7.13.08 |
Settings | |
Screen Resolution | 1024×768, 60Hz |
Graphics Benchmarks | |
MDK2 | Downloadable Demo Version T&L On, trilinear, high texture detail |
Mercedes Benz Truck Racing | Triple buffering, reflections, bright mode, hires textures, alpha textures, all filters |
Quake III Arena | Retail Version 1.11 command line = +set s_initsound 0 + set cd_nocd 1 |
I focused on 3D tests only. The refresh rate is set to 60Hz, which is typical for notebook flat panels. The BX chipset of the L8400B Notebook is very old. New notebooks designs will most likely utilise a modern chipset in combination with those graphic chips.
MDK2
Lets start with the OpenGL game MDK2.
The frame rate of Savage/MX is below 1frame/sec, probably due to driver issues or other problems. It takes 1.2 to 4 seconds to calculate a single frame by software. The simulated GeForce2 Go dominates in this benchmark, logically due to its higher fill rate.
With increased color depth the ATI chip drops to 89%, while the NVIDIA chip drops to 70% of the frame rate at 1024x768x32. What we see is not very surprising, since it is known that RADEON is able to get closer to GeForce2 once the memory bandwidth becomes tight.
Mercedes Benz Truck Racing
The representative for DirectX games is MBTR.
The simulated Mobility RADEON takes the lead, while the NVIDIA chips seem to have some problems with this game. It goes to show that Mobility RADEON doesn’t always have to come in behind GeForce2Go.
With only 8MB this Savage/MX solution cannot provide the memory needed for this game in such a high resolution. The simulated Mobility RADEON is once more in front of the simulated GeForce2Go.
Quake 3 Arena
The simulated GeForce2 Go is clearly able to outperform the other mobile graphics solutions.
With increased resolution at 16bit color, the simulated Mobility RADEON drops to 52% while the Geforce2 Go goes down to 66% of the performance at 640×480.
Our simulated Mobility Radeon behaves the same way as all the other Radeon-solutions we already know. The score in 32-bit color is identical to the score in 16-bit color. The simulated GeForce2 Go is still in the lead however.
Conclusion
Please keep in mind that the performance evaluation between Ati’s new Mobility RADEON and NVIDIA’s GeForce2Go is only of theoretical nature. The 3D-performance in actual notebooks can significantly differ from our best-case results, because it is up to the notebook maker to decide which core and memory clock rates he will use. The situation is very different to the desktop graphics cards, because in a notebook size and power requirements as well as the thermal situation matter more than pure 3D-performance.
The race has just begun. In the last years ATi had no performance competitors in the notebook sector. NVIDIA is pushing hard to conquer yet another market segment, using the strong performance of the Geforce2 Go as its weapon. However, things are different in the notebook arena. Pure 3D-power is not one of the most important things.
ATi’s new Mobility RADEON offers some real nice features for notebook makers, in addition to its respectable 3D-performance. ‘DualDie’ will save space inside of the notebook and combined with its low power consumption the ATI chip will probably get plenty design wins.
The performance results that we show are simulations and don’t reflect the shipping product. Notebook makers have many options for the implementation. The complete spectrum will be available in the market from low voltage, low speed with 8MB to high voltage, high speed and 64MB.
Without the T&L and with only 1 rendering pipeline, the ATi Mobility Radeon offers some good DirectX performance. The external TV Tuner is a nice value-added feature for users and unique in this market segment.
Will the NVIDIA brand be strong enough to gain a considerable market share in the mobile business? We don’t doubt that GeForce2 Go will be a successful chip. For smaller mobile graphics maker like Silicon Motion, S3 and Trident the world will look different in a year from now. ATi will lose some market share in the high end, but the market will increase too.
Will the implemented solution be as good as the theoretical solution? The performance of the two chips will be limited by notebook-typical thermal and power considerations. The chips will start a new era for notebook users. We are one step closer to a real desktop replacement.