<!–#set var="article_header" value="Attacking the GeForce2 MX:
ATi’s Radeon 32MB SDRAM” –>
Delays Are Bad for the Business
Graphics cards that cost $300+ are not attractive for everyone. The main business for the end user still takes place in the price range between 150 and 200 dollars. She/he wants the best bang for the buck. ATi now offers two new versions of its Radeon series that differentiate in price from the expensive high-end card Radeon 64MB DDR. The recommended retail price for the latter is $329. Sporting with of graphics memory the ‘smaller’ DDR version starts at $200 and the SDR version at $150. ATi positions the DDR card for the retail market, while the SDR version comes in a white box and is officially positioned for OEMs. You will find however that most dealers will also offer the SDR card on retail shelves.
For $150 we expect at least the performance that GeForce2 MX products are able to offer. You should note that MX cards are now available for only $100. A fellow competitor in this price segment is also the 3dfx Voodoo4 4500. This one has however, in comparison to the GeForce2 MX, a worse performance. ATi almost missed to jump on the bandwagon because the GeForce2 MX and the 3dfx cards have been available for two months now. It is somewhat incomprehensible how ATi sets its priorities. The Radeon 32MB is nothing more than a light version of an existing product! With the Radeon SDR coming rather late to the market, the company might lose important revenues.
Competitors and Differences to High-end Products
When ATi introduced its new flagship Radeon 64MB DDR in July, they pulled some jokers out of their sleeves that came unexpected. With a new technology, which ATi calls ‘HyperZ’, the Radeon chip was not quite as exposed to the usual memory bandwidth problems as NVIDIA’s GeForce2 GTS. At true color (32-bit color depth) and high resolutions the new ATi chip was even able to overtake the GeForce2 GTS from NVIDIA. However, NVIDIA reacted quickly with its so-called Detonator3 drivers. Those enabled a more effective memory management similar to ATi’s HyperZ technology – the only difference is its software management instead of the hard-wired design by ATi. The new drivers helped NVIDIA gain the performance crown again. Meanwhile NVIDIA has upgraded its product line with the GeForce2 Pro and the GeForce2 Ultra. Both graphics chips have a higher performance than the GeForce2 GTS.
Serving the enthusiast is one source of revenue, but NVIDIA entered the mainstream market with a modified version of its GeForce2 GTS. The company lowered the chip clock from 200MHz to 175MHz, reduced the number of rendering pipelines from four to two and baptized its new baby ‘GeForce2 MX’. In addition to that NVIDIA recommended cheaper SDRAM instead of DDR modules to meet the price expectations. ATi goes a similar way with its Radeon SDR. No architectural modifications were made to the chip but ATi reduced the clock for its 32MB version from 183 to 166MHz. As a customer you can now choose between the retail version with DDR SDRAM (DDR = double data rate) or the white-boxed SDRAM version with single data rate.
In the picture below you can see the SDRAM module from Micron found on the Radeon 32MB SDR. At the lower right corner you can find the ‘- 6’ marking which represents the access time of 6.0 nanoseconds.
The Radeon 32MB DDR uses memory from Samsung, also with access times of 6.0 nanoseconds. Only the Radeon 64MB DDR has faster 5.5ns modules from Hyundai.
The True Chip Clock – Marketing Versus Reality
ATi has a long tradition concerning the clock rates of older chips such as the Rage Pro, Rage 128 and Rage 128 Pro. Whether marketing papers or press releases, you always found higher clock rates in the specifications compared to what the actual products revealed. With Radeon SDR ATi seems to have picked up this habit once more. ATI announced a chip clock of 166MHz for the 32 MB DDR as well as the SDR cards. This specification is only valid for the Radeon 32MB DDR, but not for the SDR boards our Californian and German lab received. Here we could observe the following:
Instead of 166MHz clock we measure only 159.8MHz on both SDR test samples.
Image Quality
For this test ATi provided us with the driver version 4.12.3063 as you can see on the screenshot.
In Evolva we observed small artifacts when playing the game. The image quality became worse in the new benchmark Mercedes Benz Truck Racing (MBTR). In the latter the image quality was unbearable, as you can see in this picture.
In reality the deformations (see windshield and grille) look even worse. When we captured this screenshots the scene play stopped for a fraction of a second, which ‘improved’ the image quality to the state shown above.
Furthermore we cannot report anything positive about the state of the Windows 2000 drivers. Compared with Windows 98 the performance was so bad that we decided against publishing any results under Windows 2000.
The driver programmers at ATi have indeed a lot of homework to do. Although the Radeon 64MB has already been available since July we cannot say that the drivers have matured.
Driver Features
Since the introduction of the Radeon 64MB DDR little has changed.
Here you can make adjustments for Direct3D games. After the installation the setting for vertical sync is enabled, which leads to a better image quality. In this screenshot you can see that we disabled the vertical sync. Only under these settings you can reach optimal performance in games. As far as anti-aliasing is concerned we recommend leaving the switch disabled. Enabling anti-aliasing results in a drastic performance drop. Please note that this recommendation only applies to the type of games we prefer. You can try it out for yourself if anti-aliasing is an option for you. Edges will indeed get smoother but you have to decide if you can live with much lower frame rates.
The OpenGL tab offers more functions. You can find the most important buttons ‘quality’ and ‘performance’ at the upper left corner, which are pretty much self-explanatory.
Driver Features, Continued
Under ‘Advanced’ you won’t find much on the 32MB SDR. Only Radeon cards with TV-out offer more options here.
For those of you who don’t want to fiddle on a complicated OSD at your monitor, you can also make changes to the screen position and the screen size with the driver software.
Brightness and colors can be changed in the ‘color’ tab. However, it is not possible to make individual changes for OpenGL and Direct3D apps.
Test Setup
In the following benchmarks we compare the two 32-meg versions of the Radeon with the GeForce2 MX. All these graphics cards cost between $100 and $200. Additionally we publish the results of the high-end Radeon 64MB DDR and GeForce2 GTS based boards to give you an idea how much performance boost they offer for the higher price.
Since 2D performance does not show any noticeable difference anymore, we focused to test all contenders with Quake 3 Arena, MDK2, Evolva and Mercedes Benz Truck Racing (MBTR).
We chose a minimum resolution of 1024×768 to stress the 3D pipeline of the graphics cards and not the chipset of the platform. We then increased the resolution up to 1600×1200 pixels. At this point we want to repeat what the following average frame rates (fps) actually mean:
<ul
Graphics Cards and Clock Rates | |
GeForce2 GTS | Asus AGP-V7700 TVR 32MB (200/333MHz) |
GeForce2 MX | Asus AGP-V7100 Pure 32MB (175/166MHz) |
ATI cards | Radeon 32MB SDR (160/160MHz) Radeon 32MB DDR (165/333MHz) Radeon 64MB DDR (183/366MHz) |
Platform | |
CPU | Intel Pentium III 1GHz |
Motherboard | Asus CUSL2 (bios 1003 BETA 003) |
Memory | Crucial PC133 CAS2, setting 2-2-2-5/7 |
Network | 3Com Etherlink 3C905-TX |
Software, Drivers and Settings | |
OS version | Windows 98 SE 4.10.2222 A |
DirectX version | 7.0 |
GeForce2 GTS/MX | 4.12.01.0631 |
Radeon 32/64 | 4.12.3063 |
Quake 3 Arena | Retail version command line = +set cd_nocd 1 +set s_initsound 0 |
Evolva | Rolling Demo v1.2 Build 944 Standard command line = -benchmark Bump Mapped command line = -benchmark -dotbump |
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 |
Benchmark Results: Quake 3 Arena
At ‘NORMAL’ settings, i.e. 16-bit High Color, all Radeon cards have weaknesses compared to the GeForce2 chips, which is due to the 3D engine. It is remarkable that the high-end version of the Radeon with 64MB memory (see yellow bar) only reaches frames rates that already its mainstream competitor GeForce2 MX from NVIDIA is able to produce.
In games the trend goes to 32 bits of color depth. HIGH QUALITY slows down the GeForce2 MX due to its bad memory bandwidth. Comparing the Radeon 32MB SDR to 16-bit the performance delta is less than with the MX. Therefore, the Radeon is slightly ahead of the MX.
Benchmark Results: MDK2
A similar picture reveals MDK2. At 16 bits the GeForce2 MX overtakes the Radeon SDR.
MDK2 doesn’t seem to stress memory bandwidth as much as Quake 3. The GeForce2 MX and the Radeon 32 SDR are on par in this case. It can also be seen that neither the Radeon 32 SDR nor the GeForce2 MX is suitable for resolutions of 1152×864 and higher as frame rates drop below 45 fps. Only the GeForce2 GTS manages to achieve acceptable results – at least up to 1280×1024…
Benchmark Results: Evolva Bump Mapped
If you turn on Bump Mapping in Evolva the complexity for any 3D chip is increased. Nevertheless, all contenders achieve respectable results up to 1280×1024. The GeForce2 GTS is a bit of a overkill for this game when played at 16 bit.
At more demanding 32 bits the MX is definitely the slowest in this benchmark but the lead of Radeon 32 SDR is only small.
Benchmark Results: Mercedes Benz Truck Racing
The new addition to our benchmark suite MBTR demands a lot of rendering power of all 3D chips tested. Contrary to all other results the Radeon chips seem to get a lot of ground if you compare them the GeForce2 chips. The main problem is still unsolved by ATi – the horrendous image quality. All ATi chips drop a lot of frames and show unbearable artifacts.
We already mentioned it in the paragraph concerning the image quality. The current ATi drivers have to be improved! We cannot take those results into our performance evaluation, as the scenes are not rendered properly. NVIDIA provides matured drivers that produce a perfect image quality in this benchmark.
Overclocking
Many of our readers probably know that Radeon 64MB DDR can be overclocked from 183 to 200MHz. Unfortunately, we had to realize that 32MB SDR board practically offers no headroom at all. When we increased the clock only 5MHz above its default setting the card became unstable.
Conclusion
The benchmarks show that the GeForce2 MX is still the performance leader at high color (16-bit). The Radeon 32 SDR flexes its muscles at 32-bit true color, where it is practically on par with the GeForce2 MX in many 3D applications. As many gamers prefer 32-bit color now, ATi offers indeed a competitive product for the mainstream market as far as performance is concerned. However, Radeon SDR is e.g. missing nifty features such as NVIDIA’s ‘Twin-View. It should also be noted that neither the GeForce2 MX nor the Radeon SDR are suitable for 3D-games at resolutions of higher than 1024×768 pixels. This still remains the playground for high-end graphics cards such as the GeForce2 GTS.
Even if we can confirm that ATi can now provide the counterpart to the GeForce2 MX, the manufacturer has a problem – the price. $150 for the SDR board and $200 for Radeon 32MB DDR is too much for a card of that caliber from our point of view. As we mentioned above, many vendors already offer GeForce2 MX cards at $100.
This little roundup reveals another issue. All results of our initial tests are now overruled by the new NVIDIA Detonator3 drivers. These boost the performance of GeForce2 GTS chips to levels where the Radeon 64MB DDR is left in the dust. We can only hope that ATi reconsiders its pricing towards a more favorable point for all of its Radeon cards.