The chipset
In TurboCache and HyperMemory technologies, the chipset has an important role to play. The graphic card uses it to access system memory. To be accurate, the graphic card has to go through two busses to access system memory, the PCI Express bus and then the memory system bus. For HyperMemory and TurboCache technologies to be efficient, you need to have enough bandwidth on the two busses.
We tested UT2004 and Far Cry with 3 different chipsets: the i915G, i945G and the nForce 4 Ultra:
The nForce4 is slightly more efficient, but the gap is so reduced that it isn’t really significant.
Of course, we used Dual Channel with all 3 platforms, as it’s very useful when the graphic card partly uses system memory. For that you need to use two memory modules instead of one with doubled capacity for all computers equipped with TurboCache or HyperMemory graphic cards.
2D
Is it still possible to have performance differences in 2D between current solutions? Yes, it is. For most cases it is insignificant, but we noticed a visible difference between graphic cards and Intel’s integrated cores. These cores have some difficulty in rendering some effect of Window’s interface. For example, the blending under menus isn’t flowing. Of course, it is only a small detail, but it would be best to completely deactivate blending to avoid this unpleasant feeling of lack of fluidity in 2D.
In their drivers, NVIDIA and ATI have a function, which gives the possibility of substituting them for flat monitor interpolation when connected in DVI. It’s interesting to take a look at when testing entry level products as chances are that they won’t be able to run games in standard resolution (usually 1280 x 1024). For ATI the option is purely decorative as it doesn’t have any effects (bug?). For NVIDIA the option is functional (but well hidden in the driver:
nView display-> Monitor parameters -> Monitor adjustments) and the result is conclusive.
Instead of sending a 800 x 600 image to the TFT monitor, which resizes it in 1280 x 1024, the GPU does the operation before sending it. The monitor receives an image with standard resolution and doesn’t have to resize it. What’s the point? Don’t we obtain the same result? Yes and no. Most TFT monitors (especially entry level ones) use poor quality interpolation algorithms whereas NVIDIA’s graphic cards use an excellent one. Of course, using the standard resolution is always the best thing to do, but when it’s not possible, this option is more than welcome! Icing on the cake, the performance incidence is negligible.
Unfortunately we won’t be able to show you the quality difference as it’s not possible to take a screenshot image once it has been interpolated, and photo results aren’t presentable.
Video
If MPEG-2 decompression is no longer a problem for GPUs, it isn’t necessarily the case for the last HD codec such as WMV. What are the performances of the different graphic cards and integrated cores in continuous WMV file reading? We conducted this test with a computer equipped with a P4 3.8 GHz and deactivated Hyper Threading to have a more representative CPU use rate. The rate reported with Hyper Threading and CPU wasn’t always reliable because of the presence of the second logic core.
In 720p, acceleration worked correctly with all NVIDIA graphic cards except for the 6200 TC 16 MB, which doesn’t support it. The Radeon X700 Pro and X700 SE also accelerate 720p reading, but the X300, X550 and X600 do not. In the end, all solutions can read 720p content without much difficulty.
It is a little bit more difficult with 1080p. The GeForce 6600 and 6200 provide nice results thanks to their WMV HD acceleration. It is important to note that NVIDIA indicates that TurboCache 32 and 64 MB only support acceleration in 720p, but it is also functional in 1080p. The GeForce PCX 5750 reads the content without much difficulty even if it doesn’t have any acceleration and that the CPU rate is very high. It is also very high with the TurboCache 16 MB. With this card, however, several frames are left out, which causes small cuts here and there.
For ATI only the X700 has hardware acceleration. No graphic card, however, play the video corectly. An undetermined problem makes video turn at 17 to 20 FPS instead of 24 FPS, making it slower than usual. We also found the problem with an nForce 4 Athlon 64 4000+ platform. With the X800, however, it disappears. We asked ATI and they told us that they were indeed able to reproduce the same problem, but only under certain still undetermined circumstances (problem due to the regionalisation or drivers, or Microsoft’s patch activating the acceleration? A lack of power of entry and middle level graphic cards?). Let’s hope that they will soon find a solution via new drivers. Anyway for the moment this doesn’t work.
The test
For this test we used different protocol. To find out if it was possible to play games with these graphic cards… we played with them! In addition to the usual benches, we played 19 games with all graphic cards to determine the maximum quality in comfortable play with a nice acceptable framerate.
For each game we selected a scene we felt to be representative. Games chosen cover a wide range and were released from 2003 to 2005. All don’t have the same needs as Doom 3 or Far Cry and use less evolved engines. We felt it was important not to only evaluate entry level graphic cards on their capacity to run the latest FPS.
In order to stick to a realistic entry level memory, we used 512 MB only with the test computer. Under those circumstances, loading periods were awfully long and disc swaps were annoying with recent games, becoming even worse with TurboCache / HyperMemory and integrated chipsets. To play with 1 GB is more than recommended!
Test configuration:
- Leadtek / Foxconn nForce 4 Ultra
- AMD Athlon 64 4000+
- 2x256 MB of PC-3200 memory
- ForceWare 71.89
- Catalyst 5.5
- ASUS P5GDC-V (i915G)
- Intel D945GTP (i945G)
- Intel Pentium 4 670 (3.8 GHz, 2 MB)
- 2x256 MB of PC2-4200 memory.
