Asus A7N266-C in the test: The nForce 415-D now also without graphics
- 1 Foreword
- The chipset in detail
- 2 Graphics included
- 3 MCP, MCP-D?
- 4 Scope of delivery and documentation
- 5 Features
- 6 Stability
- 7 Test system
- 12 Conclusion A7N266-C
If the key data of the stand-alone graphics card and its 'embedded' brother with a clock rate of 175MHz and the internally identical structure are still the same, opinions ultimately differ with regard to the memory bandwidth. While the GeForce2 MX available on graphics cards can draw a maximum of 2.7GB/s from the built-in SDRAM at a clock rate of 166MHz, this point has to be considered separately for the individual IGP versions. While the TwinBank architecture of the nForce 420 with its 4.2GB/s seems the ideal partner for the GPU at first glance, it must be taken into account that all PC components have to share the available resources. The CPU alone uses up to 2.1GB/s at peak times, so we estimate the average bandwidth kept available for the MX at a little under 2GB/s. What sounds rather sobering compared to the 2.7GB/s, thanks to 'TwinBank' is still way ahead of the solutions of the competition or the IGP-64 of the nForce220. Because the CPU, graphics and co. Together have to draw from only 2.1GB/s - a bottleneck. In terms of graphics memory, the IGP-128 and IGP-64 are similar. The current mainboards reserve either 32 or even 64MB for the GPU. If data has to be reloaded via the AGP bus, the nForce can score again against the GF2 MX card. Because instead of the usual 66MHz, the internal bus of the IGP clocks with 100MHz. Upgraders or owners of powerful graphics cards have also been thought of and the connection of an external AGP 4x port has been made possible.
Another feature that has just started with AMD with the Palomino core under the name 'Data Prefetch Logic' Entering the Athlon/Duron, nVidia has again integrated it into all three Northbridge versions (IGP-128, IGP-64 and SPP).
Dynamic Adaptive Speculative Ultimately, the DASP only has one task: It should use routines that are as clever as possible to predict which data could soon be queried from the memory and which have already been loaded, even before they are needed. If the request then comes, the information can be sent without the slightest loading delay. nVidia claims to have patented the installation of this unit in the northbridge, which would force the competition to an expensive licensing.
The most important properties of the northbridge would be dealt with. For the sake of a better overview, the following table lists the nForce again in comparison to the current class leader KT266A.
On the next page: MCP, MCP-D?