The TurboCache Architecture
When nVidia brought the original 6200 core out, the simple fact was that this video processor was simply a 6600 core with four of the pixel pipelines disabled but the 6200 TurboCache (aka TC) is a totally different kettle of fish. The NV44 core is actually a full redesign in order to remove the additional memory channel that gives the 6200 its 128-bit memory interface. Also in order to reduce the price by reducing the die side, the extra four pipelines that aren't needed are removed to save on space and silicon usage.
Apart from the removed unused components, the TC uses the same memory management that Intel uses for its onboard graphics, that being Dynamic Memory Assignment. The TurboCache 16 and TurboCache 32 can have up to a maximum of 128MB frame buffer. This means that a 16MB model will use 112MB and the 32MB model will use 96MB system memory in order to make up the full 128MB - the system automatically assigns a total of 128MB for the video card. The 64MB model can have up to 256MB frame buffer, sucking 192MB of system memory. Like Intel, nVidia has a MMU or Memory Management Unit that interfaces with the Forceware drivers. When the system is running only Windows or any 2D application, only the onboard cache memory is used, leaving the system memory free for the CPU to use. When a 3D application is loaded, the MMU removes the extra memory from the system RAM in order to increase the framebuffer size, when the application finishes, the memory is released back to the system to use as usual.
Now we have seen this technology before used by nVidia with the nForce 2 onboard graphics, VIA with its S3 integrated graphics chipset as well as Intel on its Express graphics system, however, performance has been somewhat lax and this has been because of two major issues - Firstly the graphics core on the Intel and VIA systems has been extremely value based, the only real solution that showed promise was the nForce 2 with the GeForce 4MX controller built in, on graphics cards this chip produced reasonable graphics scores however the major problem is the memory bandwidth.
System memory is extremely slow compared to that of the video cards. Graphics cards use up to a 256-bit memory bus delivering bandwidths over 30GB/s, where as system memory even to this point using DDR-2 can only go as high as 8.5GB/s. Accessing this memory is also an extremely painful affair. AGP even at its full 8x capabilities can only handle 2.1GB/s, with some of the bandwidth being used to transfer data to the GPU processor, leaving around 1GB/s free when you count the overheads. This means that you have to push all this data to and from system memory using about 1GB/s and this is simply not going to give the greatest performance.
The TC uses the new PCI Express x16 link to transfer data to the GPU as well as access system memory. Using a full duplex 4GB/s (8GB/s total) you are looking at about 6GB/s of usable bandwidth, while this isn't as good and the 6200 with dedicated memory, it certainly does give it a bit more to play with when it comes to transferring data to and from the card and memory.
Find the lowest price on nVidia GeForce 6200 Graphics Cards!