NVIDIA's TurboCache cards are a part of the company's GeForce 6200 series, and use the NV44 GPU. Like the RV370, the NV44 has four pixel pipelines. However, unlike the ATI chip, which has a raster operator (ROP) attached to each of its four pixel pipelines, NV44 has only two ROPs, which limit it to writing only two pixels per clock. NVIDIA uses a similar arrangement and crossbar with its GeForce 6600 series, which has eight pixel pipes with only four ROPs, and it seems to work reasonably well. Halving the number of ROPs allows NVIDIA to save die space, which should make the NV44 cheaper to produce. To further reduce die size, NVIDIA has also removed color and z-compression, and OpenEXR 16-bit blending and filtering from the NV44 GPU. Removing color compression will likely hurt the chip's antialiasing performance, but budget graphics cards tend not to be powerful enough to offer smooth frame rates with antialiasing, at least with newer games.
Although NVIDIA removed a couple of ROPs and several other features to help reduce the NV44's die size, it did add a memory management unit designed specifically with TurboCache in mind. The memory management unit links the chip's pixel pipelines and ROPs to system memory, which according to NVIDIA, allows the chip to render to system memory with "100% efficiency." NVIDIA says that its design is more efficient than HyperMemory, a claim that we'll try to put to the test.
While NVIDIA claims to have the more efficient implementation, ATI's HyperMemory family is definitely less confusing than the TurboCache lineup. NVIDIA offers four TurboCache configurations, each with a different mix of memory bus width, clock speed, and on-board memory. We've spelled out all the possible configurations below.
|Core clock (MHz)||Memory clock (MHz)||Memory bus width (bits)||On-board memory (MB)|
|GeForce 6200 128MB TurboCache||350||700||32||16|
|GeForce 6200 128MB TurboCache||350||550||32||32|
|GeForce 6200 128MB TurboCache||350||700||64||32|
|GeForce 6200 256MB TurboCache||350||550||64||64|
NVIDIA is offering TurboCache cards with either a 32- or 64-bit memory interface, and as little as 16MB of memory on-board. Further adding to the confusion, cards are available with either an effective 550MHz or 700MHz memory clock. This wide variety of configurations offers between 2.2GB/sec and 5.6GB/sec of local memory bandwidth, and as we'll see, performance can vary quite a bit.
Wide performance deltas between TurboCache cards wouldn't be a problem if it was easier to identify a specific card. Unfortunately, a quick scan of online vendors reveals that TurboCache memory clock speeds and bus widths are rarely provided. Tracking down the size of a card's on-board memory is a little easier, but buyers are going to have a rough time figuring out whether a GeForce 6200 128MB TurboCache with 32MB of on-board memory has a 32-bit bus and 550MHz memory clock, or a 64-bit bus and 700MHz memory clock. Of the retail cards we gathered, including a 64-bit/32MB card from XFX, and 64-bit/64MB and 32-bit/32MB cards from eVGA, none of the boxes displayed the memory bus width or clock speed. In fact, only the eVGA packaging specified the amount of on-board memory.