17 entry and mid level graphic cards (page 1: Introduction, TurboCache & HyperMemory)

Announced a few days ago, the NVIDIA GeForce 7800 GTX is the new highest performance achiever. The downside is its price of $500, something only a few gamers and computer users are willing to spend for their graphic card. On the other hand, there is a whole spectrum of less expensive products costing from $50 to $150.

Of course, you often get what you pay for. These cards have lesser performances, but what is the performance difference? Can you play with them? Do they bring a significant contribution compared to Intel’s integrated graphic solutions, which dominate the market?

Reducing costs

In their evolution GPUs have increased their complexity. In consequence, they have bigger memory space, and bandwidth needs to access bigger and varying data (textures etc.) for more complex and less repetitive environments. On the other, hand to remain competitive compared to integrated chipset solutions, GPU and graphic card manufacturers have to keep production costs low. To resolve this conflict and make the GeForce 6200 a competitive product, NVIDIA released its TurboCache technology in December, which allows the card to extend its onboard video to include system memory. Consequentlly, NVIDIA avoids increasing memory, and even allows them to reduce it significantly.

Is using system memory as video memory new? Integrated GPUs do it as well as AGP texturing. Nothing new then? Yes and no. The difference is that PCI Express enables a better efficiency thanks to bigger (or even much bigger) bandwidth to send data from the GPU to the system memory. So it’s possible to use memory as a rendering area and not only a place to read textures. Also, the GPU can access two memory areas in parallel. This all comes mainly from PCI Express and new drivers efficiently supporting two memory spaces, even if NIVIDA speaks of in-depth GPU modifications to make TurboCache technology possible. Of course, there probably have been a couple of small modifications. TurboCache is PCI Express + optimised drivers to 99%.

TurboCache gives the possibility for NVIDIA to release cards with 16 MB, 32 MB and 64 MB and still keep (this is what we are going to verify) decent performances. As many users only differentiate graphic cards by the amount of embedded memory, the marketing department came up with announcing the quantity of addressable memory (graphic card memory + system memory). 16 and 32 MB cards are introduced as 128 MB and 64 MB as 256 MB. During the TurboCache release, NVIDIA assured us that the real memory quantity will be clearly displayed as they recommended their partners to do. 6 months later we can see that nothing has been done. The objective of many manufacturers and stores is that consumers notice the 128 or 256 MB but not the 16 MB!

ATI is next

When one of the two manufacturers releases an innovation, you can be sure the other will follow. ATI released a similar technology to TurboCache called HyperMemory. ATI hasn’t tried to convince us that major architectural modifications were made. They only optimised their drivers, removed a couple of memory chips and a new product appeared as the Radeon X300 SE HyperMemory. Just like NVIDIA the adopted strategy is to indicate addressable memory and not the real embedded memory, 128 MB for the 32 MB and 256 MB for the 128 MB version.

As the technology is linked to drivers and PCI Express, why not use it to increase performance instead of reducing costs? ATI visibly already answered that question as HyperMemory will be activated on all graphic cards with 128 MB or more with the next drivers! Gains will be variable and depend on the graphic card and game, but even if they are reduced it is still good to have it. Why not activate it for 256 MB graphic cards? Because it would probably reduce performances, the memory space and memory bandwidth would no longer be homogeneous. It’s preferable to stick to 256 MB rather than trying to address 512 MB with HyperMemory even if in specific cases it might help a little.