TexBench
I was interested in how each product was able to make use of system memory when on-card memory is full. I engineered a situation using TexBench where 40MiB of textures were being pushed back to the CPU via the application. The results were suprising and not really in line with the performance results seen.
The HyperMemory board can push texture data back to the CPU at nearly 800MB/sec while the NVIDIA hardware, at least in TexBench's simple test scenario, is a fair bit slower (almost by half). The game benchmark results show the TurboCache board to be generally ahead in performance terms, while TexBench proclaims otherwise. Pushing large textures back across the bus isn't what the hardware is doing in games, though, which is what TexBench simulates.
It shows that as resolution increases, the ATI hardware should pull ahead usefully, but there's such a lack of bandwidth and memory space overall that it never materialises. Neither product is particularly fast in modern games, so you never see it.
