3. Benchmark Anomalies Due To Higher Resolutions And 2D/3D Graphics Accelerator

Using 2D/3D accelerators and VGA resolutions in a PDA has its consequences when running benchmarks, or rather, when interpreting their results.

The first issue is that the benchmarks currently available are written for PDAs with a display resolution of 240x320 (QVGA). If they are run on PDAs with higher resolutions, the operating system has to more or less translate the image output into a VGA resolution. The resulting process overhead for the operating system would seem to suggest that the graphics subsystem of a VGA-PDA is significantly slower. In fact, however, it's the operating system that "eats up" all that time with rescaling, not the graphics hardware.

The second factor is that the functions and benefits of a graphics accelerator like the 2700G can be exploited in real life only if the developer uses programming interfaces like Open-GL ES or DirectX Mobile when designing the applications. That's because their function libraries help utilize the chip's features. If the developer uses standard programming interfaces like GAPI when writing the application, systems with a dedicated graphics chip fare even worse than those without one in some benchmark results. The reason lies in the way the frame buffer is approached: software developers in the "GAPI world" often work by accessing the frame buffer directly using what's known as DFB (Direct Frame Buffer) accesses.

DFB access path in PDAs with no dedicated graphics chip

In a PDA with no dedicated graphics chip, the CPU will logically write the frame-buffer content to the on-die memory, if there is one, or to the system RAM otherwise. These accesses can occur very rapidly, depending on the architecture, and on-die memory of course works faster than the system RAM. Graphics benchmarks based on this principle measure the bandwidth between the CPU and the graphics memory.

When a graphics controller is available, however, the content of the frame buffer will be written to the on-die memory of either the graphics chip or its video memory.

DFB access path in PDAs with dedicated graphics controller

If a benchmark based on the CPU's DFB access is run on this kind of system, the results will appear to be worse than they should be. That's because it naturally takes longer for the CPU to access the on-die memory of the graphics chip or its video memory than it does for it to access the CPU's internal memory or the system memory. But it would be wrong to conclude that the graphics performance of this type of system is inferior; the processor access to video memory is just slower.

After that brief excursion into the benchmark problems associated with new system architectures, we'd now like to present the Axim X50v - in living color.