Benchmarks I
As is usual, SiSoft's SANDRA starts off the benchmark batting. Memory performance is usually a reasonable indicator of expected performance. Let's see how the 648 does.
The benefits of running faster memory are clearly evident here. We're approaching 3GB/s bandwidth with the processor still at 133FSB. This may sound impressive, and indeed it is for DDR, but remember that the quad-pumped P4 can utilise up to 4.26GB/s at 133FSB, we're still far off satiating its need. Still, the number posted by the SiS648 do look impressive in any context.
How do these numbers translate to real-world performance ?. Let's investigate as we turn to our first practical test, Pifast. Pifast simply calculates the constant Pi to X million decimal places. It requires masses of bandwidth allied to pure CPU grunt. An efficient memory controller is imperative here.
DDR400 is showing its worth here. Nothing has changed apart from the speed of system memory. The DDR400 results would have been that much faster had I been able to use stricter memory timings.
Let's move on to DVD - DivX encoding. I'm now using a quality-driven approach. 2-pass encoding via VirtualDub. Three Kings is the DVD of choice, resized to 720x304, precise bilinear and black borders cropped. YUV2 spacing is used and the bit rate is set to 1700 kbps. I encode a 15 minute section and calculate the average fps from there. DivX 4.12 is still my CODEC of choice.
I've used two highly memory-dependant benchmarks so far. Both show a noticeable gain going from DDR33 to DDR400 memory. Almost 3GB/s of bandwidth certainly plays its part in the DVD benchmark.
