|
Stability and Memory Issues
DDR
Using a single DDR DIMM in the first slot, the system was running stable at 100 MHz (PC1600 mode, 2:2:2:7 timing setting). However, if a second DIMM was installed, the system would immediately develop registry errors. As it turns out, the problem does not so much relate to a 2 DIMM configuration as it does to populating the second DIMM slot. Of all DIMMs tested, only the Mushkin PC2400 prototype was capable of running in 2:2:2 mode in DIMM slot #2 at 100 MHz bus speed. The Corsair PC2100 DIMMs made it to a full boot into Windows but did not finish graphics applications. None of the other DIMMs tested in the second DIMM slot managed to finish a full boot into Windows. Increasing the latencies to 2,5:3:3 Increased stability, albeit at the cost of a substantial performance hit.
|
[an error occurred while processing this directive]
Switching CPUs (133 MHz FSB) changed the picture in that the second DIMM slot became functional with the drawback that now the first DDR slot gave errors. In addition, none of the Micron DIMMs would boot even at the lower latencies. Of all production samples tested, only the Mushkin 16Mx8, and the Corsair DIMMs would boot at 133 MHz and run stable only in the second slot.
SDRAM
For a performance comparison, I tried running the Enhanced Memory Systems HSDRAM150 and the new Corsair XSDRAM 150, with the main difference that the HSDRAM would stop after high pitched beeping while the Corsair DIMMs would simply play dead. The only DIMMs that were actually running in PC100 Mode and 2:2:2 were the Mushkin Rev.2 (both 128 and 256 MB). However, it was impossible to finish a single benchmark without errors. Keep in mind that all DIMMs used here are capable of running at least 150 MHz at CAS-2.
Setting the memory to SPD did not change things at 100 MHz. Interestingly, however, at 133 MHz FSB, (1200 MHz CPU speed) both the Mushkin and the Corsair DIMMs booted at SPD (2:2:2:7) settings as well as at 3:3:3:7 latency cycles. However, it was not possible to finish one single run of Expendable timedemo which did not crash but went into a loop of erratic mass slaughter at various stages of the benchmark. Projected frame rates in 133 MHz and 2:2:2:7 was around 105 fps, dropping to some 100 fps at the higher latency setting.
Performance
In PC1600 Mode (single DIMM at 2:2:2), the A7A266 shows the typical sagging performance of a DDR system running at 100 MHz. Even considering the fact that the ATi Radeon was used, Expendable Timedemo scores of 81.5 (or 77.8 in 2,5:3:3 mode) are far from being competitive for a 1 GHz setup. With the Mushkin Rev.2 DIMMs, the benchmark would run but exit at random points or else get stuck in loops of the same sequence. Despite these errors, the estimated average frame rate would be at some 74-75 fps in PC 100 mode (2:2:2).
In 133 MHz mode (1.2 GHz, DDR at 2.5: 3:3), a score of 108 fps looks far better but would still place the A7A266 at the bottom of the stack.
In Sysmark2000, using two 128 MB Mushkin PC2100 DIMMs at 2,5:3:3, the system did not convince either, achieving at 1 GHz (10 x 100 MHz) a score of 180 (177 Internet Content Creation; 182 Office Productivity).
There is no point in beating on a dead horse, reason enough to leave it like this.
|
