We’ve recently done a number of articles looking at the performance of DDR and RDRAM on the Pentium 4 and this is as good as place as any to follow up with these and talk about some of the things we came across during our initial testing and the second round of testing we’ve just completed. The first article I’m referring too is ‘Rambus PC-1066 and PC-1200, pushing the envelope
’ were we run RDRAM modules and the i850 chipset at up to 50% over their designed clockspeed. That article definitely made a lasting impression as we showed that the scalability of current RDRAM was beyond what many, ourselves included, had initially expected. The stability of the system when running at 33% over specification was excellent, the PC-1066 system was just as stable as the system running at stock speed and finished hours of strenuous testing without a single glitch. However, the PC-1200 system was a bit more problematic and during our initial round of testing we experienced some instability that we couldn’t seem to remedy. Nevertheless we were able to show the ground-breaking performance of a PC-1200 RDRAM system.
During our second round of testing, which lasted for two weeks and was completed early this week, we set out to tackle that stability problem and if at all possible diagnose what could be causing the instabilities. After spending some time with the datasheets of the i850 chipset and the RDRAM modules and doing some board level measurements with an oscilloscope and a digital multimeter, it turned out that at these high busspeeds and under heavy loads, the voltage levels of the power lines that fed the i850 chipset and the RIMMs dropped below their minimal specified voltage.
Fig 1. Power routing on the Asus P4T-E motherboard, large capacitors and wide copper traces, signs of a well thought out design.
Clearly the power routing on the motherboard, or the power supply itself was having problems handling the load. But upon looking at the power supply itself it should be perfectly capable of handling the load, so there must be something else going on, something which engineers refer to as ‘power supply slew rate’. In plain English the slew rate of a power supply tells you how quickly the power supply is able to respond to a power surge without the voltage level dropping too much, thus the higher the number the better.
Fig 2. RDRAM interconnecting traces with the i850 chipset, these are impedance matched to provide even loading on all traces.
After some calculations on the capacitance and the resistance of the power routing on the motherboard we quickly concluded that the motherboard itself was not causing any problems and had enough reserve built in to handle the power surges, provided the power supply itself can also keep up with the load. It turned out that the slew rate demand that we we’re putting on the 300-watts power supply was too much and thus we set out looking for one that would fit our needs. After some looking around we found a 400-watts power supply, the Antec PP-412X
that offered a much improved slew rate and we were happy to see that this remedied our initial problems with getting the PC-1200 system up and running stable.