Pentium 66 MHz vs 75 MHz shoot out

Everybody knows the P75 is better right?

The Triton Bonus

Most benchmarks comparing these two are biased in favour of the P75 by running it with a mature P54-era chipset, such as the Intel 430FX (Triton). The Triton chipset was a big hit. It introduced pipelined-burst L2 cache significantly improving performance. Also present was EDO RAM support and deeper post write buffers helping memory and PCI performance. These things combined boosted the performance by almost a one CPU class up. Don’t forget vastly superior IDE with DMA support. There is no surprise the P75 came well ahead in contemporary benchmarks comparing P5 with P54C.

However, it should be noted that the Triton chipset wasn’t available when P90/P100 or even P75 launched in March resp. October 1994. The first buyers had to run these chips on Intel 430NX (Neptune) chipset and it wasn’t until 1995 when buyers were able to get 430FX based chipsets. So many compared them with the previous P5 system based on 430LX (Mercury).

What if we do the comparison on level ground and compare if the P75 is able to hold its own against the P66 even without the boost from a “modern” chipset.

Testbeds

Here are two almost identical siblings. The first is the popular Batman’s Revenge Socket 4 board. It sports the first Intel’s Pentium chipset – 430LX (Mercury). It is by no means Triton but still was a solid performer back in its days. This particular board powered many OEM PCs.

And then it’s an almost identical twin – Intel Plato. Based on 430NX (Neptune). One would be pushed to find more than 10 differences between the boards 😀. The Neptune chipset is practically identical to Mercury as far as performance is concerned. Essentially a 3.3V of the same. Intel even couldn’t be bothered creating separate datasheets for these parts and created a single document covering both.

So we have the Pentium 66 MHz on 430LX board running its frontside bus at the same clock and PCI @ 33 MHz. On the other side, we have Pentium 75 MHz with 50 MHz bus and only 25 MHz PCI.

There is the same amount of RAM, save graphics – S3 Vision 868 PCI, standard Intel (neither slow nor aggressive) timings. Is 9 extra internal MHz enough to maintain P75’s lead when considering 16 MHz less on the bus and slower PCI?

Results

• 

The results confirm the expectation that the P75 stripped of its main advantage – the superior chipset – it is outperformed by the older P66 part. The P54C platform needs to go up one CPU class to P90 to re-establish the lead.

Sure, chosen benchmarks are all of mixed type (CPU, Mem, Graphics) load. There is no pure-CPU workload where more megahertz would win. However, that was just the case back in those days. Saving for the synthetic benchmarks most application software was showing more balanced load patterns where memory, cache and I/O played a part.