Linux Kernel Compilation Benchmark

I have purchased a K6-2/266Mhz processor and a Soyo 5EHM Super7 motherboard specifically so that I would be able to benchmark the K6-2 against Intel PII and Cyrix processors. Since I have been running Linux since the v0.10 days, I thought it would be useful to perform some benchmarks under Linux. Here are my findings.

As I have time (and access to equipment) to add additional results, I will update this page. Soon I hope to add PII results (ABIT LX6, 64Mb 10ns Hyundai SDRAM, Riva 128, same 1.6Gb WD hard drive).

System Description - Super Socket 7

• Soyo 5EHM motherboard (MVP3, AGP, 1M cache, Super7)
• 64Mb 10ns SDRAM (Hyundai, 2x32Mb sticks)
• 1.6Gb Western Digital hard drive
• Asus 3DExplorer AGP (Riva 128)

System Description - Slot 1

• ABIT LX6 Slot 1 motherboard
• 64Mb 10ns SDRAM (Hyundai, 2x32Mb sticks)
• 1.6Gb Western Digital hard drive
• Asus 3DExplorer AGP (Riva 128)

Methodology

• 'make clean; make dep; time make zdisk'
• using the same kernel configuration under Redhat 5.0
• Linux v2.0.32, with this .config file
• GCC v2.7.2.3

Cyrix/IBM PR233 Results

PR Rating	Voltage	Setting	BogoM	User	System	Elapsed	CPU util
PR200	2.9	2.5X66	166.30	283.22	21.26	5:28.56	92%
PR233	2.9	2X100	199.88	236.4	17.48	4:35.97	91%
PR233	2.9	2.5X75	187.19	257.99	20.17	5:01.32	92%
PR266	2.9	2.5X83	207.67	233.75	19.51	4:35.40	91%

AMD K6-2 266 Results

Intel Pentium-II 233 Results

Simulated Celeron Results - Intel Pentium-II 233 with L2 Cache disabled

Discussion of results

Please note, in the comparisons below, only the "User" time of the kernel compilations is used.

In order to be able to make a direct, Mhz-to-Mhz comparison of the processors, I underclocked the K6-2 to run at 2.5x75 (Cyrix PR233 rating) and 2.5x83 (PR266 rating). Comparing the total elapsed time for the compilation, we find that:

Comparison by actual Mhz

The AMD K6-2 processor seems to be 5.25% faster than a Cyrix MX processor at the same clock speed. The PR rating system would not seem to apply to Linux kernel compilations.

Comparison by PR rating

The AMD K6-2 is 14.5% faster than a Cyrix/IBM 686MX when comparing a K6-2 at the same Mhz as a Cyrix chip is PR rated for.

How about the Pentium II?

I wanted to see how the P2 would compare to the K6-2. As I only have a P2-233, I had to overclock it to approach 300Mhz. Please note, I used an extra 3" fan blowing air at the CPU.

Comparison between P2 and K6-2

NOTES

1. P2 at 262.5Mhz (75x3.5), K6-2 at 250Mhz (2.5x100)
2. P2 at 291.6Mhz (83x3.5), K6-2 at 300Mhz (3x100)

The P2 was faster for compiling the kernel by less than three percent. There is no point in comparing the K6-2 to the Celeron - see the simulated Celeron benchmarks on the previous page. The Celeron is not suitable for use as a Linux development machine.

Price Comparison

As we all know, absolute performance is just part of deciding which processor to get. If cost was no object, we would all be running Kryotech Alpha 767's, or dual PII-400's. For reference purposes, here are some prices, in US$, as 1:14pm PST of July 5 from PriceWatch.

Conclusion

There is no question that the Cyrix processors provide excellent performance for a low cost. The K6 (non-3d) processors are also an excellent value, however as I don't have such a CPU I was unable to run tests on one - but I would expect that on the same motherboard with similar memory and hard disk the performance of the plain K6's would be very close to the K6-2's.

The K6-2 appears to be an excellent value for a developer's machine. A 14.5% increase in speed over the 686MX is difficult to ignore. The P2 is less than three percent faster than the K6-2 at comparable speeds. I do not think that such a small difference in speed justifies the price differential between the P2 and the K6-2.

I hope you found this article to be of use. Please remember that I welcome feedback on this (or other) articles. I can be contacted at [email protected].

Regards,