K6-2+ 550MHz or K6-III+ 500MHz, is cache or frequency more important?

After my last post comparing the K6-2+ and K6-III+ cores at the same 500MHz frequency, I wondered if it is better to run a K6-III+ 500 or a K6-2+ 550 MHz in my system to obtain the maximum performance. So the basic idea is to know if gaining 50MHz is better than gaining 128KB of L2 cache.

So here are my 2 contenders as seen by CPU-Z. The K6-2+ is probably the easiest chip to find if you build a retro-K6 system.

And here are the scores from 3DMark 2001 SE. Before the test I was sure the 50MHz increase would be the dominant factor, as it represents a 10% increase. From the previous test we knew that the 128KB increase at a fixed 500MHz speed yielded a 6% performance increase.

Surprisingly, the K6-2+ obtained a score of 2001 and the K6-III+ a score of 2016! So the cache increase actually outweights (slightly) the frequency gain. In practice I doubt that you will feel any difference, so if you have one chip or the other and wonder if you should upgrade, just stay with the one you already have :)

K6-2+ or K6-III+ at 500MHz, which is better?

I have quite a bunch of AMD Socket 7 CPUs around, from the humble K5 to the latest K6-III+ versions produced. So yesterday I made a little test with 2 CPUs I clocked at 500MHz to compare the 3DMark 2001 SE results on my system.

The first contender is a K6-2+ 500 (really a 550 downclocked), which is the low-power (0.18 micron, 2.0V) variant of the K6-III core with only 128KB L2 cache as you can see on the CPU-Z screenshot.

The second chip is a K6-III+ 500, which is the same chip but with the full 256KB of L2 cache, this one runs at an even lower 1.8V.

Running 3DMark 2001 SE gives us 1900 points on the K6-2+ and 2016 points on the K6-III+, which is a 6% increase only die to the aditionnal 128KB of L2 cache. This is of course only one benchmark which also takes into account graphics, but it gives us a good idea of the performance increase between the 2 “high end” cores produced at the end of the K6 family life.

K6-2 CPU performance reduced by 50% because of USB 2 card

Recently I found that my K6 system was running slow in some games, the mouse was for example lagging and games like C&C Red Alert were unexpectedly slow. So I ran a 3DMark 2001 benchmark with the default setting and got this, 1312 points. This is not normal for such a system.

One change I recently made was to add an USB-2 PCI card (NEC chipset) to my system so I can use faster USB keys to transfer data around. Could it be the problem? I went to the device manager and disabled the card under Windows XP. After a reboot I ran the benchmark again…1990 points this time. This is a huge difference.

Going even further, I physically removed the card and ran the benchmark once again, this time 2011 points! That’s around 55% increase in performance and it was all due to the USB 2 controller, even when nothing was connected to it.

So beware if you add hardware to your system…All the benchmarks were run 2 times and the results were close, so it’s not a fluke. And my system feels responsive again.

SATA to IDE converters

If like me you are (or are looking to) running an SATA HDD/SSD in you K6 machine, you are probably wondering if you should buy a dedicated bootable SATA host-bus adapter PCI card or maybe use an IDE to SATA converter to plug your SATA drive on the motherboard’s IDE ports.


A good Promise PCI card will cost you around 30€ and it works flawlessly and performs very well, but what about these adapters from China you can get for 10€? Well, I tested 2 of them; they all use the same chipset and only the PCB differs a bit. Both adapter’s PCB look like shit with soldering marks all over the place; they are cheaply made but at first glance they work well.


Unfortunately, they perform horribly bad to the point where I had the feeling my K6 system was 4-5 slower doing any action using the disk…so do yourself a favour and avoid them. Buy a good SATA controller instead!

Old Hardware Shop

One of the most frustrating experiences when building an old system is to be able to buy cheap parts but then have to pay a lot for all the shipping costs from different sources and countries…at least this is what I was faced with when building my K6 system :)

Fortunately there are some specialised online shops which actually sell old hardware for a very low price so one can combine shipping costs; I ordered several parts from CBO-DO.de and they are very nice and helpful :)

Gigabyte GA-5AX performance optimisation

Having my working K6 system since a couple of weeks now I began tweaking the BIOS settings for my Gigabyte GA-5AX rev 4.1 to find the best performance, as it took quite some testing I’m now sharing what I found ;)

First of all let’s remember the system:

  • AMD K6-2+ 550 MHz
  • 768 MB PC133 CL2  RAM
  • nVidia GeForce 3 Ti200 64MB
  • Fast SATA HDD
  • Windows XP SP3

I began by setting reasonably good BIOS settings and after each individual setting change I  ran a series of 5 benchmarks using the good old Quake 3 v1.32 with “timedemo 1, demo four”. Quake 3 has the advantage of giving very consistent and reproducible results over any other benchmark tool and after all I’m using this system for old games…

The first tests I made is not BIOS related though…I tried to find the best nVidia driver working with my card. The 23.01 gave me an average of 28 FPS and the 21.83 WHQL an average of 28.7 FPS so there is the first base…I’ll be using the ForceWare 21.83 WHQL for the rest of the tests.

Now I’ll go trough each BIOS setting and give the fps change from this baseline…note that the changes are “incremental” by always keeping the previous test’s best performance.

Primary Frame Buffer

  • 2 MB (BIOS default): 28.7 FPS
  • All: 28.2 FPS
  • Disabled: 28.8 FPS

VGA Frame Buffer

  • Enabled: 28.8 FPS
  • Disabled: 27.7 FPS

Data Merge

  • Enabled: 28.8 FPS
  • Disabled: 27.7 FPS

Tweaking “IO Recovery Period” or “AGP Texture Size” didn’t change anything. Then I changed the CPU from an AMD K6-2+ 550MHz to an AMD K6-3+ 550 MHz to see how the 128kB additional L2 cache influenced the result. Well the fps jumped from 28.8 to 31.1 FPS which is nice!

As you can see I wasn’t that wrong with my default configuration as I only managed to “lower” the result by tweaking some settings…that’s experience talking ;) Here are the pictures of my final best result settings…

GA-5AX BIOS – Base Menu

GA-5AX BIOS – Power Management Setup

GA-5AX BIOS – PNP/PCI Configuration

GA-5AX BIOS – Integrated Peripherals

GA-5AX BIOS – BIOS Features Setup

GA-5AX BIOS – Chipset Features Setup

Arctic Cooling Copper Silent 3

Quickly after installing my K6 system I realised how noisy all the fans were back then…the stock CPU cooler just runs a small fan at full speed regardless of the CPU’s generated heat and I wanted to throw the whole system out of the window after 30 minutes.

So I began searching a good and quiet cooler for the K6 box…as nobody is selling Socket 7 coolers nowadays it wasn’t easy at first. Then I remembered that the Socket 7, 370 and A actually had the same fixations and it turns out Arctic Cooling is still selling a very nice cooler for them!

So I bought a couple of Arctic Cooling Copper Silent 3 (also known as Copper Lite) for Socket 370/A. I doubt they will continue selling them for much longer so grab one if you can…as you can see on the picture above it’s a massive cooler with a copper core. In the package there is also a tube of thermal paste which is nice.

On the GA-5AX you’ll have to cut away an unused part of the fixing system because there are capacitors on the way…not a big issue.

The biggest issue you’ll have with such a large cooler are the capacitors on the side of the Socket. In the picture above with the 12.5mm tall capacitors I replaced it works just fine, but with the stock 15mm capacitors found on the GA-5AX it will not fit unless you work with your Dremel.

The fixing mechanism has to be installed this way, with the pression point at the center of the CPU. I found that the best way to mount the cooler is to remove the screw, gently put the cooler into place while the mounting pins are lose and then apply a moderate pressure with the screw.

As you can see the final result is quite massive but also very quiet and the K6 runs at a completely cool temperature…no wonder there as it was designed for the very hot Athlons at the time. On the package it’s even written “Cooling Capacity: 90W” so don’t worry too much about cooling a K6 ;)