Overclocking The Very Low-End

Cheap upgrade, excellent performance – Stephan Höneise

Or: How an Applebred can lengthen the lifespan of your secondary/backup computer for really little money – like 30 bucks or so.

The purpose of this article is to follow my little adventure in upgrading my living-room machine the really, really cheap way. In my opinion, this would be a nice upgrade for any really budget-conscious overclocker – it should allow any secondary machine to last until A64s become more affordable next summer while still using mostly old equipment.

My idea was to upgrade the current setup with about only 30 Euros* that I had to spare to the best possible degree while maintaining the following guidelines:

• Reuse of present hardware as much as possible
• Low-noise and low-heat
• No new software installation required (I hate having to spend about a day just to properly reinstall W2K and Linux on a machine that we use only every other day…)
• Should be able to play Battlefield with low/medium settings at 800 x 600, 16 bit resolution
• Should be able to play DVDs, MP3s and DivX movies on my TV without black borders at the edge

These guidelines derived from my currently somewhat financially deprived situation – not working at my old job anymore limits, at the moment, my possible personal finances a lot. Anyway…

The old setup consisted of:

• Duron Morgan 1300
• ECS K7S5A Mainboard with the Cheepobios
• 512 MB SD-RAM PC133 Kingston, CL3
• Titan TTC 5DB Aluminum cooler with an 80mm fan (not the TC version!)
• 4MB GF4MX440 from Chaintech that uses SD-RAM (!), thus its actually more a MX420, clocked at 260/180 Mhz from 250/166
• 40GB HDD

All these components are mounted in a cheap, oversized aluminum briefcase for ease of transport to LAN parties, where the computer doubles its role as a server for Battlefield 1924. Unfortunately, the computer usually lives in a shelf under our TV, so ventilation is not overall very good there. An even bigger minus is that the computer’s main air intake is on its top (by a regulated Enermax 92mm fan, set to low rpms) and warm air exits the machines on each side of the case via the PSU and a 50 and 60mm fan @ 7 Volts on the CPU side of the case.

What happens is that the hot air exits at the side, then elevates and gets sucked in almost immediately by the top fan again. Reversing the direction of the fans in the case yielded little effect – now hot air exits on top, but most fresh air would get sucked out of the case before the hot components could get cooled enough, so the case temperature would decrease a little, but some parts, for example the VGA card, became too hot after a while.

So I decided to stick with the old case, despite its obvious flaws. *shrug*

The best chance for a cheap upgrade came when a coworker of mine wanted to upgrade her aging machine for cheap. She had a Duron 600 @ 800 on an Abit KT7 board that supports no FSB 133 – thus the biggest “new” CPU she could get to work on it would be a Duron 1300 – after doing a BIOS-Update – the one I had, for example…

Plus she wanted a new graphics card. For Solitaire, you don’t exactly need a Geforce4MX, but for watching movies on her TV in decent full screen quality, it would take more than her old S3 Savage. So we made a deal: she would get my old graphics card (30) and the processor (20) for 50 Euros* total (including me installing everything and stuff) and I’d go and replace my components with slightly better new ones.

My total budget was now up to 80 Euros* – the less I could spend, the more beer I could buy from the leftovers. 😉

What I did then was a little research on what would be the best bang for the almost non-existent buck. I really wanted to rather get a passively cooled FX5200 than a GF4MX again – but all the FX cards that I saw with passive cooling were 64 bit variants and thus not much faster than the GF4MX. Plus they would be more than 60 Euros*, leaving not enough money for a new CPU. So no thanks; for about zero extra performance, I’d rather stick with the proven-yet-sluggish GF4MX440SE.

I looked at some reviews for the PixelView MX440SE and found some favorable reviews. Not a great overclocker, certainly not fast, but the connectors were at the right place (if they are placed too high in the slot, the suitcase can’t close) and it was passively cooled. Plus it is really dirt cheap – so I gave it a try.

For a processor, I first thought about getting a XP2000+ CPU. They are sold for about 59 Euros* around here, but I saw the need for a new heat sink, since my 2400+ in my main rig runs pretty hot even with a Tbred-b core and using a Taisol heat sink/fan with a copper core.

The Titan cooler in the suitcase doesn’t sport a copper base – a colleague of mine had severe problems getting his XP2100 stable with this cooler, so I did not want to take chances on it and looked for a less demanding CPU. Alas, it would add up to be above my budget anyway, so I opted for a cheaper solution.

Duron Applebred to my rescue: while I was looking for a CPU, the new Duron was released. I read some forum comments about it at Planet3dnow.de and decided I should give it a try, since people claimed to have run the 1600 at 2 Ghz and more – air cooled! With little VCore increase. Hmm… Tempting. Plus I loved my old Duron 650 – that did 927 Mhz 3 years ago.

The store I went to had only the D1400 MIXHB variety – and for a good price: Deal!

So what I bought in the end:

• PixelView GF4MX440SE, 64 MB DDR-RAM: 40.99 Euros*
• AMD Duron 1400 – Applebred with a MIXHB stepping: 39 Euros*
• Total amount: 80 Euros*; minus the 50 Euros* I got for the old stuff, I was at exactly 30 Euros*. Well, I had a spare cent… A nice fit, right?

Back home I decided to get some beer first – I had to stretch the budget a little for that, but for really necessary things… well, you get the picture.

The K7S5A mainboard with its SiS735 chipset is definitely not an overclocking wonder – in fact, only with the hacked BIOS you can change the FSB a little. No voltage control, no multiplier settings – nothing.

But: If you get it to work right, then it is extremely stable and reliable – and it has no issues with running Tbred-Bs. In my opinion, this is very respectable for a two year old chipset! It’s like a VW Beetle: A bit slow and without bells and whistles, but it works, and works, and works…

I had used the OC-friendlier “cheepoBIOS” before, which enables some extra FSB-settings. Unfortunately, the old Duron 1300 would never run with FSB 110 as stable as its predecessor, a Duron 650, which even ran FSB 133 if I pushed the VCore up to 1.85V. That one got really hot, so I usually ran it with FSB 110 and later replaced it with the 1300.

First I tested the CPU under stock conditions – with a BIOS from January 2003, the CPU was detected properly (to my surprise) as a Duron 1400 and Win2K ran properly. I ran a quick benchmark and was amazed by the increase from 3500 points in 3DMark2001 (that identified the Duron as an AMD Thoroughbred) to almost 4000 – Nice.

But this site is called Overclockers.com – not stockspeed.com. So I opted to try to overclock the poor little thing.

Stephan Höneise – Germany

*One Euro = ~ US $1.17

More research revealed that with the “pin-mod” method, I could change the multiplier (hard-lock!) of the new Duron even on the K7S5A. So I went to www.ocinside.de and looked up the pin-mod-description there and the charts for the Tbred-b. The new Duron is nothing but a neutered Tbred-B, so I could use the given charts for it there. I looked up three settings, which I was willing to try out: 1600 Mhz, 1800 Mhz and an insane 2000 Mhz.

I also looked up the VCore settings, which I checked for 1.6 and 1.65 Volts instead of the 1.5V default.
I printed the sheets out so I could lay them next to the socket when I was putting the shorting wires into it. Time for a beer to steady my hands. 😉

I cut up an old floppy cable to get to the small wires inside. Cutting them to 10 mm in length, I then bent them into a U-shape. With some pliers, it was now pretty easy to get the U-shaped wires stuck in the appropriate holes in the socket in order to change the multiplier. I shot for the 1600 at stock voltage first.

After putting in the wires and applying the heat sink again (of course with *cheap* silicon thermal grease) – boot. A Duron 1200.

Whoops?

After a few seconds, I realized the mainboard defaulted back to FSB 100. I fixed that and booted again.

Surprise: A Duron 1600. Much better!

Then I tested the system, let some benchmarks run and decided I’d go for more. I upped the FSB to 143 and the Duron was now running at 1700 Mhz. Nice already, but I hoped for still more.

Next try was 1800 Mhz, FSB 133 at 1.6 Volts – no problem booting at that speed. Again benchmarks and Prime95 – no problems whatsoever. OK, up the FSB again to 143. At 1900 Mhz, the Duron still ran stable. Now I guessed this was already in the range of an XP2000, so I opted to try to get something extra out of it.

2000 Mhz settings and I set the VCore to 1.65 Volts.

POST, Windows – no problems.

With this old cheap aluminum heat sink, the CPU was reported to be running at 42C idle. That is a lot less than the old Duron not even overclocked. I was amazed. So I started the usual set of benchmarks and the Duron passed them all without any problems. It didn’t get hotter than 47C, which for this setup is really good.

OK – time to get mean: I upped the FSB to 143 and the Duron was now running at 2.14 Ghz. No problems. 43C idle. Benchmarks passed, Prime95 worked – shocking.

The Duron 1400 ran now as fast (clockwise) as my slightly overclocked XP2400+ @ 2.14 Ghz.

At this point I decided that for now this was enough CPU power, so I went to overclock the new VGA card a bit. 3DMark2001 reported 4500 points with just the CPU clocked this high. I started the machine and fired up Rivatuner to alter the settings of the VGAcard. I upped the settings from running at 270 core and 420 Mhz memory (default) to 278 core and 500 Mhz on the RAM. No artifacts, no crashes – the machine passed all the benchmarks I threw at it.

3DMark2001 now reported 5029 points – with some minor changes even up to 5190 points, but at the cost of lowered quality settings, which aren’t really visible on the puny 15 inch CRT I tend to carry to LAN parties or on the TV set, where the resolution isn’t that high either.

Now this was definitely nice – I increased the benchmark result by about 40% and in Battlefield I can now run the system with more or less medium settings at 800 x 600 – exactly what I was looking for.

I also managed to get a 55% overclocked CPU – and this is even a bit better than the Celeron 300A @ 450 trick that I remember from some years ago. Still a better overclock than the Duron 650, which managed to run at 927 Mhz on my old KT133.

Since a lot of people were interested in how the speed of the Duron compares to an Athlon XP clocked at the same speed, I did some testing with the machines at my place and posted them into the forum of planet3dnow.de, where I got the initial idea of going “the Duron way”. My results (in German) can be found
HERE.

For as little as the price of three crates of decent German beer (beer isn’t expensive in Germany), I managed to more or less double the gaming performance of my living room system. Plus, in total it’s now running cooler than before – not to mention the novelty appeal of having a competitive Duron system rather than a regular Athlon XP system.

To my knowledge, this is at the moment the fastest clocked suitcase worldwide – until tomorrow probably… 😉

Stephan Höneise – Germany

*One Euro = ~ US $1.17

Benchmarks with the Duron1400 @ 2.143 Ghz, the Athlon XP2400 @ 2.143 Ghz and the Athlon XP2400:

UT-Demobench was done with 1024 x 768, 32 bit settings. 3DMark2000 and 2001 were running in their respective default settings. No AA, No AF or whatever was used, Vsync was off. For games I usually switch on Vsync – that’s it.

Just to see the impact of using DDR-RAM instead of the aging SD-RAM, I tested the Duron also with DDR333 RAM @ 143 FSB, but I only used the GF4Ti, since the 4MX was severely underpowered to make any difference on the result:

GF4Ti + DDR-RAM:

• 3DMark2001: 9973
• 3DMark2000: 9988; CPU: whoops – I skipped writing this one down, but I’m pretty sure, it was 506 too.
• UT2003Demobench Flyby: 124.2
• Bot: 42.9

I also benchmarked the XP 2400 system with the GF4Ti at its default settings with 2 Ghz, to see how it compares to the Duron:

GF4Ti, DDR-RAM and XP 2400+

• 3DMark2001: 10372
• 3DMark2000: 11636; CPU: 656
• UT2003Demobench Flyby: 128.41
• Bot: 48.1

To sum up the results:

1) The Duron, using the same system setup (in this case DDR-RAM and the GF4Ti) is slower than a similarly clocked Athlon XP T-Bred. But not by a big margin – it is about 11% slower.

If the VGA card limits the test result, the margin is about 5%, whereas if it is CPU-limited, it increases up to 18%. An 11% tradeoff for a third less the price – if you are really on a minimal budget, you should keep this in mind.

Compared with the XP 2400 at stock speed, the Duron manages to close the margin up to 3% at best, and 15% at worst. If all measurements are averaged, we can see this development:

• XP 2400 @ 2.143 Ghz:     100%
• XP 2400 @ 2.000 Ghz:     96.5%
• Duron 1400 @ 2.143 Ghz:    89%

If this would be linear, a Duron Applebred clocked at 2.143 Ghz equals roughly an XP 2100 running at 1.733 Ghz – and which costs about 20 Euros*, or 33% more.

Detailed results:

XP 2400 @ 2.143 Ghz vs. Duron @ 2.143 Ghz:

• 2001 3Dmark: 10609 vs. 9973; Difference: 630 Points. Duron achieves 94% of the XPs speed
• 2000 3Dmark: 12172 vs. 9988; Difference: 2200 Points. Duron achieves 82% of the XPs speed
• UT2003Demo – Fly: 130.7 vs. 124.2; Difference: 6.5 fps. Duron achieves 95% of the XPs speed

• UT2003Demo – Bot: 50.2 vs. 42.9; Difference: 7.3fps. Duron achieves 85% of the XPs speed

XP2400+ at default settings:

• 2001: 10372 (97% of XP@ 2.143 Ghz) vs. 9973. Duron achieves 96% of the XP’s speed
• 2000: 11636 (96% of XP@ 2.143 Ghz) vs. 9988. Duron achieves 85% of the XP’s speed
• UT2003 – Fly: 128.4 (98% of XP @ 2.143 Ghz) vs. 124.2. Duron achieves 97% of the XP’s speed

• UT2003 – Bot: 48.1 (96% of XP @ 2.143 Ghz) vs. 42.9. Duron achieves 89% of the XP’s speed

2) SD-RAM, which was still sufficient to power the Morgan Duron, severely limits the new Durons’ power. The results dipped overall by a whopping 10% – meaning a total speed difference of 15-25% to a similar AthlonXP system. That hurts – on the other hand an AthlonXPs’ score would also dip in the same manner with SD-RAM. But the Duron system still manages to put out quite playable framerates with UT2003 even with the puny GF4MX440.

• 2001 3Dmark: 10609 vs. 9270; Difference: 1340 Points. Duron gets 87% of the Athlon XP’s speed
• 2000 3Dmark: 12172 vs. 8947; Difference: A measly 73 % of what the Athlon achieved with DDR-RAM
• UT2003Demo Fly: 130.7 vs. 116.73; Difference: 89% of the XP’s speed

• UT2003Demo Bot: 50.2 vs. 38.7; Difference: 77% of the XP’s speed

3) With a GF4MX, both systems are terribly limited by the weak VGA card. Benchmarking the Duron against an AthlonXP or a Barton with such a card is, IMHO, futile –
others may think differently though.

4) The Duron has a big problem using SD-RAM with 3DMark2000 – though I repeated the tests twice as often as with the other setups, the result was remarkably low. :/

5) Please note: All benchmarks were done using my “everyday” configuration on the machines – the results could be higher if I used a specialized “benchmark setup” – but this was not the goal of this comparison, since I was looking for typical day-by-day differences in the machines.

6) The K7S5A with the Duron Applebred reaches the max the mainboard is capable of – but for a new K7S5A, you pay about 40 Euros*, maybe 40 for a Duron 1400 and maybe 10 for a cheap cooler. The remaining 10 Euros* could be invested in a crate of beer – and with 100 Euros* total (given a similar VGA card, and even when using SD-RAM still) you can achieve a pretty decent system that should last you at least until next summer, when the Athlon 64 finally gets interesting.

You can contact me if you have questions about this article.

Stephan Höneise – Germany

*One Euro = ~ US $1.17