Winamp & Shoutcast Forums - Overclocking/tweaking/3dmark tips and tricks.

Firstly, i should say that if anyone doesnt like reading long posts, they should stop here and go back to the game threads. :p Secondly, if you try anything described here and it leaves your pc a smoking pile of rubble i take no responsibility. For obvious reasons, im not going to go into detail describing how to actually make these changes, but google holds all the answers.

With another squeaky clean fresh install of windows (i know my windows product key off by heart now), and a finally trouble free system, i had an opurtunity to do test some tweaks, do some overclocking and get some benchmark scores.

Instead of just getting a baseline score, doing every tweak i know, overclocking, then benchmarking again and comparing the two, i decided to test every tweak to see how much difference there was afterwards. Each time i ran the benchmark twice, and averaged the two scores. The system was restarted after each change. I've never tried doing this before, because im usually too lazy to bother, but since my parents were tying up the internet connection, i didnt have anything better to do...

I found some results pretty interesting, so i thought i'd share them. Some may help you with tweaking your own system, or decide which is the best path for youre next upgrade. This is not a guide to getting the best 3dmark score.

To get a baseline score, all bios defaults were loaded, the cpu multiplier and fsb were set to standard, windows xp was installed, then service pack 1. ver 4.42 via 4 in 1 drivers, ver 30.82 nvidia drivers, sound card, mouse, network drivers etc. were then installed.

This is pretty much as close to a stock system as you'll ever get. Basically the way a pc would be set up if it was bought straight from a store.

The only other aplications installed were msi system monitor (to keep an eye on temps), nvmax (best freeware nvidia card tweaking program bar none) and 3dmark2001se. No settings were changed.

The system was left to idle for 10 minutes, with all case fans turned off. Base cpu temperature was 38 degrees celsius. Ambient temp was around 23 degrees.

I ran the default 3dmark benchmark - 1024x768, 32bit color, compressed textures, d3d pure hardware t&l, double frame buffer, anti-aliasing turned off.

Like that the system managed 8413.

By default, windows xp has a heap of background services and programs that arent really neccesary for everyday use. Disabling the unneccesary services and programs gave 8585. A small improvement, but if several programs were running (like firewalls, anti virus, download managers etc) the difference would have most likely been bigger.

Next i opened up regedit, disabled the paging executive (keeps the kernel in ram at all times), set the large system cache, changed the I/O page lock limit, and prioritsed irq 8 (cmos/realtime clock). The result - 8576. Lower than the previous score, perhaps the previous score was just an anomally, but i had to check. So I went back and set them all to defaults, the score didnt change, so i re-enabled them. None are really going to help 3d performance, but i thought it would be worthwhile testing them, like disabling background services, these tweaks would really only be useful if several other programs were running.

Next i moved the pagefile to the second hard drive. Again, this shouldnt have made any difference in 3d performance, and it didnt - 8578.

Setting the priority of 3dmark to high made no difference, since there werent any other programs hogging resources, but again, if there were other programs running it would have helped.

Then i went into display properties, set the card to render a max of 5 frames ahead (default 3) and changed the mipmap detail level to best performance. This is really a per taste thing, i normally have it set to best quality, but was interested to see how much faster it would run. The score - 8871, nearly 200 3dmarks, but not really a noticeable difference. In nvmax, it allows you to set the LOD even lower, and the result was 8810.

Next i applied the winxp/2000 athlon patch in nvmax, not expecting an increase in performance, but i wanted to check if it caused a drop. The scores remained unchanged.

So with windows and driver tweaks, id managed to pick up less than 400 3dmarks. However this was to be expected, the windows tweaks wouldnt make much difference unless there were several background programs running, and the driver tweaks dont usually make much off a difference. In fact its not worth turning the mipmap detail down for gaming, some games just look plain fugly like that.

A quick temp check showed 47 degrees, not bad considering i'd been pretty much non stop testing - rebooting - testing - rebooting.

Next i went into the bios, and set the ram latency agressively. The result was 9448, showing that if you can run ddr ram stable at cas2 its definatley worth it.

Back into the bios again, set the cpu multiplier from 10.5x (1400mhz) to 12x (1600mhz), and set its voltage to 1.80 (1.75 stock). 9901, showing that raw cpu speed still makes quite a difference in 3d performance.

Set the multiplier back to 11.5, set the fsb to 139 (1592mhz). That gave 9972, meaning the faster cpu-chipset-ram transactions made a difference.

Multiplier back to 11, fsb up to 145 (1594mhz), the result was 10142, again, the scores were scaling well with the increased fsb, while the cpu speed remained pretty much unchanged. I was pleased to hit 10,000 without touching the video card at all.

Next i set the agp aperture at 256mb (from stock 128), the score was 10189, an almost unnoticeable difference, but still an increase.

Cpu Temp was up around the 51 degree mark after each test, so the rear and front case fans were turned on.

Figuring that was going to be about the best i was going to get without touching the video card, i set the gpu to 300 (stock 275) and the video ram to 650 (stock 555), these are the exact speeds of the ti4600. The result - 10567, less than 5% increase, for a 10% gpu and 20% ram overclock. Strange, i had expected a much larger score that that.

I decided to bump up the fsb to 147 (1617mhz), gpu to 305, video ram to 655, giving 10538. Lower than the last runs, which was even stranger.

Thinking i may have reached the limit of the video card, i ran it at 310/665 to see if it crashed, id didnt, and it scored 10625. Perhaps it just didnt like the previous speeds.

The next run, with the fsb at 148, gpu at 315, video ram at 670 caused 3dmark to crash halfway through the lobby game test, but the system didnt. I was just booted back to the desktop. There were some occasional artifacts, meaning the video card had reached it's limit.

With the fsb back at 147, I was curious to see what score i would get with the video card set to ti4200 specs (250/444). It gave 9864.

Hmm..Seems that with an amd cpu around 1600mhz, the different geforce4 cards really dont have much of a performance difference at all. With the average ti4200 costing around half as much as a ti4600, it seems that the aditional cost isnt justified. Perhaps with an overclocked athlon xp 2200+ or a pentium 4 2.53 the gaps between the cards might have been bigger. But it seems that the bottleneck in most of todays systems with geforce4 cards is the cpu and ram, at least at 1024x768. With my old geforce2 mx400 i was able to get a 33% increase by just overclocking the card, while the jump in performance overclocking the fsb and cpu was smaller.

After 3 1/2 hours of testing the cpu temp was sitting around 48 degrees, so the top and side fans were also switched on, the (now extremely noisy) system left to idle for 10 mins. Temp dropped to 39 degrees. At this stage i hadnt seen a single blue screen or system hang, so i was pretty happy.

Next i set the ram back to defalt settings and bumped up the fsb to 166, cpu multiplier at 9.5x (1577 mhz), set the video card back at 305/655. The result was rather dissapointing. 10226. At 170x9 (1530mhz) gave 10398. 174x9 (1566mhz) gave 10476. Obviously lower because cas2.5 ddr333 is slower than cas2 ddr266, but with high quality ddr333 ram running agressive timings im sure the scores would have been much better.

Setting the ram at cas2 with 166 fsb caused the system to hang during post. Even trying the ram at 2.8v at cas2 it wouldnt boot properly, hanging even in safe mode. Running ram at higher than stock voltages can be a bit of a hazzard, definately not recomended for long term use, i dont even like running it for a few hours like that. I figured i'd reached the limits of my system, at least with its curent configuration.

So i decided to go back to what i know to be stable everyday settings, of 145x11, and 300/650 on the video card. I ran the benchmark 5 times to see if temperatures got too high, and they didnt, maxing out at 44 degrees with all fans on. 10434 was the average score, and i was quite pleased with that. I then switched off all the case fans and ran it 5 times, it got to 53 degrees. Warm, but still a fair way from the danger zone for athlon processors. It also shows that the geforce4 cards throw out some heat, especially when overclocked.

Anyways, in total the maximum 3d performance increase was 27% over stock, the settings i settled for were just under 25% over stock, a noticable difference. I could have possibly pushed it further, by upping the cpu, agp and ram voltage, The fsb may have been stable at 150mhz, but this would have most likely required removal of all pci cards, and with more voltage to the video card it would most likely have run at 320/690 but theres no way i'd run it like that every day, and theres no point benchmarking unless i will it run all day every day at those speeds....and i wont. It would just be too much stress on the hardware.

I was surprised to see that the biggest increase in performance was setting the ram latency agressively, so if you have recent video card and decent ram and youre running it with tame timing, there may be a good performance boost there. The windows tweaks made bugger all difference, but are still worthwhile if you run a lot of programs at once. If you have an unlocked cpu, then an increase in both the multiplier and cpu is definately worthwhile, but increasing the fsb speed more gives the best results. With an older card, overclocking can give some really big performance increases, but at low resolutions like 1024x768 the geforce4 cards dont seem to benefit much. I suppose the extra ram bandwith would really come into play at 1600x1200, but my monitor wont run at that res, so i couldnt test it.

Still, this goes to show that just tweaking/overclocking one or two parts of your system on their own wont give the best results. It was definately worthwhile testing everything instead of just making all the tweaks and settings, then benchmarking.

Anyways, if anyone else has any tips or tricks which have managed to give a decent improvement in 3d benchmarks or had strange results when tweaking or overclocking, or just wants to post their current best 3dmark scores and how they got them it would be cool.

Also, if you'd like to flame me for writing such a frigging long post about someting so trivial then go ahead.