3DMark Tips

[Ace]

Note: Please excuse the pathetic scores - this is with 256MB RAM, not properly overclocked, not properly tweaked... TBH in quite a mess.

PROCEDURE
I first ran 3 tests doing the following:

1. Booting
2. Starting Task Manager
3. Shutting down everything except:

svchost
svchost
svchost
lsass
csrss
services
winlogon
smss
system

4. Starting 3DMark2001se
5. Closing Task Manager
6. Selecting the four game tests only
7. Running
8. Rebooting

After the three runs, I started regedit and deleted the value in the "Shell" string at:

HKEY_LOCAL_MACHINE/SOFTWARE/MICROSOFT/WINDOWS NT/WINLOGON/

I found that deleting the string itself didn't work, but I had to delete the value in it so in this case Explorer.exe

I also set the AutoRestartShell value in the same place to 0, but this is not necessary.

I then proceeded to try with no Shell starting at boot.

1. Booting
2. Starting Task Manager
3. Starting 3DMark2001se
4. Closing Task Manager
5. Selecting the four game tests only
6. Running
7. Rebooting

RESULTS

I took the average of each of the three runs.

SHUTTING DOWN SHELL MANUALLY

17567
17608
17572

=

17582 overall

REGISTRY TWEAK TO PREVENT SHELL FROM STARTING

17581
17621
17583

=

17595 overall


Therefore, I propose that it MAY be advantageous to set the shell NOT to start at boot, although gains may be small.

[Ace]

We all have our own little methods to tweak our systems, so I thought maybe we could pool our ideas and hopefully help a few people out

I may as well start...

Disable all hardware not required for the benchmark. As far as possible do this in the BIOS, and then move on to the Device Manager.

Go through the services.msc list and disable any service not required for Windows to run.

Also check msconfig for any unwanted startups.

Use the Task Manager to end Explorer (the shell) before running the benchmark.

Delete unnecessary fonts (don't know if this makes a difference).

Increase the AGP Apperture in the BIOS to about half the system RAM.

Increase the initial and maximum size of the paging file - I set mine to 3GB each.

Use the L2 cache registry tweak (instructions here).

Disk cleanup, defrag.

Best to use a dedicated hard drive and a clean install.

Uninstall anything not explicitly required for the benchmark - control panel/add remove progs/add remove windows components - remove all unnecessary programs like MSN Explorer etc.

Set ratio box to performance in control panel/system/advanced/performance - settings.

Set sliders to performance in control panel/display properties/settings/advanced.

[Colin]

Ooo tips...

I'm at work right now but when I get home I'll post what I do in WindowsXP.

[ElrichMeister]

OoOOoOoOoh i cant wait for this hehe, cause the only thing i do is take uneeded programs from the task manager, and set the all the sliders to performance in the advanced settings of the VC.

[SSL]

i disable almost all the windows services

take out any unnecessary hardware

close unnecessary running processes

i have everything i do writtin down at home, lol, does closing explorer seriously work tho? doesnt it jus reopen itself rite away?

[Ace]

In XP you can close the shell OK.

[Colin]

Sorry about the delay guys. Was kinda pissed off about the Vapochill SE/Classic thingy...

Anyway here are some of my tips.

As Ace suggested turn off all unnecessary devices (Serial/Parallel/USB, audio, RAID, etc), they eat up valuable system resources. You can turn them back on after you've finished benchmarking. Always set your memory to their most aggressive 2-2-2-5 with all the trimings enabled (PAT, CMD RT 1, etc). This is vital for AMD users, P4 ppl can get away with a higher CAS Latency but make sure RAS to CAS is set to 2. Myself I always set the AGP Aperature size to 128, I find that setting the thing to 256 causes some weird slowdows sometimes but YMMV of course. A few hardcore benchmarkign friends do like setting of 256 but it's really your personal preference.

Next it's important to select the proper OS. For some reason I find anyway that AthlonXP based systems score best with Windows 2000 Pro, up to 500 points higher than an identical system equipped with WindowsXP. Why is that, I have no idea! For P4 users it's the other way around, WindowsXP is about 1000 points faster than anything else.

The tweaks I'm going to be suggesting are mainly for WindowsXP but they do apply to Win2k as well.

After OS and driver installation the first thing I do is disable services... You can disable services via the Services.msc or MSCONFIG, both will turn the selected services off.

Here's the list of stuff I disable.
Alerter
Application Layer Gateway Service
Application Management
Windows Audio
Background Intelligent Transfer Service
Computer Browser
Indexing Service
ClipBook
COM+ System Application
DHCP Client
Logical Disk Manager Administrative Service
Logical Disk Manager
DNS Client
Error Reporting Service
Event Log
COM+ Event System
Fast User Switching Compatibility
Help and Support
IMAPI CD-Burning COM Service
Server
Workstation
TCP/IP NetBIOS Helper
Messenger
Netmeeting Remote Desktop Sharing
Distributed Transaction Coordinator
Network DDE
Network DDE DSDM
Net Logon
Network Connections
Network Location Awareness (NLA)
NT LM Security Support Provider
Removeable Storage
IPSEC Services
Protected Storage
Remote Access Auto Connection Manager
Remote Access Connection Manager
Remote Desktop Help Session Manager
Remote Registry
QoS RSVP
Security Accounts Manager
Smart Card Helper
Smart Card
Secondary Logon
System Event Notification
Internet Connection Firewall (ICF) / Internet Connection Sharing (ICS)
Shell Hardware Detection
Print Spooler
System Restore Service
SSDP Discovery Service
Windows Image Acquisition (WIA)
MS Shoftware Shadow Copy Provider
Performance Logs and Alerts
Telephony
Terminal Services
Themes
Telnet
Distributed Link Tracking Client
Upload Manager
Universal Plug and Play Device Host
Uninterruptible Power Supply
Volume Shadow Copy
Windows Time
WebClient
Windows Management Intrumentation
Portable Media Serial Number Service
Windows Management Instrumentation Driver Extensions
WMI Performance Adapter
Automatic Updates
Wireless Zero Configuration

As for the non microsoft stuff I also disable the ATi/nVIDIA driver services too. Please note that some of the services listed above are required for everyday computer use but my goal here is to create a lean dedicated benchmarking OS. To use yoru system properly you may be required to turn on some of the services above.

Sorry Ace I'm going to be copying you again on the swapfile thingy.

No matter how much physical memory you have in your system you must have a swapfile in Win2k/XP. Disabling the thing will cause your windows to crawl because some system services and applications require a damn swap file. You're going to want to set it to something large so Windows does not adjust the thing while you're running a benchmark. Myself I set the initial size to at least 1GB, setting the maximum size is not as necessary, you just want the initial size to be big enough that Windows will leave it alone when the benchmarks are being run.



While in System Properties in the Performance options window, set Windows (XP only) to for best performance.



After that's done defrag the HDD, this places the swapfile at the front of the HDD where things are fastest should Windows need to access it for whatever reason during benchmarks. I use Norton Speed Disk myself but you can use the built in Microsoft Defragger, it's just much slower.

Set all the sliders in yoru display driver properties to Fastest performance/Lowest IQ, we're benchmarking here not playing games.

As everyone knows, load up task manager and clear out all the unnecessary programs running in the background...



HSDX is the snapshot tool I use.

In 3DMark2001 there's a bug with the Game 1 tests. The Car Chase - Low Detail test can sometimes varry by 10-20 fps, and the Car Chase - High Detail can also vary by up to 10 fps. There's no known fix for this what I do is I run the Car Chase tests only first and see if I get decent fps's. If I don't, I reboot the system and retry again.





Same configuration and as you can see Game 1 in the first window is about 10 fps higher and game 2 is about 5, that's 300 points right there. True it's not a big difference but it can mean the difference between a bad run and a good run. This bug is more evident when the videocard is heavily overclocked, with my old 9800XT oc'ed sometimes the game 1 would vary 40fps, that's 400 points alone!

That's all I can think of for now, it's 7AM here in TO on a Saturday morning... I'm usually asleep but something I ate yesterday did not agree with me and kept me up pretty much all night. I'll post more tips when they come to me.

[therealwesty]

Well it's going to be hard to add anything after what Colin said. I'll list some of my tricks still though...

For 3DMark2001 DX8.1 scores higher than DX9.0x. I used Win2KPro when I am really trying to push my rig, and if I am going for extreme 2001 scores I reinstall Windows so I can use DirectX 8.1. A clean instally scores highest anyhow.

After the clean install make sure you install all the drivers, forgeting something like the chipset and AGP bus driver can be devistating to the score.

Next I do the L2 cache regestry tweak and disable a ton of services. And apply the 3DMark2001 profile from Rage3DTweak.

I benchmark on a dedicated HDD with only Win2KPro, drivers and the BM applications on them so there are no tray applications to shut down.

I always run the BM after a fresh reboot for best results.

That's all I got that hasn't been mentioned.

[Ace]

Nice post Colin!

Just thinking... you can set the location of the shell in the registry. Would it make any difference if you removed that key, so the shell didn't start at boot? Might that save RAM or something? Anything?

[Colin]

<< Nice post Colin!

Just thinking... you can set the location of the shell in the registry. Would it make any difference if you removed that key, so the shell didn't start at boot? Might that save RAM or something? Anything? >>

Ace, I never thought about that. Give it a shot and report back to us!

Man I'm dead tired afte replying to a few posts I'm going to take a nap. I'll try and post some more tips tonight.

[Colin]

You have a forbidden word in your message post. Please click the back button in your browser and remove this word from your post. The words that are forbidden are highlighted for you.


Great info Ace! nVIDIA users you are aware that you can tone down the LOD and texture filtering options even further in RivaTuner right? I don't do that myself because it makes the screen look like crap but it does increase score...

[ElrichMeister]

whoo these tweaks gave me a about 200 more points

new score 16248

[Ace]

I read somewhere that if you goto display properties, advanced, 3d, compatibliity and enable alternative pixel centre that increases your score a little, although I haven't had the time to verify this yet.

[Nova]

yea it does somethimes, i think mainly in 2001, since it forces the vid cards to quit using its current shaders (lets say 2.0) and revert back to its old (lets say 1.1) for example, and that will give you a bit of a performance boost

[Ace]

Oh excellent. Thanks for that Nova. I'd better remember to turn it off for 03 in that case.

[c0ldweller]

I don't know if anyone else's tried this. I have 1GB of RAM and one thing I do is load up Apophysis 2.0 (it's a special program for making fractal art and you can get it at ******.apophysis.org for free), not to make art, but to start a render. I start a render with a VERY high resolution setting. Then I let it run for about 1 minute or so. After that I close the program and start up the benchmarking utility right away (I do all my other tweaks prior to starting this program). The result of doing something like this is that while rendering, it takes up A LOT of system resources, making other processes take less amount of memory. So lets say when I start Windows I'll have about 820mb of free RAM. But once I exit the render, I have around 900mb of free RAM. I haven't done any specific tests to tell the difference, but I do know that it increases the PCMark2002 Memory score quite a bit so I would assume it helps in other benchmarks as well. Hope this helps .

[wacko]

ok please dont flame me.....
i dont see the point of going through all that trouble to gain a few points in a benchmark....i think benchmark are kinda useless. the condition in wich you put your PC to run the benches is not the same of course as when playing a game..and its there that the performance matters the most

Personnaly i think a better way to compare your PC to others should be with a specific games at specific settings....

Its prolly just me but i dont get that whole benchmark bragging rights...all id like is to enjoy my game running as high as possible

[blueice]

Wacko, you are absolutely right about this. It is like running ones machine on steroids.

[Zefram]

Wacko, you are absolutely right about this. It is like running ones machine on steroids.

No, its just helping to squeeze out every marks for the benchmarks.............. so, to be ontop of the benchmark list..........
I agree that you practically need to stimulate the real-world condition, but for once, you want to see a high benchmark score coming out from your PC