

![]() |
![]() |
|
|||||||
![]() |
![]() |
| Video Cards Nvidia, ATI and all video card related subjects |
![]() |
|
|
LinkBack | Thread Tools |
![]() |
![]() |
|
|
#1 (permalink) |
|
Civilian
Join Date: Jan 2006
Posts: 4
|
Benchmark Tweaking Guide
Per Rhino56's request, cut and paste from my guide at OCF for the most part for now, a few updates here and there. More to be done, especially with ATi LOD out now. Input from nVidia tweakers also appreciated.
Thanks goes to OC-Rookie from XtremeResources, D]g[ts and Glock19Owner from the Overclockers Forums So you've got a brand spankin' new rig with a killer overclock. Now its time to put your hardware to the test. Hopefully you have already found your optimal overclock and latency settings, but some settings to consider are: CPU Interface/1T Command Rate/CPC Enabled RAS to CAS delay (tRCD) (lower the better, 3DMark loves it) Row precharge time (tRP) same AGP Aperture (64mb usually works the best) Side Band Adressing/Fast Writes Enabled (I like it better this way, helps scores slightly) For Athlon64's, can be accessed using A64 Tweaker or some BIOSes: Write to Read Delay Time (tRTW) Read to Write Delay Time (tWTR) Max Async Latency Try messing with AGP speeds if you can. Even ATi cards can benefit. X800 AGP cards do best at 75-80MHz. X800's also can benefit vastly from tweaked memory timings. Some are able to pull way better scores at the same MHz by tweaking them. I, on the other hand, was able to get far higher speeds by sacrificing a little performance per MHz. For my X800XT PE AGP, and d]g[ts' X850XT PE PCI-E: TRCDRD: 7 TRCDWR: 4 TRP: 6 TRAS: 13 TRRD: 4 TWR: 6 TR2W: CL+2 TW2R: 4 TR2R: USE TWR RULE TR2R: 2 WR_LATENCY: 1.5 CAS_LATENCY: 7 CMD_LATENCY: 0 STR_LATENCY: WR LATENCY REFRESH_RATE: 0x75 TRFC: 23 For Sentential's X800Pro PCI-E: TRCDRD: 7 TRCDWR: 4 TRP: 5 TRAS: 13 TRRD: 4 TWR: 6 TR2W: CL+2 TW2R: 4 TR2R: USE TWR RULE TR2R: 2 WR_LATENCY: 1.5 CAS_LATENCY: 7 CMD_LATENCY: 0 STR_LATENCY: WR LATENCY REFRESH_RATE: 0x75 TRFC: 23 When running any benchmark, it is preferable to using the cleanest OS installation possible. The best option is to start on a brand-new, fresh installation. Serious benchmarkers often use seperate hard drives or partitions with a clean installation used exclusively for this. If you've got the room on your hard drive, a 5 gig partition benchmarking/testing installation can be very helpful. Begin with a fresh defrag. Likewise, you also want the cleanest hardware configuration possible. Remove any extraneous PCI devices, such as sound cards, Firewire cards, etc. For AMD systems and Dothans- most important tweak of all for 3D benchmarks, not for Pi. When beginning the installation of Windows (any version), srike F5 and when prompted, select "MPS Uniprocessor System." A dramatic increase in benchmark scores results, no loss of overclock, but a lot of general qurikiness. Definitely do not use this for 24/7 daily usage. Beginning with a fresh installation, the first thing that you'd want to do is install the latest chipset drivers for your motherboard. One of the most common causes of low 3DMark scores is not doing this. Secondly, install any necessary Service Packs, then DirectX, and finish off with the graphics drivers. This, along with whatever benchmarks you are running, are the only things that should ideally be installed on your partition. The drivers straight from ATi and nVidia are preferable. Omegas, etc, are geared towards image quality, and are not optimal for this purpose. For AMD's and Dothans benchmarking 3DMark2001SE, the optimal setup is Windows 2000 Professional paired with DirectX8.1 and Service Pack 4. Intels work better with Windows XP and either version of DirectX. (I'd prefer 8.1) Windows 2000 Service Pack 4 DirectX8.1 for Windows 2000 Now that you're finished installing, we move on to the tweaks. Begin by making sure that all quality settings, including mipmapping, antialiasing and anisotropic filtering are either set to the lowest possible detail levels or disabled in your ATi or nVidia control panel. Basic "tweak" but when overlooked can destroy your score. Glock suggested these services to be shut down: Automatic Updates Alerter COM+ Event System Distributed Link Tracking Client Distributed Transaction Coordinator Fax Service Logical Disk Manager IPSEC Policy Agent Print Spooler Protected Storage Messenger Remote Registry Service Server Telephony System Event Notification RunAs Service TCP/IP NetBIOS Helper Service Upload Manager Windows Installer Windows Management Instrumentation Computer Browser Workstation System Event Notification Removable Storage DHCP Client DNS Client One word of forewarning about the services- they can be a pain to disable sometimes, and the gains are generally minimal. They can bail you out though if you need a handful of points to break a certain barrier, etc. They take away internet access and other things necessary for 24/7 usage. More Tweaks: Under HKEY_LOCAL_MACHINE/System/CurrentControlset/Control/PriorityControl Create a DWORD Value named "IRQxPriority" where x is the IRQ of your video card, and set the value to 1. People often use 8 for the system timer, but the video card's IRQ works the best for me. Under HKEY_LOCAL_MACHINE/System/CurrentControlset/Session Manager/Memory Management: Set DisablePagingExecutive to 1 This forces drivers and the kernel to be run in the physical memory and not paged to the hard drive. Set LargeSystemCache to 1 This forces less paging to the hard drive, at the expense of free memory, which we don't care about anyways for benchmarking. Set SecondLevelDataCache to the decimal value of how much L2 cache your processor has. This is a controversial tweak. It's purpose is to optimize off-die cache as seen in ancient processors, but seems to provide some gains even for modern day ones. I don't question it and just do it. Use it for 3D benchmarks, but not for Pi. Now to move on to the actual benchmarking. Have your resolution at 1024x768x32bit@60Hz. Set the benchmark to either Realtime or High priority using the Task Manager. In 3DMark2001, you can get huge point gains simply based on how you run the benchmark from here. You can select each test individually, and by doing this, can alter the order, but additionally, push your overclocks higher than normal since you can allow your system to cool down between each test. In the "Selected Tests" area, click Change... and go to the custom tab. Deselect all tests but the one you want to run, and run one at a time. The famous OPPAINTER let this order out of the bag on XtremeSystems: Dragothic Low Detail Lobby Low Detail Car Chase High Detail Dragothic High Detail Lobby High Detail Nature Car Chase Low Detail- try and run this one about 5 to 10 times, saving each result in between. It's possible to gain up to 500 points simply by repeating this test. Bear in mind that for all three of these, only the game tests are necessary to calculate a score. Simply press the escape key when the CPU tests begin loading to get a score immediately in 3DMark03 and 3DMark05. Total scores are calculated as follows: For 3DMark2001- 10 * Low Detail tests + 20 * High Detail tests + 20 * Nature For 3DMark03- 7.3 * GT1 + GT2 * 37 + GT3 * 47.1 + GT4 * 38.7 For 3DMark05- (GT1 * GT2 * GT3)^(1/3)*250 Last tip: Never quit. If you've got a score barrier you've just got to break, be determined and you will be sure to achieve it. Good luck and happy benchmarking! |
|
|
|
![]() |
![]() |
![]() |
| Thread Tools | |
|
|
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Asus p4 motherboards overclocking guide | rhino56 | Processors/Motherboards | 10 | 09-19-2005 02:54 PM |
| How voltage modding works. Can't find a voltage mod guide? | rhino56 | Video Cards | 0 | 06-19-2005 09:28 AM |
| Official Benchmark Thread | Anonymous | Video Cards | 18 | 06-06-2005 06:10 PM |
| All times are GMT -5. The time now is 10:36 AM. |

