By Vito Cassisi - 14th April 2008 | < Digg It

Skip to Section: Danger of Overclocking Overclocking and Warrantee Overclocking Theory Components of a GPU - Core Clock - Memory Clock - Working out the rating of GPU Memory - Shader Clock Artifacts and Errors AGP/PCI-e Bus Voltage Alteration Determining a Cards Heat Output Memory Latencies/Timings Cooling the Card The GPU BIOS Getting the Right Ratio Getting out of Trouble Overclock Practical Required Software Find the Max Heat and Artifacts Find the Max Core/Mem/Shader Clock BIOS Editing/Flashing - Nvidia - ATI Tweaking the Overclock - Advanced - Exact Mode - VID Mode Advanced BIOS Editing Physical Modifications - Voltage Mods - Cooling Methods Software Modifications - Fan Control/Emergency Shutdown

Default | Printer Friendly Version | PDF

Overclocking is the process of increasing the speed of a device beyond its specifications. The concept behind overclocking GPUs is similar to that of CPUs and RAM, but the methods are quite different. In this tutorial I will delve into several overclocking techniques, processes and tricks to teach you how to get the most from your GPU.

Dangers of Overclocking

Overclocking is safe if done sensibly. I will not attempt to scare you away from overclocking. I believe that if you learn properly you can overclock without trouble, as long as you know your limits. This tutorial explains how to determine these limits. I am not liable if you damage anything by following this tutorial. Overclock at your own risk. Please note that overclocking can decrease the lifespan of your components. You are welcome to ask me any questions via the official forum.

Overclocking and Warranty

Before you leap into the world of overclocking, check that you aren't voiding your warranty. Most devices will forfeit your warranty when overclocking is performed. The main rule of thumb is to not overclock anything you cannot afford to replace. Even though overclocking is fairly safe to perform, it is not very smart to risk breaking something if you cannot replace it.

---

Overclock Theory

Components of a GPU

The three main components of modern GPUs are the memory clock, core clock and shader clock. The memory clock is the speed at which the VRAM (Video Random Access Memory) runs, the core clock is the speed at which the processing unit in the GPU runs, and the shader clock is the speed at which the pixel shaders run. All these speeds are rated in MHz/GHz. The aim is to overclock all these components as far as possible without producing too much heat or any artifacts/errors.

GPU overclocking is quite simple compared to that of CPUs. Overclocking is done via software applications such as ATITool or RivaTuner until the best overclock combination is found between the three GPU components. The universal way to read/write the clock speeds of these components is core clock/shader clock/memory clock.

For example, a GPU with a core clock of 570MHz, memory clock of 780MHz and a a shader clock of 756MHz would be written like this:

570/780/756

Although this isn't vital to know, it helps when comparing results with other people.

Below is a quick rundown regarding overclocking theory for each GPU component. Read both this theory section and the practical section before attempting an overclock.

Core Clock

The core clock is overclocked with incremental increases, usually in 5MHz jumps, using software applications. There is no multiplier or dividers unlike CPUs, so the value you set is the speed is runs at.

Memory Clock

The memory clock is overclocked similar to the core clock. If your GPU is using GDDR RAM, the value you put in is the real memory clock which is doubled to give you the rated memory clock, also known as the effective memory speed. The rated/effective memory speed is generally the advertised speed.

For example, if you were to set the real memory clock to 100MHz:

100MHz (real memory clock) x 2 (GDDR, Graphics Double Data Rate) = 200MHz (rated memory clock)

Overclock in small steps of 5MHz.

Working out the Rating of GPU Memory Chips

Memory ICs (Integrated Circuits) on GPUs have a particular rating in which the manufacturer guarantees it can run stable. Not all GPUs take advantage of this speed and are clocked below this rating. By finding the rating of your GPUs memory chips you can raise your memory clock to this speed straight away. After this you can overclock further in small steps.

To find the manufacturers rating, locate the small memory ICs on your GPU. Below is an example of memory ICs on an Nvidia 7900GT. They are generally scattered around the core of the GPU.



Select one of the ICs and look at the numbers after the dash. Most modern cards will have a two digit number. Older cards will be a single digit. When there are two digits, place a decimal between them. This number is the latency of the memory measured in ns. Now divide 1000 by the number you find. This is the real memory clock rating of your GPU. Remember, for GDDR cards, double this result to find the rated memory clock.

For example, the 7900GT below has 1.4ns ICs as show here:



Which equates to 1000÷1.4 = 713MHz

And on an older GeForce 2 MX200 the ICs are 7ns:



1000÷7 = 143MHz

Now that you've learnt this technique you can quickly determine a safe speed to start your overclocking.

Note: On some Samsung ICs there will be 2A shown. This is actually 2.8ns not 2ns.

Shader Clock

Not all GPUs support overclocking of the shaders independently. The latest Nvidia and ATI cards do however, allowing more control over the overclock as a whole. On older cards the shader clock is linked to the core clock, so by increasing the core you are also increasing the shaders. These shaders increase with the core at a particular ratio, depending on the model of the card.

Overclocking the shaders is the same process as the core clock. Choose the speed and set it. As with core and memory clocks, increase the speed in steps of 5MHz.

---

Artifacts and Errors

Artifacts are graphical errors which are visible when the GPU is under stress and overclocked a little too far. Graphic processing errors are not always visible but can be picked up by specialised overclocking tools. Having errors/artifacts does not necessarily mean your GPU is damaged, most of the time you just have to underclock a little and the errors will no longer appear. Another method is increasing the voltage of the components; this will be discussed later in the tutorial.

AGP/PCI-e Bus

The AGP (Accelerated Graphics Port) and PCI-e (Peripheral Component Interconnect Express) busses are what connects and allows communication with the GPU to the chipset on the motherboard. Overclocking this bus usually makes no difference to performance, although there are rare cases which may benefit from a small overclock. It is not recommended that you overclock this bus unless it is a bottleneck (holds back other components) as it can easily cause system instability.

Voltage Alteration

Increasing the voltage of the GPU core allows a higher stable overclock. The downside of increasing the voltage is that the heat also increases, which is an overclockers nemesis. Different cards can handle different temperatures safely. If a voltage increase doesn't help stability, or only helps slightly, it is recommended to lower the voltage again to avoid unnecessary excess heat. If the temperature of your card is too high, invest in come aftermarket cooling.

Determining a Card's Heat Output

Software such as ATITool can monitor GPU temperature. This is useful because it allows you to monitor heat during stress tests, which essentially generates the maximum heat output of the card (depending on ambient/room temperature). Finding this maximum heat output ensures that your card is not overheating, and if it is, you can adjust the overclock/voltage/cooling accordingly.

Memory Latencies/Timings

Memory latencies describe the time it takes between the request of a byte of data from RAM, and its arrival. The smaller the latency, the faster this process is. Overclockers experiment with latencies to get the most out of their overclock. Raising the latency commonly allows a higher memory clock. Although it doesn't sound beneficial to raise the latency, it is not uncommon that the resulting higher memory clock provides greater performance. If you want to squeeze the most performance from your GPU, it is advised that you experiment to find the best combination of latency and memory clock.

Cooling the Card

All graphics cards have either passive or active cooling systems. Active cooling systems either use fans, water pumps, or other forms of accelerated cooling, and are often combined heatsinks/heatpipes. Passive systems only use heatsinks/heatpipes and rely on passing air to cool the GPU. It is not recommended to overclock passive cards as they usually run hot at stock.


Passive Cooling


Active Cooling

The main issue with many GPUs is that the core is cooled well but the memory ICs are bare. This means overclocking the memory ICs is harder due to excess heat. Either buying passive IC heatsinks, or a whole new aftermarket cooling system which caters for both the core and memory chips, will help solve the heat issue.

The GPU BIOS

Overclocking your GPU with software is not permanent, the overclock is only effective when applied at Windows startup. To make the overclock permanent, the core, memory and shader values must be altered in the GPU BIOS. This action is only recommended after extensive testing of the overclock to ensure stability. Once the BIOS is extracted, altered and flashed, any instability will be permanent, so ensure the settings are stable by stress testing the card and scanning for errors using ATITool for a few hours. Only when you are sure that the settings are stable you should flash the BIOS.

Getting the Right Ratio

Although getting the highest clock for the shader, core and memory components is the ultimate goal for overclockers, it may not always lead to performance gains. Synthetic benchmarks such as 3DMark thrive on every last MHz of an overclock, but actual games prefer a particular ratio or balance between the three components. This preferred 'ratio' is different for each model card, so if overclocking to the limit of each component results in degradation of in-game performance, then sticking to the stock ratio is best. The ratio could be something like 1:2.5 (core:shader) at stock. Always aim to overclock everything to its limit, then benchmark. Try the stock ratio at overclocked speeds and then benchmark again. Benchmark using in-game tests unless you're only goal is to get a high 3DMark. Choose the best performing setup using these benchmark results.

Getting out of Trouble

The most important thing when overclocking is having a backup plan. If something goes wrong, you want to be able to fix it without replacing the card. For example, if you were to make Windows boot up with an overclock that is too high, you'll have to boot into safe mode and change the settings. To do this, repeatedly press [F8] during bootup of the PC until a list of options appear. Choose 'Save Mode' and then login. Locate RivaTuner (or the program you used to boot with an overclock) and change the settings, or even disable it altogether. Now reboot like normal.

If you plan to do some BIOS editing, then you'll be smart to invest in a backup PCI GPU. Not PCI-e, but a good old PCI card. This backup card is used when your main card refuses to function properly, most likely due to a bad BIOS update/flash. The PCI card allows you to bootup and reflash the BIOS of the faulty card.

---

Overclock Practical

Welcome to the fun part of this tutorial. This is where you learn how to perform the alterations mentioned above. Be sure to follow the steps properly, and most of all, have fun!

Required Software

BIOS Editing

ATI:
RaBiT   [Download - Freeware]
Used to edit ATI GPU BIOSs.

ATI BIOS Editor   [Download - Freeware]
Used to edit ATI GPU BIOSs.

Computer Base ATi BIOS Editor   [Download - Freeware]
Used to edit ATI GPU BIOSs. Use this for the HD 38xx series cards.

Nvidia:
NiBiTor   [Download - Freeware]
Extracts and edits Nvidia GPU BIOSs.

BIOS Programming/Flashing

ATI:
WinFlash   [Download - Freeware]
Extracts and flashes ATI GPU BIOSs

Nvidia:
NvFlash   [Download - Freeware]
Flashes ATI GPU BIOSs.

Realtime Clock Alteration

RivaTuner   [Download - Freeware]
Adjusts core, shader and memory clocks and much more, in realtime via Windows.

ATITool   [Download - Freeware]
Adjusts core, shader and memory clocks in realtime via Windows. Also has an effective stress test.

USB/CD Boot Tools

wBIOSboot   [Download - Freeware]
Boot image designed for flashing BIOS ROMs.

IZArc   [Download - Freeware]
A compression/extraction tool.

physdiskwrite   [Download - Freeware]
Writes .ima files to storage devices.

Benchmarking

3DMark06   [Download - Freeware]
Benchmarks your CPU and GPU with a series of graphical tests.

---

Finding the Maximum Heat Output and Artifacts

This is the first and probably the most important step before overclocking. Depending on the model GPU you have and the cooling your system uses, there is a chance that your card is already at its maximum stable performance. The following steps help determine if your card can be overclocked further.

Firstly, fire up ATITool. Take notice of the temperature of your card, if it is too hot, then invest in better cooling. Do a Google search to find what temperatures are safe for your card. If the card is at a good temperature, hit the 'Scan for Artifacts' button.



During this scan, there are a couple of things to watch out for. Heat and artifacts/errors. As the test runs, keep your eye on the temperature of the card. If the card gets too hot, stop the test and don't overclock until you get better cooling. Also look out for any errors. With no errors, the test should look something like this:



Sometimes ATITool doesn't detect the artifacts, so it's best if you also look for abnormal rendering. Here is an example of this situation:



And this is what it looks like when ATITool does find errors:



Leave the test running for about an hour. If there are no errors or heat issues in this period of time, then continue with the tutorial.

Finding the Maximum Core, Memory and Shader Clock

The core is generally the first component that is overclocked. The process of overclocking the three components is essentially the same, but getting it stable and at its best possible performance is a little harder.

Open RivaTuner and click the small arrow button under 'Driver settings' and next to 'Customize...' Click on the first icon, 'System settings'. The first tab should be 'Overclocking' and show the Core, Shader (if applicable to your card) and Memory clocks. It also provides options such as 'Startup settings' and profiles. Don't let it apply the overclock at startup just yet.




As you can see, my 7900GT doesn't have the option to alter the shader clocks. Be sure to overclock the 3D and 3D performance states simultaneously, unless you wish to tweak them separately. Use the dropdown list to choose between each. Starting with the core, increase the clock in 5MHz increments. Click 'Apply' to set the new clock. Have ATITool running and ready, after every increment stress test the GPU by scanning for artifacts. Stress the card for 5-10mins every increment and keep watching for heat problems or errors. When you've either reached the maximum safe heat, or errors start to occur, underclock a little and test again. When the problems no longer appear, move onto the shaders (if applicable) and then the memory clock. Test the same way for each.

Now go play a graphically intensive game. If the game runs without graphical distortion then you probably have a stable overclock. Run the ATITool stress test for a couple of hours to seek out any long term errors. Now open 3DMark06 and do some benchmarking! If you wish to try different combinations and ratios of the core, shader and memory clocks, test them as above. Usually the shader clock has the most effect on gaming performance, so try overclocking that component first, and then the core and memory. Experimentation is the key to performance.

When you are confident that you have the best stable combination, either make RivaTuner boot with the new speeds, or flash the settings into the GPU BIOS. To edit and flash the BIOS, see 'BIOS Editing/Flashing' below.

BIOS Editing/Flashing

Editing the BIOS is the only way you can store your overclock permanently to the GPU. This means that you can take your GPU to any other PC and it will still retain the overclock. Editing your GPUs BIOS is fairly risky, so have another read of the 'Getting out of Trouble' section above. The steps for both Nvidia and ATI are provided below.

Note: It is possible to download a pre-customised BIOS and just flash it straight to your card. This is not recommended unless you know what these alterations are (and if your card can handle them safely), and even then only use it if you can't customise it yourself. Always ensure that the BIOS is functional and has good feedback before flashing.

Nvidia BIOS Editing/Flashing

Nvidia cards at this time are the easiest cards for BIOS editing. The first step is to extract the BIOS from the card using NiBiTor. Open NiBiTor then navigate to 'Tools' > 'Read BIOS' > 'Select Device...'



Now select the GPU in which you want to extract the BIOS from. Then click 'OK'.



Now navigate to 'Tools' > 'Read BIOS' > 'Read into File...' Save the file to a safe place, this will be your backup BIOS. Make a copy of this backup for editing. Import the copied BIOS into NiBiTor by clicking 'File' > 'Open BIOS...', then find and select the BIOS. Make sure the BIOS is detected properly and that the model is correct. Watch for the integrity of the BIOS as shown here:



The green icon is what you're after. If you're having problems, have a look at the help provided by NiBiTor, otherwise the risk of causing damage to your card will be increased significantly. Now that the BIOS is loaded you can alter the clocks for your card. Using the results from your overclocking and stability testing, fill in the following sections:



Note: You can alter the bootup name and settings of the BIOS by clicking the 'Bootup Settings' tab.

Now that you have input the values, go to 'File' > 'Save BIOS...' and then save the new customised BIOS to a bootable start-up floppy disk. If you don't have a floppy disk (or floppy drive for that matter), then follow the 'Boot nvFlash from CD' section.

Now download and extract 'nvFlash' to the floppy/flash drive. Reboot your PC and boot off the floppy/flash drive (you might need to change boot priority to do this). At the nvFlash prompt, type:

nvflash bios

Where 'bios' is the name of the BIOS you saved. After the flash is done, you have a permanently overclocked GPU.

ATI BIOS Editing/Flashing

At present, editing the BIOS of an ATI card can be difficult due to the limited card compatibility in ATI BIOS software. Some BIOS editing tools only support certain cards, so try different editors until you find one that is compatible. The following steps will show you how to edit and flash the BIOS of an ATI card if your card is compatible.

Firstly, open WinFlash and hit the 'Save' button to save the current GPU BIOS. If your card isn't supported, do not continue (or look for an alternative program). Save the BIOS to a safe place and make a copy of it for editing.

[Image]

Fire up RaBiT and load the copied BIOS by clicking 'Open' > 'ROM BIOS File'. Hit the 'Clocking' tab and adjust the clocks according to the results of your overclocking and stability testing.

[image]

---

Tweaking the Overclock - Advanced

This section explains how to alter voltages on your GPU. The alterations here can cause damage to the card easily, so take things slowly and be patient. Do as much research as you can before changing anything, Google is your friend. See what problems and successes other people have had. And remember, just because one guy can reach a really high overclock, it doesn't necessarily mean your card can too.

Changing the voltage of your graphics card is the best way to gain more stability while overclocking. The main problem is that it also significantly increases the heat. Only increase the voltage if your card is running well below its heat threshold. The other performance booster is the alteration on memory latencies. This is very complex, and only the method will be covered here, not the theory. DO not alter the latencies if you are not experienced in doing so,

Advanced Nvidia BIOS Editing

Below are instructions on how to alter the GPU core voltage and memory timings using NiBiTor. Please be cautious, I don't take any responsibility for any damage this tutorial may cause. Only attempt this if you are confident in your overclocking knowledge.

Open your overclocked BIOS using NiBiTor (see 'Nvidia BIOS Editing/Flashing'). Click the 'Voltages' tab.



There are two sub-tabs here, 'Exact Mode' and 'VID Mode'. Exact Mode is the safest because it only shows 'known good' configurations. If 'Exact Mode' doesn't have the options you need, go to the 'VID Mode' section.

Exact Mode

This is the default tab. You will be met with few dropdown menus; the only one that concerns most people is the '3D' voltage. Increasing the 2D voltage is not required unless you are overclocking the GPUs 2D clocks. Most people, however, don't have the need for more 2D power.

Change the 3D voltage to a higher value.



If the voltage you want is not available, tick the 'Extend Voltage Table' for more options. As you can see, my particular BIOS doesn't provide this option. If you cannot select the voltage you need, try the VID Mode option below.

VID Mode

If you can't select the voltage you want with Exact Mode, click 'Tools' > 'Voltage Table Editor'.




To add another voltage option, click the dropdown menu beside 'Amount of active entries' and change it to '02'.



Now that you have two active entries, you must assign a different 'VID' or 'Voltage Identification' to each. You can select more 'active entries' if you wish, but each should have their own VID (this particular GPU only has 2 VIDs, '00' and '01').



Click 'OK', you should now be able to select the new voltage.



Note: Although the voltage appears to have been raised, it doesn't necessarily mean it has. Some cards don't support voltage alteration, or the VIDs don't match/support the hardware. This is why Exact Mode is recommended over VID mode, but if you want to experiment, VID mode may be your only option.

Now that you have you new voltage, we can adjust the memory timings. This process is completely manual and only recommended for veteran overclockers.

Click the 'Timings' tab and hit the 'Autoselect timingset' button. Choose your GPU from the device list and click 'OK' twice. This sets the timings to the cards default timingset.





It's best to edit the default timingset, otherwise the card might not use your tweaked timings. Adjusting memory timings involves the following:

Edit the default timingset by hitting 'Detailed timings'. You'll be introduced to a selection of timings, adjust these as you see fit.




Click 'OK' to save the settings. Now you should test these timings to ensure they are stable using the 'Test timings' feature. Click 'Test timings' and copy across the corresponding hex values.



Make sure you save any work you are doing on your PC. Click 'SetTimings' to set the GPU to the timings temporarily. If your system locks up, the timings are most likely inadequate. Rebooting the system will reset the GPU to its original timings. Experiment to find the most stable/suitable timings.

If you've 'loosened' the timings, try overclocking a little further using ATITool/RivaTuner to test, and NiBiTor/nvFlash to edit and flash the BIOS.

Congratulations, you have now tweaked your GPU overclock. Remember to use nvFlash to flash the customised BIOS to the GPU.

Advanced ATI BIOS Editing

Under Construction

---

Physical Modifications

Physical modifications can be performed on cards to unlock disabled features or to manually increase voltages. These alterations are often left as a last resort, and have the highest risk of damaging the card. Below are the most common types of card modifications.

Voltage Mods

Voltage modifications are used when software applications (including BIOS editors) cannot provide enough voltage. These mods vary from card to card, so there are no generic methods of performing this particular modification. If you want to attempt a voltage mod, use Google to search for the best method for your particular card.

Cooling Modifications

This one is the most obvious. Cooling modifications usually involve replacing the default stock cooling device and replacing it with a better aftermarket one. After all, better temperatures means a higher overclock. Different model cards use specific aftermarket cooling devices, so an definite method is not available. Aftermarket coolers come with specific instructions.

Software Modifications

This section describes ways to alter software based features such as fan speed alteration and emergency shutdown procedures.

Fan Control/Emergency Shutdown

Fan control is very helpful for many reasons other than reducing heat. Firstly, it allows the fan speed to be reduced when the GPU is under little stress; this reduces system noise. Another advantage is setting up fan profiles which can be activated when GPU temperatures hit pre-defined values. This customized automated process gives you the best of both worlds, cooling vs noise, depending on the card's temperature. These settings can also be modified to shutdown the PC after a pre-determined temperature value is reached.

Due to my 7900GT not being able to utilise these features, I will instead temporarily link to a great guide which shows how to setup these settings. I'll create my own version ASAP (when I get access to a compatible machine). Rivatuner V2.02 Guide - Derek Wood.