Overclocking ati: How To: Overclock Your AMD / ATi GPU

Extreme Overclocking ATI Radeon HD 3870

• Introduction
• ATI Radeon HD 3870 Extreme Overclocking: Technologies and Methods
• Extreme overclocking ATI Radeon HD 3870:

results

• Extreme overclocking: power consumption
• Test bench configuration and test methods
• Test results. 3D FPS Shooters
• Battlefield 2142
• Bioshock
• Call of Juarez
• Call of Duty 4: Modern Warfare
• Crysis
• Enemy Territory: Quake Wars
• Half-Life 2: Episode Two
• S.T.A.L.K.E.R.: Shadow of Chernobyl
• Test results. 3D Third Person Shooters
• Lost Planet: Extreme Condition
• Tomb Raider: Legend
• Test results. RPG
• Hellgate: London
• The Elder Scrolls IV: Oblivion
• Test results. Realtime Strategy Games
• Company of Heroes: Opposing Fronts
• Command & Conquer 3: Tiberium Wars
• World in Conflict
• Test results. Synthetic tests
• Futuremark 3DMark06
• Futuremark 3DMark Vantage
• Conclusion

As our readers know, in our reviews we regularly touch upon the topic of overclocking graphics cards, however, adhering to the so-called «reasonable overclocking» ideology. In other words, overclocking that does not require the use of any special tools — for example, replacing the cooling system with a water one or modifying the power supply system of the video adapter. Another hallmark of «intelligent overclocking» is that it can be used for extended periods of time in daily activities, and while the resulting performance gains are generally no substitute for practical benefits, many gamers are interested in and enjoying the concept.

However, in addition to «reasonable overclocking» there is also «extreme», which differs from the first one in the most fundamental way. It is called so for a reason, since it provides for truly extreme methods, up to cooling the overclocked card with liquid nitrogen and increasing the voltages that feed the GPU and memory in order to increase their overclocking potential. We know that almost every processor and memory chip is often able to operate at a frequency higher than that set by the manufacturer if their supply voltage is increased. Such measures lead to a sharp increase in the level of heat generation and a sharp drop in the life of components, therefore, extreme overclocking is rarely used by players, being, in fact, a sport, since it does not put stable operation in the first place, but the achievement of record performance at any cost. The excitement is high, but so are the stakes — almost every extreme overclocking enthusiast has encountered a situation where the pursuit of 3DMark points led to the failure of a modified card, and some even boast a whole collection of such “corpses”.

As a rule, flagship solutions are subjected to extreme overclocking, since they give the greatest chance of setting a record that you can rightly be proud of and show off to other enthusiasts. But we set ourselves another task — to squeeze the most out of the ATI Radeon HD 3870 without resorting to extreme cooling in order to find out how high the potential of the RV670 GPU is and whether it can at least theoretically compete with Nvidia’s solutions based on the G9 chip2. This is also interesting because at the moment there is a serious gap in the ATI Radeon HD 3000 line — between the relatively inexpensive single-chip ATI Radeon HD 3870 and the two-chip monster ATI Radeon HD 3870 X2, while in the Nvidia product line there is such a serious there is no break.

Real enthusiasts of extreme overclocking often limit themselves to fixing the set record in the form of 3DMark results. We do not pretend to be champions, but we will try to reach the 1 GHz air-cooled graphics core clock, and test the ATI Radeon HD 3870 in this mode using our standard set of test software. This approach will allow us to get a much more complete picture of performance, which is of significant interest to any player, even not interested in overclocking.

Note that the results of our experiments are not a guide to action and do not call for the modification of any products. Anyone who decides to repeat them must evaluate both their strength and the possible consequences.

A lot has changed since our last extreme overclocking experiment. First of all, graphics card power systems have become much more complicated — now they are widely used intelligent multi-channel PWM controllers with software control, which are used, among other things, to implement energy-saving technologies that reduce not only the frequency of the graphics core, but also its voltage .

Second, modern GPUs have multiple zones running at different clock speeds. The canonical example is the Nvidia GeForce 8/9 family, but this fully applies to modern ATI Radeon HD chips. Although officially one frequency is indicated for them, internally these cores are divided into many zones with different clock frequencies, moreover, they can change dynamically as part of the PowerPlay power-saving technology.

recommendation

These features often have to be taken into account when overclocking modern cards. For example, the original BIOS of the ATI Radeon HD 3870 does not contain dividers for frequencies above 862 MHz, as a result of which, any attempt to increase the core frequency above this mark instantly leads to a system freeze, regardless of the GPU voltage. Fortunately, this limitation is not fundamental and can be easily bypassed by flashing a modified BIOS version, which can be downloaded from this
link
. For firmware, you can use the ATIFlash utility — we used version 3.59. After that, you can start hardware modification of the ATI Radeon HD 3870 power circuits.

The heart of the ATI Radeon HD 3870 GPU’s power system is the uPI uP6201 chip, which is extremely difficult to find documentation on. Unfortunately, we were not able to do this either. However, the volt mod method is extremely simple and consists of connecting a trimmer resistor between pin 13 of the uP6201 and ground. We used a 100 kΩ multi-turn resistor. The connection points of the resistor are marked in blue in the picture below:

The easiest way to control the voltage is to use one of the Vgpu control points marked in red. By default, the voltage in 3D mode should be within 1.33V.

Volt mod memory power supply is even easier — a similar tuning resistor with a resistance of 100 KΩ is connected between pins 3 and 7 of the uPI uP6101BSA chip. These points are marked in green on the image:

For voltage control, the point marked as Vgddr control seems to be the most convenient for us. The default memory supply voltage is 1.89V.

After flashing the BIOS, which rid the card of the «862 MHz problem», the above hardware modification allowed us to control the GPU and memory supply voltages over a wide range. At the same time, the rotation speed of the cooling system fan was set to the maximum using the RivaTuner utility, which made it possible to keep the temperature of the graphics core within acceptable limits. Attempts to overclock the GPU led to the following results:

• 900 MHz (891 MHz according to RivaTuner) at 1.44V
• 950 MHz (945 MHz according to RivaTuner) at 1.51V
• 975 MHz (972 MHz according to RivaTuner) at 1.55V
• 1000 MHz (999 MHz according to RivaTuner) at 1.71V

As you can see, we came across a not very successful instance of ATI Radeon HD 3870 — we managed to achieve 1 GHz frequency only at a supply voltage of 1. 71V, despite the fact that there are cases when a similar result was achieved at 1.55-1.6V. An attempt to conquer the peak of 1.1 GHz at a voltage of 1.75-1.77V did not give results, and it was decided to stop further attempts, since even 1.7V is extremely dangerous for a 55-nm chip with a nominal supply voltage of 1.33V. This is indirectly evidenced by temperature indicators — even at maximum fan speed, the temperature of the graphics core in idle was 69-72 degrees Celsius, and under load it easily exceeded 90 degrees. We do not recommend repeating our experiment, since the chances of irreversibly damaging or destroying the card are very high even with powerful cooling. However, ordinary users do not engage in such overclocking, and no warnings can stop real overclocking enthusiasts in any case.

We managed to overclock the memory to 1242 (2484) MHz by increasing the supply voltage to 2.05V. This raised the throughput from 72 to 79.5 GB / sec., which cannot be called a serious increase; in any case, it will not be the main influence on performance in games.

At 999/2484 MHz, the card successfully passed 2-3 test cycles in 3DMark06 and 3DMark Vantage, however, testing in games had to be abandoned due to freezes that began after a while. The instability was overcome only by lowering the core frequency to 972 MHz, and the memory frequency to 1224 (2448) MHz. In this mode, the card successfully passed the entire test cycle.

Detailed information about the power consumption of overclocked graphics cards is of great interest, and we decided to find out how much the ATI Radeon HD 3870 can consume, the graphics core of which is overclocked to 1 GHz. For this, a specially equipped test bench was used with the following configuration:

• Processor AMD Athlon 64 FX-55 (2.6 GHz)
• EPoX EP9-NPA+ SLI motherboard (Nvidia nForce4 SLI)
• Memory PC3200 (2×512 MB, 200 MHz)
• Western Digital Raptor WD360ADFD hard drive (36GB, SATA-150, 16MB buffer)
• Chieftec ATX-410-212 Power Supply (Rated Power 410 Watts)
• Microsoft Windows Vista Ultimate 32-bit
• Futuremark PCMark05 Build 1. 2.0
• Futuremark 3DMark06 Build 1.1.0

In order to create a load in 3D mode, we used the first SM3.0/HDR test of the 3DMark06 package, run in a loop at a resolution of 1600×1200 with forced FSAA 4x and AF 16x. The peak 2D mode was emulated using the PCMark05 2D Transparent Windows test. As a result, the following data were obtained:

Under extreme overclocking, PowerPlay is disabled and the core frequency does not decrease dynamically depending on the operating mode, which explains the high performance in 2D and peak 2D modes. But the last figure obtained in 3D mode is of the greatest interest: our guinea pig managed to overtake the ATI Radeon HD 3870 X2 by 10 watts, and the load on both 12V power circuits significantly exceeded the maximum allowable 75 watts! Here it is, the price of extreme overclocking with an increase in the GPU supply voltage.

At the same time, we doubt that the performance level of the overclocked ATI Radeon HD 3870 will be comparable to that of the ATI Radeon HD 3870 X2. This is another argument in favor of the thesis that the linear increase in clock frequencies is a dead end path of development, and the future belongs to multi-core solutions. Whether they will be homogeneous or heterogeneous — time will tell.

Test benches with the following configuration were used to test the overclocked ATI Radeon HD 3870:

• Intel Core 2 Extreme X6800 (3.0 GHz, FSB 333 MHz x 9)
• DFI LANParty UT ICFX3200-T2R/G (ATI CrossFire Xpress 3200) Motherboard for ATI Radeon HD
• Asus P5N32-E SLI (Nvidia nForce 680i SLI) motherboard for Nvidia GeForce
• Memory Corsair TWIN2X2048-8500C5 (2×1 GB, 1066 MHz, 5-5-5-15, 2T)
• Maxtor MaXLine III 7B250S0 hard drive (250GB, SATA-150, 16MB buffer)
• Enermax Galaxy DXX EGX1000EWL Power Supply (Rated Power 1000 Watts)
• Dell 3007WFP Monitor (30”, maximum resolution 2560×1600@60Hz)
• Microsoft Windows Vista Ultimate 32-bit
• Nvidia GeForce 175. 16 WHQL for GeForce 8/9
• ATI Catalyst 8.5 for Radeon HD

According to our testing methodology, the drivers were tuned in such a way as to provide the highest possible quality of texture filtering and minimize the impact of software optimizations used by default by both ATI and Nvidia. In order to achieve the highest image quality, the anti-aliasing options for transparent textures were enabled: Adaptive Anti-Aliasing/Multi-sampling for ATI Catalyst and Antialiasing — Transparency: Multisampling for Nvidia GeForce. As a result, the driver settings looked like this: ATI Catalyst:

• Catalyst A.I.: Standard
• Mipmap Detail Level: High Quality
• High Quality AF: On
• Wait for vertical refresh: Always Off
• Enable Adaptive Anti-Aliasing: On/Quality
• Method: Multi-sampling
• Temporal Anti-Aliasing: Off
• Other settings: default

Nvidia GeForce:

• Texture filtering – Quality: High quality
• Texture filtering – Trilinear optimization: Off
• Texture filtering – Anisotropic sample optimization: Off
• Vertical sync: Force off
• Antialiasing — Gamma correction: On
• Antialiasing — Transparency: Multisampling
• Other settings: default

We used the following set of games and synthetic tests as test software:

S. T.A.L.K.E.R.: Shadow of Chernobyl

Third-Person 3D Shooters

• Lost Planet: Extreme Condition
• Tomb Raider: Legend

RPG

• Hellgate: London
• The Elder Scrolls IV: Oblivion

Strategies

• Company of Heroes: Opposing Fronts
• Command & Conquer 3: Tiberium Wars
• World in Conflict

Semi-synthetic Benchmarks

• Futuremark 3DMark06
• Futuremark 3DMark Vantage

Each of the games included in the test set of software was tuned by us to the highest possible level of image detail, and we used only the tools of the game itself, which were initially available to a simple user — no configuration files were manually modified. The exception was the shooter Enemy Territory: Quake Wars, which had the built-in fps limiter disabled, fixed at around 30 frames per second. Games that can take advantage of DirectX 10 features have been tested in this mode. In addition to ATI Radeon HD 3870 (in normal and overclocked modes), the following cards took part in testing:

• ATI Radeon HD 3870 X2 (2xRV670, 825/825/1800MHz, 640sp, 32tmu, 32rop, 2×256-bit, 2x512MB GDDR3)
• Nvidia GeForce 9800 GTX (G92, 675/1688/2200MHz, 128sp, 32tmu, 16rop, 256-bit, 512MB GDDR3)
• Nvidia GeForce 8800 GT 512MB (G92, 600/1500/1800MHz, 112sp, 28tmu, 16rop, 256-bit, 512MB GDDR3)
• Nvidia GeForce 9600 GT (G94, 650/1625/1800MHz, 64sp, 16tmu, 16rop, 256-bit, 512MB GDDR3)

Tested in 1280×1024/9 resolutions60, 1600×1200 and 1920×1200. If the game did not support the 16:10 screen format, the 1920×1440 resolution was used instead. In all cases, where it was possible without disabling the effects of SM3.0/4.0/HDR, the mode was used, which implies the use of MSAA 4x in addition to the standard 16x anisotropic filtering. Activation of anti-aliasing and anisotropic filtering was carried out either by the means of the game itself, or, in their absence, was forced using the appropriate settings of the ATI and Nvidia drivers. To obtain performance data, either the tools built into the game were used, with the obligatory recording of original demos, if possible, or, in their absence, the Fraps 2.9 utility.1 in manual mode. Whenever possible, data were recorded not only on the average, but also on the minimum productivity.

• Battlefield 2142

The game doesn’t support 16:10 screen format by default, so instead of 1920×1200 resolution we have to use 1920×1440 resolution with 4:3 aspect ratio.

The extreme version of the ATI Radeon HD 3870 is quite solidly ahead of its counterpart operating at nominal frequencies: for example, at 1600×1200, the average performance gain is 20%, while the minimum one has grown by as much as 35%.

At 1920×1200 resolution, the gain is slightly less, but there is a certain benefit from it — in some situations the player could feel more confident. However, Nvidia solutions are still able to provide better performance.