The remarquable thing with this GTX 580 FTW HC2 is the cooling performance of the Hydro Copper waterblock. The guys at HotHardware have done a stress test with FurMark (res: 2560×1600) and the GPU temperature didn’t exceed 35°C! This is just impressive!

If you look at the following results, 35°C is lower than most of the cards at idle state:

I just stress-tested with FurMark 1.9.0 my EVGA GTX 580 SC, overclocked at 900MHz with a Vcore (or VDDC or GPU voltage) of 1.100V. At first glance, the card was stable, no visual artifacts, a very high GPU temperature (96°C) with an insane power consumption: 385W for the card alone!

I even ran 3DMark11 without problem: P6534 and X2218.

I also launched Unigine Heaven 2.1 Direct3D 11 test (1920×1080, tessellation normal, 16X aniso, 4X MSAA): FPS: 52.9, Scores: 1333

But in Heaven, I noticed few visual artifacts. Then I launched an artifact scanning with the latest EVGA OC Scanner 1.6.0 and actually, 900MHz was a bit too high. Let’s see how to find stable and robust overclocking settings thanks to OC Scanner.

Test bench
- CPU: Core i7 960 @ 3.2GHz
- RAM: 4GB DDR3 Corsair Dominator
- Motherboard: GIGABYTE X58-A UD5
- Windows 7 64-bit
- Graphics drivers: R266.58
- PSU: Corsair AX1200

First thing to do for overclocking a GTX 580 is to disable the over current protection (OCP). I simply used the latest GPU-Z 0.5.1 with the /gtx500ocp option (for information, FurMark 1.9.0 makes this step simpler).

To detect artifacts with OC Scanner, I recommend to enable the heavy mode, in the settings box:

Then set the high resolution like 1920×1080 fullscreen:

Okay OC Scanner is ready for artifacts scanning. Press the Start stress test button and you should see this rendering:

The most important line is the last one:

Artifacts: cur:0 - max:0 - total:0

In short, as long as all values are 0, that means you overclocking is stable. As soon as one value is greater than 0, your overclocking settings are too high:

Here are the different steps of my GTX 580 overclocking (I set the core clock and core voltage with Afterburner 2.1.0 beta 7 –with Afterburner, 1.138V is the max GPU voltage we can set):

• 1 – GPU @ 890MHz, VDDC @ 1.113V –> 4 artifacts dectected after 1 min
• 2 – GPU @ 890MHz, VDDC @ 1.125V –> 2 artifacts dectected after 2 min
• 3 – GPU @ 890MHz, VDDC @ 1.138V –> 2 artifacts dectected after 7 min
• 4 – GPU @ 880MHz, VDDC @ 1.138V –> 1 artifact dectected after 4 min
• 5 – GPU @ 880MHz, VDDC @ 1.125V –> 1 artifact dectected after 15 min
• 6 – GPU @ 880MHz, VDDC @ 1.113V –> 1 artifact dectected after 3 min
• 7 – GPU @ 880MHz, VDDC @ 1.088V –> 5 artifacts detected after 1 min
• 8 – GPU @ 870MHz, VDDC @ 1.088V –> 2 artifacts detected after 1 min
• 9 – GPU @ 870MHz, VDDC @ 1.100V –> 1 artifact detected after 4 min
• 10 – GPU @ 870MHz, VDDC @ 1.125V –> no artifact detected after 15 min

A you can see, OC Scanner is able to detect very small artifacts: few pixels in a full HD scene. Those few pixels are undetectable with our eyes in 3D scenes and that explains why I thought my OC was stable at 900MHz!

With these settings {870MHz, 1.125V}, the GPU temperature reaches 93°C and the power consumption of the testbed is 525W under OC Scanner. The power consumption of the testbed at idle is 120W and the PSU efficiency factor is around 0.9: (525-120) * 0.9 = 364W for the EVGA GTX 580 SC alone.

I validated these overclocking settings with a FurMark 1.9.0 burn-in test session (1920×1080 fullscreen, 15min): GPU: 98°C, fan speed 85% (3810RPM), total power consumption: 540W: (540-120)*0.9 = 378W for the GTX 580 alone!

378W ? Is it not too much? Sure my friend, it’s too much, this is even a very, ultra high value for the power consumption of a graphics card . NVIDIA says the TDP of a GTX 580 is… 240W. The GTX 580 has one 8-pin + one 6-pin power connectors leading to a max of 300W for the board (see here for more details: Maximum Power Consumption of Graphics Card Connectors). Then we are completely out of range with 378W! That’s why you have to have a solid PSU, a solid motherboard to overclock a high-end graphics card.

3DMark11 and Unigine Heaven scores with the new OC settings {870MHz, 1.125V}:
- 3DMark11: P6391
- Unigine Heaven 2.1: FPS: 51.7 , Scores: 1303

With a GPU core clocked at 870MHz and fed with a voltage of 1.125V, EVGA’s GTX 580 SC is stable. Finding the best overclocking settings for a GPU is a rather long process. And if you want to be sure that your OC settings are stable, I recommend you to burn your GPU with FurMark, or EVGA OC Scanner for at least one hour. After what, play with your favorite game to be sure the OC is okay.

EVGA’s GTX 580 SC deserves this nice badge:

A detail about RTI tc. RTI tc is the Return To Idle time constant. This time constant is the time required for a reduction of 63% of the GPU temperature after a full loading (98°C is 100% and 42°C is the 0%). In our case, the system took 29 seconds to reach 62°C (decreasing of 63%). We can determine a kind of cooling factor:
(98-62) / 29 = 1.24 °C/sec.

FurMark, the popular GPU burn-in utility, is, at last, available in version 1.9.0.

1 – FurMark 1.9.0 Download

Webmasters: hotlinking is not allowed (it will cause an error message), please use the post url as download link.

Translation files
Just copy the following files in the localization folder of FurMark:

• Chinese by nhldp
• Polish by cienislaw

Remark: some strings are not translated: they will be fixed in the next update of FurMark.

You can also download FurMark 1.9.0 from FilePlanet.

2 – FurMark 1.9.0 Release Highlights

First thing, FurMark is still an OpenGL 2.0 GPU stress test. That means all available GPUs can be tested with FurMark including:
- NVIDIA GeForce 6 to latest GeForce GTX 500
- AMD Radeon HD 2000 to latest Radeon HD 6000
- Intel Sandy Bridge GPUs HD 2000 / HD 3000
- S3 Graphics Chrome 400, 500
- and all other models I forget.

You can even stress test your GPU under Linux using Wine, waiting for the Linux version of FurMark

Okay, now let’s see the new features of FurMark 1.9.0.

The graphics workload is heavier than in FurMark 1.8.2 to follow the GPU power processing of recent graphics cards. For example, a GTX 480 has a score of 6478 points (108FPS) in FurMark 1.8.2 in 1920×1080 while in FurMark 1.9.0 the score of the same card is only 1769 points (29FPS).

FurMark 1.9.0 adds a dynamic background instead of an static image. But don’t panic, this is an option, and if you want to find your favorite dragon, just uncheck the dynamic background. I also added a mode with two interlaced furry torus and a moving camera.

To make benchmark process simpler, there are now two presets that allow to submit an online score: Preset:1080 and Preset:720. Just click on a preset button and the benchmark starts with predefined settings:

• Preset:1080 : 1920×1080 fullscreen, no AA, Post FX disabled, dynamic background, burn-in enabled.
• Preset:720 : 1280×720 fullscreen, no AA, Post FX disabled, dynamic background, burn-in enabled.

Two GPU monitoring utilities are now embedded in FurMark: GPU-Z and GPU Shark. For GPU-Z, I added the option to disable the OCP (over current protection) on GeForce GTX 500 cards.

OCP is not available on every GeForce GTX 500. For example, OCP is actived on EVGA GTX 580, while this protection has been disabled by ASUS on its GTX 560 Ti DC2.

Remark: NVIDIA power limiter / OCP is based on an application black list. FurMark 1.8.2 is black listed and you can easily see the impact of the OCP on a GTX 580: with OCP, the FurMark 1.8.2 score is 4256 points (71FPS) in 1920×1080 while with OCP disabled, the score jumps to 7772 points (130FPS). Currently the new FurMark 1.9.0 is not yet black listed so you won’t see differences between OCP enabled and disabled…

Another cool little feature is an audio alarm for the GPU temperature. You can set the max GPU temperature in the settings box:

As soon as the GPU temperature exceeds the max value, an atomic alert siren will be played.

Another important change is related to spikes in GPU temperature monitoring. These spikes occur when several monitoring utilities try to access to the same hardware at the same time. FurMark 1.9.0 includes a spike filtering module that clamps spikes when difference between two consecutive temperature exceed a threashold.

Multi-GPU: SLI and CrossFire

And multi-GPU support like NVIDIA SLI or AMD CrossFire? Of course FurMark supports both technolgies but you have to apply some little tricks to properly enable the multi-GPU support. See these articles for more details:

• How to Enable SLI and CrossFire Support for OpenGL Applications
• MSI Kombustor: SLI and CrossFire Support

What is FurMark?

FurMark is an OpenGL-based GPU stress test utility. It makes it possible to push the GPU to the max in order to test the stability of the graphics card (and the PSU too!) leading to maximal GPU and VRM temperatures. That’s why FurMark is often used by overclockers and graphics cards fanatics to validate an overclocking, to test a new VGA cooler or to check the max power consumption of a video card.

FurMark 1.9.0 changelog

• New: support of all latest GeForce GTX 500 and Radeon HD 6000 Series…
• New: added /gtx500ocp checkbox for GPU-Z to disable GTX 500 OCP (over current protection).
• New: GPU-Z 0.5.1 and GPU Shark 0.4.1 are now embedded in FurMark.
• New: added programmble GPU temperature alarm. When GPU temperature exceeds a threshold, an alarm sound is played.
• New: added warming-up step to benchmarking.
• New: improved OSD (on-screen display data) readability (key V).
• New: added GPU monitoring in 3D window.
• New: added GPU-Z full GPU power consumption data (power, voltage and current) if available.
• New: caution message displayed before every burn-in test.
• New: score submission is now limited to two presets: Preset:1080 and Preset:720.
• New: increased graphics workload.
• New: added animated background image option based on iq’s code.
• New: added animated camera option with multiple furry torus.
• New: two burn-in modes are now available: normal burn-in mode and Xtreme burn-in mode
• Bugfix: spikes in temperature graphs are now filtered.
• Bugfix: many minor bugs…

I’d like to thank ASUS, Corsair and Sapphire for their hardware support.

At the CEBIT 2011, AMD has demonstrated the superiority of its LIano APU vs Intel Sandy Bridge processor. Among the applications used, there was FurMark 1.8.2. In 1280×1024, AMD LIano (1.18GHz) plays FurMark at around 18FPS while Intel Sandy Bridge (Core i7 2630QM @ 2GHz, with a HD 3000 GPU) runs FurMark at around 10FPS.

And when AMD enabled the displacement mapping option, AMD was happy to show the rendering bug of Sandy Bridge…

But this bug is strange, because Intel has recently updated its drivers with optimizations for some Geeks3D’s apps like FurMark or FluidMark (see HERE for more details). Maybe AMD has used an older driver. I’ll try to test the Sandy Bridge processor with my OpenGL app asap.

[source] | [via]

There is an odd thing with video: it looks like both GPUs of the HD 6990 are loaded by FurMark in… windowed mode. CrossFire on gaming cards works only in fullscreen mode (see HERE). So why both GPUs are stressed in windowed mode?

GPU temperatures: 82°C for one GPU and 87°C for the second one. I don’t know why but I’m nearly sure that in fullscreen mode, the temperatures can reach 90°C and more…

As you can see on the video, there is a little lighting problem with FurMark 1.9.0 and Radeon cards: the furry torus is not lit properly (reversed). During the dev of FurMark 1.9.0 I never noticed this bug (maybe a change in recent Catalyst…). It doesn’t matter, I will check it and fix it for the next update of FurMark.

Another test by Linus Tech Tips, this time focused on the new GeForce GTX 550 Ti cooleing system: MSI’s Cyclone II VGA cooler vs NVIDIA reference cooler.

Yep these hot days are the perfect moment to release some updates of FurMark: FurMark 1.9.1 and FurMark 1.8.5. Both versions of that heat generator are maintenance releases, read no new features but support of latest graphics cards or minor changes.

Since the branch 1.8.x of FurMark is still widely used and downloaded, I decided to update it (FurMark 1.8.2 has been released more than one year ago).

FurMark 1.8.5 Download

Webmasters: hotlinking is not allowed (that will cause an error message), please use the post url as download link.

What is FurMark?

FurMark is an OpenGL-based GPU stress test utility (also called GPU burn-in test). It makes it possible to push the GPU to the max in order to test the stability of the graphics card (and the PSU too!) leading to maximal GPU and VRM temperatures. That’s why FurMark is often used by overclockers and graphics cards fanatics to validate an overclocking, to test a new VGA cooler or to check the max power consumption of a video card.

Here are some links that show how FurMark can be used to torture graphics cards and PSUs:

• Thermaltake Toughpower Grand 850W PSU: The Torture Test
• ASUS ROG Matrix GTX 580 Platinum: Overclocking and Power Consumption
• SAPPHIRE Radeon HD 6970: Overclocking and Power Consumption
• ASUS GeForce GTX 560 DirectCU II TOP: Power consumption, temperature and overclocking
• ASUS GeForce GTX 570 DirectCU II: Power Consumption and Overclocking

Multi-GPU: SLI and CrossFire

Both multi-GPU technologies (NVIDIA SLI or AMD CrossFire) are supported by FurMark but you have to use some little tricks to properly enable the multi-GPU support. See these articles for more details:

• How to Enable SLI and CrossFire Support for OpenGL Applications
• MSI Kombustor: SLI and CrossFire Support

FurMark 1.9.1 changelog

• Maintenance release.
• Update: added support of recent AMD Radeon and NVIDIA GeForce graphics cards.
• Update: GPU-Z 0.5.4 and GPU Shark 0.5.1
• Change: added a workaround for a bug in AMD Catalyst GLSL compiler (GL2 for() loop –old bug now present in a beta version of Cat 11.2 and / or Cat 11.3). This bug led to a wrong lightning of the furry donut.

FurMark 1.8.5 changelog

• Maintenance release.
• Improved score submission.
• Updated the number of window resolutions available.
• Change: updated with the latest version of ZoomGPU (support of recent NVIDIA GeForce and AMD Radeon graphics cards).
• Bugfix: spikes in temperature graphs are now filtered. See here for more information: FurMark and Spikes in Temperature Monitoring Graphs.
• Change: minimizing window is no longer possible in windowed benchmark mode.
• Removed the FurMark 1.7.0 mode.

28dBa is barely audible. Nice VGA hack!

Read the complete article here: AMD’s flagship HD6990: is silent air cooling possible?

At idle, the GPU temperature is around 32°C without noise (fan speed: 20%). Stressed by FurMark 1.9.1, the Devil 13 shows a GPU temperature of 67°C with slight noise due to the rotation speed of the fan. According to the reviewer, the noise is obvious but sustainable.

Hereunder, the power consumption test:

In order to stress test and burn the new graphics cards you will receive for Xmas, here is an update of FurMark. FurMark 1.9.2 brings some bugfixes, latest versions of GPU-Z and GPU Shark and a new benchmark mode: the burn-in benchmark.

The aim of this benchmark is simple: get the highest score with the lowest GPU temperature. The benchmark lasts 15 minutes at a full HD resolution (1920×1080). Here is an example of score: Radeon HD 6970 (3630 points, 96°C).

What is FurMark?

FurMark is an OpenGL-based GPU stress test utility (also called GPU burn-in test). It makes it possible to push the GPU to the max in order to test the stability of the graphics card (and the PSU too!) leading to maximal GPU and VRM temperatures. That’s why FurMark is often used by overclockers and graphics cards fanatics to validate an overclocking, to test a new VGA cooler or to check the max power consumption of a video card.

Here are some links that show how FurMark can be used to torture graphics cards and PSUs:

• Thermaltake Toughpower Grand 850W PSU: The Torture Test
• ASUS ROG Matrix GTX 580 Platinum: Overclocking and Power Consumption
• SAPPHIRE Radeon HD 6970: Overclocking and Power Consumption
• ASUS GeForce GTX 560 DirectCU II TOP: Power consumption, temperature and overclocking
• ASUS GeForce GTX 570 DirectCU II: Power Consumption and Overclocking

Multi-GPU: SLI and CrossFire

Both multi-GPU technologies (NVIDIA SLI or AMD CrossFire) are supported by FurMark but you have to use some little tricks to properly enable the multi-GPU support. See these articles for more details:

• How to Enable SLI and CrossFire Support for OpenGL Applications
• MSI Kombustor: SLI and CrossFire Support

FurMark 1.9.2 Changelog

• New: support of latest NVIDIA GeForce and AMD Radeon cards (ZoomGPU 1.6.1).
• New: added new benchmark: the burn-in benchmark (1920×1080 only, duration:15 min). Goal: best score with the lowest GPU temperature.
• New: user can now submit a score with a login.
• Update: GPU-Z 0.5.6 and GPU Shark 0.5.4
• Change: removed the /gtx500ocp checkbox.
• Bugfix: fixed a deadlock that hung FurMark sometimes at the end the tests.

ASUS uses some popular stress test and benchmarking tools for validating the graphics cards (FurMark, 3DMark and Unigine Heaven according to the pictures).

By the way, I just received a Radeon HD 7770 DC TOP, I hope to publish a review next week!

• Page 1: 1920×1080 scores
• Page 2: 1280×720 scores
• Page 3: 1280×1024 scores

Here are the charts of FurMark scores for AMD Radeon and NVIDIA GeForce graphics cards (and Intel HD Graphics GPUs as well). These scores can be used as reference for checking your FurMark score with FurMark 1.9.x and FurMark 1.10.x (will be released shortly, currently in testing…). Each entry includes the model of the graphics card with clock speeds, the TPC (Total Power Consumption) of the X58 testbed at idle and under full load, the max GPU temperature and the graphics driver version.

FurMark is an OpenGL 2 benchmark. Almost all existing graphics cards can be benchmarked with FurMark: GeForce 6 / 7/ 8 / 9 / GTX200 / GTX400 / GTX500 / GTX600, Radeon X1900 / HD2000 / HD3000 / HD4000 / HD5000 / HD6000 / HD7000, Intel HD Graphics 2000 / 3000 / 4000.

These charts will be updated as regularly as possible with new graphics cards or new graphics drivers (only if there are significant changes in performance).

Three kind of scores are available on these charts:

• Preset:1080: 1920×1080 fullscreen, 0X MSAA, 60 seconds, dynamic background, burn-in mode checked in the Settings box
• Preset:720: 1280×720 fullscreen, 0X MSAA, 60 seconds, dynamic background, burn-in mode checked in the Settings box
• Custom preset: 1280×1024 fullscreen, 0X MSAA, 60 seconds, dynamic background, burn-in mode checked in the Settings box

1 – Preset:1080

Settings: 1920×1080 fullscreen, 0X MSAA, 60 seconds, dynamic background, burn-in test mode.

- Sapphire Radeon HD 6970 CF (GPU@880MHz, mem@1375MHz, max GPU temp:78°C, two GPUs / CrossFire ON), TPC: 87W(PSU1) + 33W(PSU2) /430W(PSU1) + 240W(PSU2) (on H67 testbed v2), Catalyst 12.3 whql (Win 7 64-bit) - 4704 points (78 FPS), PowerTune: +20%

- Sapphire Radeon HD 6970 CF (GPU@880MHz, mem@1375MHz, max GPU temp:78°C, two GPUs / CrossFire ON), TPC: 87W(PSU1) + 33W(PSU2) /352W(PSU1) + 176W(PSU2) (on H67 testbed v2), Catalyst 12.3 whql (Win 7 64-bit) - 3505 points (58 FPS), PowerTune: 0%

- MSI Radeon HD 7970 (GPU@925MHz, mem@1375MHz, max GPU temp:68°C), TPC:105W/383W, Catalyst 12.3 whql (Win 7 64-bit) - 3363 points (56 FPS), PowerTune: +20%

- MSI Radeon HD 7970 (GPU@925MHz, mem@1375MHz, max GPU temp:79°C), TPC:75W/346W (on H67 testbed), Catalyst 12.3 whql (Win 7 64-bit) - 3361 points (56 FPS), PowerTune: +20%

- EVGA GeForce GTX 680 (GPU@1005MHz, mem@3004MHz, max GPU temp:74°C), TPC:101W/312W, GPU power: 115% TDP, R301.24 beta (branch r301_07-12, Win 7 64-bit) - 2699 points (44 FPS)

- MSI Radeon HD 7970 (GPU@925MHz, mem@1375MHz, max GPU temp:76°C), TPC:75W/280W (on H67 testbed), Catalyst 12.3 whql (Win 7 64-bit) - 2582 points (43 FPS), PowerTune: 0%

- MSI Radeon HD 7970 (GPU@925MHz, mem@1375MHz, max GPU temp:68°C), TPC:105W/339W, Catalyst 12.3 whql (Win 7 64-bit) - 2582 points (43 FPS), PowerTune: 0%

- EVGA GeForce GTX 580 Classified Ultra (GPU@900MHz, mem@2106MHz, max GPU temp:87°C), TPC:116W/530W, R301.24 beta (branch r301_07-12, Win 7 64-bit) - 2403 points (40 FPS)

- Sapphire Radeon HD 6970 (GPU@880MHz, mem@1375MHz, max GPU temp:78°C), TPC:78W/352W (on H67 testbed), Catalyst 12.3 whql (Win 7 64-bit) - 2402 points (40 FPS), PowerTune: +20%

- EVGA GeForce GTX 580 SC (GPU@797MHz, mem@2025MHz, max GPU temp:91°C), TPC:85W/452W (on H67 testbed), R263.09 whql (branch r260_uda_final-58, Win 7 64-bit) - 2144 points (35 FPS)

- MSI Radeon HD 5770 Hawk CF (GPU@875MHz, mem@1200MHz, max GPU temp:69°C, two GPUs / CrossFire ON), TPC:120W/350W, Catalyst 12.3 whql (Win 7 64-bit) - 2135 points (35 FPS)

- AMD Radeon HD 5870 (GPU@850MHz, mem@1200MHz, max GPU temp:84°C), TPC:102W/320W, Catalyst 12.3 whql (Win 7 64-bit) - 2130 points (35 FPS)

- MSI GeForce GTX 580 Lightning (GPU@772MHz, mem@2004MHz, max GPU temp:97°C), TPC:125W/564W, R263.09 whql (branch r260_uda_final-58, Win 7 64-bit) - 2077 points (34 FPS)

- Sapphire Radeon HD 5850 (GPU@765MHz, mem@1125MHz, max GPU temp:66°C), TPC:110W/276W, Catalyst 12.3 whql (Win 7 64-bit) - 1805 points (30 FPS)

- Sapphire Radeon HD 6970 (GPU@880MHz, mem@1375MHz, max GPU temp:78°C), TPC:78W/286W (on H67 testbed), Catalyst 12.3 whql (Win 7 64-bit) - 1776 points (29 FPS), PowerTune: 0%

- EVGA GeForce GTX 480 (GPU@700MHz, mem@1848MHz, max GPU temp:89°C), TPC:105W/405W (on H67 testbed), R296.10 whql (branch r295_43-27, Win 7 64-bit) - 1758 points (29 FPS)

- EVGA GeForce GTX 295 CO-OP (GPU@576MHz, mem@1008MHz, max GPU temp:73°C, two GPUs enabled / SLI ON), TPC:130W/414W, R301.24 beta (branch r301_07-12, Win 7 64-bit) - 1754 points (29 FPS)

- Sapphire Radeon HD 6870 (GPU@900MHz, mem@1050MHz, max GPU temp:82°C), TPC:104W/277W, Catalyst 12.3 whql (Win 7 64-bit) - 1668 points (27 FPS)

- MSI GeForce GTX 470 (GPU@607MHz, mem@1674MHz, max GPU temp:93°C), TPC:124W/347W, R296.10 whql (branch r295_43-27, Win 7 64-bit) - 1424 points (23 FPS)

- MSI GeForce GTX 470 (GPU@607MHz, mem@1674MHz, max GPU temp:93°C), TPC:85W/294W (on H67 testbed), R296.10 whql (branch r295_43-27, Win 7 64-bit) - 1422 points (23 FPS)

- MSI GTX 460 Cyclone 768D5/OC (GPU@729MHz, mem@1804MHz, max GPU temp:67°C), TPC:100W/268W, R301.24 beta (branch r301_07-12, Win 7 64-bit) - 1187 points (19 FPS)

- MSI Radeon HD 5770 Hawk (GPU@875MHz, mem@1200MHz, max GPU temp:68°C), TPC:110W/224W, Catalyst 12.3 whql (Win 7 64-bit) - 1078 points (17 FPS)

- ASUS Radeon HD 7770 DirectCU TOP (GPU@1120MHz, mem@1150MHz, max GPU temp:59°C), TPC:75W/182W, Catalyst 12.3 whql (Win 7 64-bit) - 1036 points (17 FPS)

- MSI GeForce GTX 580 Lightning (GPU@772MHz, mem@2004MHz, max GPU temp:72°C), TPC:133/374W, R296.10 whql (branch r295_43-27, Win 7 64-bit) - 1025 points (17 FPS) – GPU is throttled down (at 192MHz with R296.10, R301.24)!

- EVGA GeForce GTX 280 (GPU@601MHz, mem@1107MHz, max GPU temp:79°C, TPC:125W/322W), R301.24 beta (branch r301_07-12, Win 7 64-bit) - 919 points (15 FPS)

- EVGA GeForce GTX 295 CO-OP (GPU@576MHz, mem@1008MHz, max GPU temp:71°C, one GPU enabled / SLI OFF, TPC:130W/294W), R301.24 beta (branch r301_07-12, Win 7 64-bit) - 879 points (14 FPS)

- MSI GeForce GTX 260 (GPU@621MHz, mem@1080MHz, max GPU temp:68°C), R301.24 beta (branch r301_07-12, Win 7 64-bit) - 739 points (12 FPS)

- MSI Radeon HD 4850 Hybrid Frozr (GPU@640MHz, mem@993MHz, max GPU temp:82°C), TPC:112W/194W, Catalyst 12.3 whql (Win 7 64-bit) - 577 points (9 FPS) (GPU throttled down by Catalyst?)

- HIS Radeon HD 3870 (GPU@777MHz, mem@1126MHz, max GPU temp:93°C), TPC:105W/230W, Catalyst 12.3 whql (Win 7 64-bit) - 576 points (9 FPS)

- NVIDIA GeForce 8800 GTX (GPU@576MHz, mem@900MHz, max GPU temp:77°C), TPC:133W/261W, R296.10 whql (branch r295_43-27, Win 7 64-bit) - 481 points (8 FPS)

- ASUS GeForce 7950 GX2 (GPU@500MHz, mem@600MHz, max GPU temp:95°C, two GPUs enabled / SLI ON), TPC:128W/249W, R296.10 whql (branch r295_43-27, Win 7 64-bit) - 415 points (6 FPS)

- Zotac GeForce GT 240 (GPU@550MHz, mem@790MHz, max GPU temp:65°C), TPC:88W/144W, R296.10 whql (branch r295_43-27, Win 7 64-bit) - 356 points (5 FPS)

- MSI GeForce 9600 GT Diamond (GPU@650MHz, mem@850MHz, max GPU temp:46°C), R301.24 beta (branch r301_07-12, Win 7 64-bit) - 349 points (5 FPS)

- ASUS GeForce 7950 GX2 (GPU@500MHz, mem@600MHz, max GPU temp:84°C, one GPU enabled / SLI OFF), TPC:128W/201W, R296.10 whql (branch r295_43-27, Win 7 64-bit) - 208 points (3 FPS)

- ASUS Radeon HD 2400 (GPU@525MHz, mem@400MHz, max GPU temp:40°C), TPC:80W/117W, Catalyst 12.3 whql (Win 7 64-bit) - 41 points (1 FPS)

• Page 1: 1920×1080 scores
• Page 2: 1280×720 scores
• Page 3: 1280×1024 scores

Et de quatre! Like GPU Shark, GPU Caps Viewer and EVGA OC Scanner X, FurMark has been updated with the full support of NVIDIA’s GeForce GTX 680 (clock speeds and power monitoring). Alongside the GTX 680 support, true clock speeds are now displayed (useful with End-Of-Life GTX 580…) and GPU monitoring code runs now in its own system thread. GPU Shark and GPU-Z have updated to their latest versions. For more information see the changelog at the end of the post.

Multi-GPU: SLI and CrossFire

Both multi-GPU technologies (NVIDIA SLI or AMD CrossFire) are supported by FurMark but you have to use some little tricks to properly enable the multi-GPU support. See these articles for more details:

• How to Enable SLI and CrossFire Support for OpenGL Applications
• MSI Kombustor: SLI and CrossFire Support

FurMark 1.10.0 Changelog

• New: added list of detected GPUs in the main interface.
• New: stress test options are now in the settings box.
• New: support of NVIDIA GeForce GTX 680 (clock speeds and power monitoring).
• New: support of latest AMD Radeon (HD 7900, HD 7800 and HD 7700) cards.
• New: added resolutions 4096×1152 and 6144×1152 for 2048×1152 monitors.
• New: display of the true clock of the GPU when available (NVIDIA only).
• Update: GPU monitoring moved in a separate system thread.
• Update: GPU data monitoring report updated (Log GPU data in Settings box).
• Update: minor changes in the way GPU monitoring information is displayed.
• Update: ZoomGPU 1.8.2 (GPU monitoring library).
• Update: GPU-Z 0.6.0 and GPU Shark 0.6.1

Just a quick note about the GPU throttling with the GeForce GTX 580 (extreme GTX 580 cards like EVGA GTX 580 Classified or ASUS GTX 580 Matrix are not concerned by this article). In recent versions of FurMark, EVGA OC Scanner X and MSI Kombustor, I added the monitoring of true GPU clock speed for NVIDIA cards. The true GPU clock speed is useful with graphics cards like the GeForce GTX 580 because this card is severely throttled down when a TDP-app (a new word coming from the GTX 680 launch) like FurMark is running. For example, with EVGA’s GTX 580 SC, under a typical gaming situation, the GPU runs at 797 MHz. When FurMark is running, the GPU clock speed is throttled down at 398 MHz (GPU-Z still displays 797MHz):

Same thing in OC Scanner X. During few seconds, the GPU runs at full speed and after it’s throttled down:

In MSI Kombustor, I added the throttling indicator in the GPU list, on the top-right part of the screen:

Fortunately, the brand-new GeForce GTX 680 (I really love this card!) nicely handles FurMark-like applications then this short note about the throttling of the GTX 580 will be shortly irrelevant!

A maintenance release of FurMark is ready for stressing and overheating your overclocked GPUs! FurMark 1.10.1 brings bugfixes and updates (ZoomGPU 1.8.3 with GeForce GTX 690 and GTX 670, GPU-Z 0.6.2 and GPU Shark 0.6.3). See the changelog for more details.

FurMark 1.10.1 Changelog

• Update: updated gpu monitoring code with memory usage (when available).
• Update: GPU-Z 0.6.2 and GPU Shark 0.6.3.
• Update: ZoomGPU 1.8.3 (GPU monitoring library) with support of latest GTX 690 and GTX 670.
• Bugfix: in gpu monitoring, max value of GTX 600 power was not saved.
• Bugfix: minor bugs fixed in score submission.
• Bugfix: gpu monitoring thread was not stopped when FurMark initialization failed.
• Bugfix: gpu monitoring xml report was not well-formed (closing tag missing).

A maintenance release of FurMark is ready for your GPU burning sessions. FurMark 1.10.2 brings the support of GeForce GTX 660 / GTX 650, as well as some updates (ZoomGPU 1.8.7, GPZU Shark 0.6.6 and GPU-Z 0.6.4).

FurMark 1.10.2 Changelog

• Update: support of NVIDIA GeForce GTX 660 and GeForce GTX 650.
• Update: ZoomGPU 1.8.7 (GPU monitoring library).
• Update: GPU-Z 0.6.4 and GPU Shark 0.6.6

FurMark 1.10.3 changelog

• Update: support of NVIDIA GeForce GTX 650 Ti.
• Update: GPU-Z 0.6.6 and GPU Shark 0.6.7
• Update: ZoomGPU 1.8.10 (GPU monitoring library).
• Bugfix: Offsets for GPU core and memory added to base and boost clocks for GTX 600.
• Bugfix: benchmark scores: average FPS was in some case lower than min FPS.

A new maintenance version of FurMark, the popular GPU burn-in utility, is available. FurMark 1.10.4 embeds the latest versions of GPU-Z and GPU Shark, updates the GPU monitoring code with ZoomGPU 1.8.11 and includes few bugfixes.

FurMark 1.10.4 changelog

• Update: GPU-Z 0.6.7 and GPU Shark 0.6.8
• Update: ZoomGPU 1.8.11 (GPU monitoring library).
• Bugfix: added some checks to prevent crash when logging GPU data to file.

A new maintenance version of FurMark, the graphics cards nightmare, is available. FurMark 1.10.5 includes the latest versions of GPU-Z and GPU Shark, as well as the latest iteration of ZoomGPU v1.8.12.

FurMark 1.10.5 changelog

• Update: GPU-Z 0.6.8 and GPU Shark 0.6.9
• Update: ZoomGPU 1.8.12 (GPU monitoring library) with NVIDIA GeForce GTX Titan and some mobile GPUs (GTX 6xxM, GT 6xxM, GT 5xxM).

A new maintenance release of FurMark, the graphics cards burn-in utility, is available. FurMark 1.10.6 includes GPU-Z 0.6.9 (remark: you have to start FurMark with adminstrative rights in order to launch GPU-Z from FurMark main interface) and GPU Shark 0.6.10, as well as the latest iteration of ZoomGPU v1.8.13 (with support of GeForce GTX Titan, GeForce GTX 650 Ti BOOST and Radeon HD 7790).

FurMark 1.10.6 changelog

• Update: support of NVIDIA GeForce GTX 650 Ti BOOST and Radeon HD 7790.
• added new 16/9 resolutions (one, double and triple HD/WHQD monitors).
• Update: GPU Shark 0.6.10 and GPU-Z 0.6.9
• Update: ZoomGPU 1.8.13 (GPU monitoring library).