The power limitation function of the GTX 580 GPU, which prevents power draw exceeding 200W, can now be bypassed by GPU-Z.
Nvidia's GTX 580 GPUs include a clever power limiter that keeps the power draw low when executing intensive operations such as the Furmark benchmark. Now, however, GPU-Z developer W1zzard has released a way to disable the restriction - but warns that it's not for the faint-hearted.
The test build, available for download on the TechPowerUp forums
, has the option to disable the clockspeed restriction logic. Although it makes no difference during daily use, when the limiter kicks in during Furmark or similar intensive benchmarking applications, the maximum power draw of GTX 580-based cards jumps from 200W to 350W.
W1zzard warns users that 'exceeding the power limitation of the card may result in damage to card and/or motherboard,
' but for those who want to push their GTX 580-based cards to the very limit it's a way to squeeze the very last drop of performance out of the GPU - providing adequate cooling is in place.
Nvidia's GTX 580's power limiting system is something that would have potentially averted the overheating issues that plagued early versions of StarCraft 2. When we investigated the issue back in August, we found that the lack of a framerate limiter
in the software meant that simple scenes, such as the menu, could over-stress the GPU - a problem that the GTX 580's power limiter neatly circumvents.
The news that Nvidia's GTX 580 can draw up to 350W when the power limitation logic is switched off will come as no surprise to rival GPU manufacturer AMD, which famously mocked its competitor on YouTube
for the high heat output of the Fermi architecture. However, the much more reasonable 200W limit of a card with the power limiter still in place will be good news for those looking to upgrade from a power-hungry early Fermi board.
Are you impressed to see Nvidia's power-saving logic circumvented already, or just wondering why anyone would risk burning out their GPU for the sake of a synthetic benchmark score? Share your thoughts over in the forums