Please note, this is a STATIC archive of website hashcat.net from 08 Oct 2020, cach3.com does not collect or store any user information, there is no "phishing" involved.

hashcat Forum

Full Version: Strange GPU behavior? (OC Nvidia GTX570)
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
During a WPA cracking session I decided to overclock my GTX570 to see what effect it has on my hashing speed. Stock voltage and clock settings in MSI afterburner are 0mV and 770MHz respectively.

Changing my core clock to 850MHz made a nice change from ~31kh/s to ~35kh/s, not bad! Here comes the weird part...

No matter how much I increase the core clock speed, I'm able to see very positive changes to the hash rate. What's interesting is that I don't have to change the voltage. (wtf)

People familiar with overclocking know that you need to increase the voltage if you want a higher core frequency, yet, for some reason, I'm able to increase my frequency to ridiculously high amounts (900MHz!) without changing the voltage. That is definitely not normal behavior if you ask me! Under normal conditions, a gpu should timeout and the drivers crash causing a temporary black screen if it's not supplied with enough voltage for the amount of processing it's trying to do. You'll also see screen artifacts!

What's even more silly is that I'm able to run 900MHz (~37kh/s!) with LESS voltage than stock. In other words, i'm undervolting the card, and still somehow getting an increase in hash rate.

Normally, I can't run 900MHz during a gaming session unless I bump the voltage up by 100mV without getting driver crashes, yet I can hash at the same frequency with an undervolt (-20mV in afterburner).

I did have to do the TDR detection registry patch to get cudahashcat64 running, so i'm thinking that this might be the root of this strange behavior. What do you guys think? Is this normal behavior when hashing with gpu's? Is what i'm doing dangerous for my gpu? Let me know what you guys think about this situation. I could just be freaking out over nothing.

For the time being, I have settings at stock until I can get to the bottom of this.
do not test overclock stability with wpa. test with MD4 and -n 800, or use md5stress.
(01-05-2014, 08:13 AM)epixoip Wrote: [ -> ]do not test overclock stability with wpa. test with MD4 and -n 800, or use md5stress.

I hadn't intended on test stability necessarily, the results of overclocking just took my by surprise. I don't understand why I'm able to increase frequency without increasing voltage.

The OC I did shouldn't be stable with even a small amount load from normal computer use, so why would it be stable under 90% load when hashing? Is hashing special? Does the patch have something to do with this?
the patch has nothing to do with it.

overvolting and overclocking do not necessarily go hand-in-hand. overvolting is only necessary in the case of aggressive overclocking. a 130 Mhz overclock is not aggressive. a 7970 can be overclocked by 275 Mhz or more before overvolting is required to sustain higher clocks.

the reason you did not get an asic hang is because wpa is not entirely alu-bound, so you were not stressing the card in a way that pushed the voltage limits. this is why you need to test with a purely alu-bound algorithm like md4 or md5.
This was the answer I was looking for, thank you.