Topic: GeForce GTX 900-series power consumption discussion thread (Read 3093 times)

If your power is expensive, mine the lyra2v2 algo. my 750ti's are only using 40 watt each

Thanks for this. I've been wondering exactly what my 970's are pulling while they keep the office warm. 

If you are mining with nVidia GPUs, this thead might be of interest for those looking for GPUs with nice discounts: https://bitcointalksearch.org/topic/wtn-interested-in-buying-brand-new-hddslaptopsgpus-w-great-discount-old-1207919 

Overclocking the memory never gave me any noticeable increase in hashrate in any modern algos. I'm curious if you can achieve a higher hashrate with it though.

Oh, and the 3rd Gigabyte GTX 750 Ti just died on me last week (with the same symptoms).

Just replied to your PM, @ltc_bilic.  I see you are overclocking the memory quite a bit on those Gigabytes.  I think I have the same cards as you, but I only overclock the GPU, not the MEM.  We'll see if it makes a difference.

bathrobehero I couldn't agree more with you. And to support my claim that these cards are not built for higher voltage,...the last two cards have died even though their temp. never surpassed 45C, because I have them in a 16-18C environment temp and additional fans in my open air case. During winter they were clocking at 33C now at 45C overclocked +40 gpu and +248 mem, and they still died. I have EVGA (+76 gpu, +210) not FTW card which runs flawlessly from the start. So I would avoid flashing Gigabyte cards at all cost.

I bought 6 x GV-N75TOC-2GI cards almost exactly a year ago and 2 of them already died on me.
When the first died I replaced the stock BIOS on all of them but a second one died anyway and is in the process of RMA.
They both bricked completely (no fan spinning, nothing). I thought they had solid build quality with plenty of headroom because these have 36 months of warranty.

I also have ASUS (GTX750TI-PH-2GD5) and MSI (N750Ti-2GD5/OC) cards, none of them have additional 6-pin connectors and all the cards have big differences in how much OC they can handle while being stable. A few of them can handle +160 Mhz with most algos while a couple of derpy ones (both MSI) crash even at around +70 Mhz.
So I pushed 60W BIOSes on them as well but it didn't help much with the derpy cards so I reverted them back to stock BIOS with only disabling boost for more consistency.


On another note I did a somewhat useless comparison. Useless because I can't be arsed for now to try each brand with different efficiency PSUs so I just did a quick comparison and the rig with the MSI cards is on a bronze PSU (XFX Pro Series 650W) and is pulling ~9% more from the wall than the ASUS rig does on a gold PSU (EVGA SuperNOVA 850W G2). But then again the MSI cards are 62-65°C while the ASUS are 55-59°C doing the same hashrate with same OC and very close fan speeds (+/- 10%) so it could be just the cards.

If the 9% difference would be solely because of the difference in PSU efficiency (which more than likely: https://en.wikipedia.org/wiki/80_Plus#Efficiency_level_certifications)
then with more investigation it might be worth it to buy really high efficiency PSUs and use 900-series cards.

I've always wondered what are exact power consumptions bathrobehero really great idea, to look into the firmware. So then the 970/980 are not as efficient as it appears at first sight.

And for anyone else into thinking of flashing the 750Ti, unlocking the extra "power" - don't. I've done it on all my cards 14 of them (Gigabyte Windforce) and had most of them replaced. They are just not built for a higher voltage,...2 of them died just recently have to RMA. Perhaps flashing the newer generation of card is another story, but I don't advise it unless you have spare money/time to throw around. If someone is going to experiment with flashing I'm all in for undervolting - to achieve even better efficiency.

You could boot up a windows, find a suitable OC which works with every algo (meaning not too high), note the frequencies and the voltage and flash a BIOS with those settings.

I wonder if you can switch between power states in Linux because if so, you could potentially use different power states as different OC profiles.

Thanks. My main reservation with this solution is I will probably have to flash them A LOT before I find optimal settings and I'm afraid the  bios will just die before I'm done :-/
The offer still stands. Forget nvidia-smi, it does not work with low-end cards...

Gigabyte 750Ti Windforce OC on quark, stock clock, with an 80+ gold PSU, consumes exactly 60W from the wall.
I can't really play with overclocking or power target since my rig is running linux and nvidia removed the coolbits option from the linux driver. Didn't find a way to control that on linux yet.
If someone has a solution that works from the command line, PM me. I'm ready to offer 48 hours of mining (4 GTX 750 Ti) to the pool of your choice :-)

Nice!

This is what lead me to start looking into the power consumption:

bathrobehero has a valid point;  a new thread on this stuff, so that we don't flood sp_'s ccminer thread.

So, here we go--->

To get this thing started, I will quote/cross-post from the previous thread:


I am by no means going to sit here and write as if I am an authority on this stuff.
However, I have taken note of something that I think deserves mention and discussion.

TDP is the term we all seem to use (myself included) to refer to and to determine the power consumption of a GPU.
But, TDP is  short for Thermal Design Power. Further definition is:
"TDP is the average power a device can dissipate when running real applications." (aka "normal" apps that an average user would run)

TDP, then, is NOT exactly equal to the device's maximum power consumption, nor is it necessarily measured at 100% load.

TDP is self-reported by manufacturers. It seems that several years ago, AMD & Intel used different percentages of processor loading to measure and report their CPU's TDP.
AMD used ~100% load and intel something like 80-85%. Well, intel's method seems to have become the norm.
This is what led to the definition of: average power dissipated when running "real applications."

So, under intense loading situations, a device can definitely consume more power and dissipate more heat than its TDP would indicate.
Of course, overclocking raises the amount of power consumed. And mining intensive algorithms certainly is not what would be called "real applications."

From http://www.cpu-world.com/Glossary/M/Minimum_Maximum_power_dissipation.html: