Power Usage of GeForce GTX 750 Ti With Various Crypto Algorithms

8 Jul
2014

palit-gtx-750-ti-gpu

Lately there is much talk about power efficiency of various mining algorithms and with the summer here people with GPU mining rigs are looking for algorithms that use less power and thus the video cards run cooler and quieter. We are starting a series of tests with GeForce GTX 750 Ti GPU first and then we are also going to move to other popular video cards for mining crypto currencies such as the Radeon R9 280X for example.

gtx-750-ti-idle-test-system-power-usage

On the photo above you can see the power usage of the GTX 750 Ti video card in idle as well as the idle power usage of the whole system we are using for testing; below you can find the specifications of the hardware. Note that one of the power meters measures only the power usage of the video card (the power meter is attached to the power line going to the card directly and all power going to it passes through the meter, so it does not take into account the PSU power efficiency) and the other one is for the whole system measured at the wall (the actual full power consumption) taking into account the efficiency of the power supply (extra power wasted as heat during the conversion).

The systems we are using for the tests include:
– Palit GeForce GTX 750 Ti StormX OC 2GB video card
– Intel i3-4130 dual-core CPU at 3.4 GHz
– Asus H81M-A Motherboard
– 2x 4GB A-DATA DDR3 1600 MHz Memory
– 1TB Seagate 7200 RPM Hard drive
– 500W Cooler Master Power Supply

gtx-750-ti-power-usage-algorithms

We have used ccMiner for our tests, the latest fork with Fresh algorithm support and we have measured the power usage of the GPU only as well as of the whole system with all of the supported algorithms by that particular version of ccMiner. Do note that if mining for Scrypt for example you will be getting higher power usage, but this is already pretty pointless to be done with GPUs with so many Scrypt ASIC miners already deployed. The results we’ve seen on the GTX 750 Ti are pretty interesting; it seems that the most power efficient algorithms are Fugue256 and HEFTY1 with the new Fresh algorithm following close by with the same power usage as Qubit. The worst performing crypto algorithms on GTX 750 Ti are the Groestl-based ones and the X-ones are pretty much in the middle. Do note however that these are the results measured on GTX 750 Ti, the situation with AMD with the same algorithms may differ significantly and we do plan to run some tests to check the situation there as well, so stay tuned for more very soon, probably tomorrow.

gtx-750-ti-power-usage-meter-2

Since there were some questions and people doubting our measurements, we have repeated the tests with another power meter connected to measure the power going only to the video card and the results are pretty much the same as with the previous meter in terms of power usage as you can see on the photo above. Do note that the Palit GeForce GTX 750 Ti video card that we have used for testing does not have an external PCI-E power connector available, so all of the power going to the video card is from the PCI-E slot. So in order to measure the exact power used by the video card we have used a powered PCI Express x1-x16 USB 3.0 Extender. This extender does not use USB 3.0 interface, just a USB 3.0 data cable for the transmission of data between the PCI-E slot on the motherboard and the video card (no power is transmitted over that cable). Instead the power provided to the video card all goes through the 4-pin Molex power connector on the extender’s board. Also do note that the power measured is coming directly from the power supply, so this measurement for the power usage of the GPU does not take into account the power efficiency of the power supply (loses of power during the conversion from 110V/220V to 12V) and depending on the power supply there will be about 10-20% of extra power lost as heat during the conversion. This power is measured by the second power meter that does measure the full system’s power consumption at the power socket however.



Other Similar Publications:

18 Responses to Power Usage of GeForce GTX 750 Ti With Various Crypto Algorithms

GrimVandal

July 9th, 2014 at 09:32

this values are totally OFF

are you meassuring only the 6 pin pci-e power plug?

you have 12v at the 6pin
you also have 12v at the pci-e slot
and you have 5v at the pci-e slot

normally you draw 1 – 2 amp at the pci-e plug at 12 volt = 12 – 24 Watt
another 2 amp at the pci-e slot at 12 volt = 24 Watt
and 0.250 amp at the pci-e slot at 5 volt = 1.25 Watt

Gives a Total of 49,25 Watt
NOW add about 10% loss due to the average psu

and you get at least 40 to 54+ Watt for a single 750ti

GrimVandal

July 9th, 2014 at 09:35

the easiest way to meassure is to put your gpu on a usb riser and measure all the 12v lanes (and the 5volt lane) with a clamp meter !!!

and dont forget to add your average 10% due to psu loss (there is no 100% power conversion and leaving that out is PLAIN FALSE)

admin

July 9th, 2014 at 09:50

GrimVandal, we are actually measuring it exactly this way, the gpu power usage that is. The video card is on a powered raiser and it does not have an external PCI-E power connector, so all the power that actually goes to the GPU passes through the power meter and is properly measured (there is a mosfet on the extender to take care of the 3.3V). The power usage for the GPU in the table is without taking into account the efficiency of the power supply as this can be actually measured at the power socket where we do measure the full power usage of the whole system.

GrimVandal

July 9th, 2014 at 11:43

Still something strange is going on here… the values are totally OFF by about a HALF.

I use a DC clamp meter to measure my gpus.

As I said above the most accurate way to do it is to use a powered usb riser cuz you can easily meassure all the power going to the gpu: (as you did)

In my case eg for Jackpot:

Gigabyte 750ti
core clock 1250 mhz
mem clock: stock

hashrate ~ 5100 khash/s

through the pci-e plug (yes this 750ti has one) flows: 1.6 amp @ 12v
through the molex plug to the pci-e slot flows: 2.2 amp @ 12v
and YES through the 5 volt molex plug flows: 0.250 amp @ 5v (which is converted to 3.3 volt onboard the pci-e usb extender)

1.6 times 12 = 19.2 Watt
2.2 times 12 = 26.4 Watt
0.250 times 5 = 1.25 Watt

Total wattage = 46.85 Watt
times 1.1 = 51.54 Watt (standard Gold power supply)

admin

July 9th, 2014 at 11:55

GrimVandal, and now you can try powering up the video card without the external PCI-E power to see if it will work and if it does start up to measure the power then, taking the external power out of the equation and you will get pretty much the same results as we do (we are testing with the card at stock frequencies, so a bit lower power usage). The card we are using does not have an external PCI-E power connector available on the PCB.

Our experience and test have shown that these cards can take up a lot of extra power if you allow them to without almost no performance gain as a result – by increasing the video BIOS TDP limit for example.

GrimVandal

July 9th, 2014 at 12:13

My 750ti is not recognized correctly under device manager if I remove the 6 pin pci-e plug.

Besides that the Palit 750ti is meassured to draw 7.65 watt idle. The lowest ever meassured 750ti was 6+ Watts in idle and up to 8 Watts. (everything stock)

And tada your meassured 3.8 Watt is exactly HALF again of what it should have (7.65).

(The above has been meassured not by me but by several online hardware testers.)

Once again your meassurement is half off this tells me your meassurement device is used in a wrong way …

GrimVandal

July 9th, 2014 at 12:17

Morover substract your system power form the wall:

eg for Jackpot

97 Watt
– 38 Watt (Idle)

= 59 Watt

see the pattern now …

admin

July 9th, 2014 at 12:23

We have tried with two different power meters and the results we are getting pretty much the same on both, the accuracy on these might not be 100%, but the tolerance should not be more than 1-3%.

But let us look at this in another way, let us say the GPU is using 8W at idle with the whole system idle at 38W, so the system uses 30W of power when idle. If the GPU uses 50W at load while mining and the whole system is at 80W under load, there are again 30W for the rest of the hardware (not 100% load, but higher than when idle), so this cannot be right. We have a motherboard, a 54W MAX TDP CPU, memory, CPU cooler plus two additional fans, hard drive…

GrimVandal

July 9th, 2014 at 12:26

now double your wattage of Jackpot

25.7 x 2

= 51.4

x 1.1 (assumed 90% efficiency)

= 56.54

(at this low wattage your psu probably delivers only 85% efficiency at max which explains the discrepancy of 56.54 to 59 watt)

admin

July 9th, 2014 at 12:27

The power supply is more like 70% according to the manufacturer specs: http://www.coolermaster.com/powersupply/extreme/extreme-power-plus-500w/

GrimVandal

July 9th, 2014 at 12:39

yep you have to substract the 8 idle watts of the gpu, thats true

so true idle is 30 Watt
your total system power is 97 watt (jackpot)

thats 67 watt for your 750ti

now times 0.7 (70% efficiency)
46,9 watt
now times 0.75 (75% efficiency)
50.25 watt
now times 0.8 (80% efficiency)
53.6 watt
now times 0.85 (85% efficiency)
56.95 watt

Looking at your psu (which is awful XD), at 97 watt it probably delivers betwenn 80 to 85 % efficiency.

admin

July 9th, 2014 at 13:10

“Our Extreme power supplies provides over 70% efficiency at typical loading” and at 100W out of the 500W rating of the PSU I doubt it can deliver more than 70% efficiency as the typical rating they talk about is probably more like with 50% load of the PSU.

What is wrong in your calculations is that you still use the idle power of the system as the power consumption when it is under load, at idle the CPU lowers the operating frequency and voltage, so the power consumption is lower as well, the same goes for the other components as well, though there the difference is smaller. While mining it is not in power saving state anymore and it does consume more power, not at 100% load of course, but mining does need some CPU power as well (varying between algorithms). So the power used by the system under load excluding the video card is at least 40W or maybe even more, with light system usage (GPU in 2D mode, not mining) the measured power at the wall is around 50-55W on the power meter (this includes the GPU usage)…

Anyway, the idea of this post was not to focus on the GTX 750 Ti power usage only, but to see the power usage of different algorithms supported by ccMiner. Different algorithms do have varying impact on the whole system usage, most notably the CPU of course, so the total power usage of the system is actually more important when considering the mining costs as you pay for the electricity used by the computer measured on the wall.

GrimVandal

July 9th, 2014 at 13:46

Please verify your results with a 3rd tester!

All I can say is that your meassured wattage of the 750ti is WRONG, and not by 1-3% but by 50% !!!

You beat about the bush too much …
You can clearly see it by simply looking at the idle and full load power at the wall that ~25 watt for a 750ti can’t be right.

Please verify your result!

chavvdarrr

July 9th, 2014 at 17:11

Why don’t you make these same measurments one more time, but this time with 2 videocards?
Case solved :)

PS: add cryptonight (monero/quazar/bytecoin)

admin

July 9th, 2014 at 22:42

We have found a better and more accurate way to properly test the GPU only power consumption, we are going to be redoing the tests in the next few days for the GTX 750 Ti along with AMD GPUs like the 280X…

lopo

July 10th, 2014 at 02:55

Where are the Scrypt, Scrypt-N, Protoshares, Primecoin and Monero results? :D

Jerry

July 11th, 2014 at 12:01

Can you make the same test with a 780 Ti to see the difference to the 750 Ti?

lojzo

August 22nd, 2014 at 02:06

GrimVandal is RIGHT!

Leave a Reply

Your email address will not be published. Required fields are marked *

top