A Bit More on the Actual Power Usage of GeForce GTX 750 Ti

28 Feb
2014

geforce-gtx-750-ti-power-usage-test-setup

The GeForce GTX 750 Ti series of graphics cards based no the new Nvidia Maxwell architecture have really captured the attention of crypto currency miners due to their good performance (hashrate) per watt. Nvidia is advertising these cards to have 60W TDP, however we have discovered that the actual TDP limit for the power target set in the video BIOS of these cards is 38.5W…

geforce-gtx-750-ti-power-usage-results

This made us dig into the matter a bit more in order to see how much power actually does a GeForce GTX 750 Ti draw from the motherboard (a reference GT 750 with no external PCI-E power connector). So we’ve attached a watt meter to the power line of an external PCI-E extender to see how much power is drawn by the GT 750 Ti card. The results we got were quite surprising, even though we knew that the default power consumption should fit in the 38.5W limit for the power target set in the BIOS. With the stock frequencies for the GPU and the video memory the GeForce GTX 750 Ti has shown to draw just about 28W ot power producing about 250 KHS hashrate for Scrypt mining. After getting the card overclocked to +135 MHz for the GPU and +650 MHz for the video memory the power consumption has risen a bit to about 34W average with a hashrate of about 300 KHS for Scrypt crypto mining.

So if you were calculating the power usage of GTX 750 Ti as 60W in order to see the ratio of hashrate per Watt of power, then you should rethink how you calculate it now. Note that this is the actual power that goes from the PSU to the video card, the real power consumption with a 80 Plus certified power supply that provides 90% efficiency for example would rise with 10% to about 30W (stock) and 37W (overclocked) respectively as the actual power drawn by the card from the wall socket.



Other Similar Publications:

30 Responses to A Bit More on the Actual Power Usage of GeForce GTX 750 Ti

martinium

February 28th, 2014 at 18:08

If 100% accurate this is even more amazing than before! Wow! Great job NVIDIA. I am excited just thinking about the mid and high end maxwell gpu’s.

jk_14

February 28th, 2014 at 22:43

I’m affraid it’s too optimistic.
Few minutes ago I have measured Gigabyte 750 Ti WF

system without card is 29W (at the wall), and mining litecoin or vertcoin – 101W

( measured with this:
http://itstore.lu/out/pictures/master/product/1/40500020.jpg )

haigent

February 28th, 2014 at 23:01

Hi, hk_14

We has the same measured, around ~75W per card for Gigabyte 750 Ti.

BTW: We only can get 290 KH for the 6 card rig(TDP 65.5), also not stable(not sure is this caused by CudaMiner).

Robin

February 28th, 2014 at 23:05

@ jk_14: That includes an increase in power draw from all system components when mining. This will increase no matter what the card used. The author is measuring the *exact* draw from the card.

admin

February 28th, 2014 at 23:19

The measurement above is only for the card, the watt meter is placed between the power supply and the powered extended measuring all the power that goes to the video card. This is the exact power consumed only by the video card with stock and with overclocked settings. When measuring the full system power draw with with card at idle state and when you start mining you cannot get adequate result only for the video card’s power consumption, because not only the card has increased power consumption, but the whole system has increased load – CPU, memory, motherboard etc. use more power as compared to when the system is in idle (power saving mode). Our goal was to measure the exact power usage of the GTC 750 Ti only, also if you are interested the idle power consumption of the card only is about 2-4W.

jk_14

March 1st, 2014 at 00:51

“when you start mining you cannot get adequate result only for the video card’s power consumption, because not only the card has increased power consumption, but the whole system has increased load – CPU, memory, motherboard”

It’s not a problem to take two items, and calculate this little factor
btw: in your opinion – additional CPU, memory, motherboard load is linear to number of cards, or not? :)

Idle power consumption is definitely 8-9W (10W at the wall, efficency let say 85%, no your “mining effect :) )

americonfusion

March 1st, 2014 at 09:25

Haigent,

Not sure what kind of cards you are running but I have 5 EVGA FTWs running at 305-315khs each, very stable as well. Try the T5X24 kernel setting or T10X24.

Venco

March 2nd, 2014 at 03:27

Which motherboard do you use? do you use power in pci-e extenders?

dkuhlman

March 2nd, 2014 at 11:58

I have an EVGA version of this card. I have the GPU clock speed at +135 and the memory at +670 but can’t get over 301khs. I get the highest hash rate with T20x20. I’ve tried increasing the voltage but I can’t seem to get the memory clock rate much higher; and can’t get the hash rate higher at all. Is there anything I can do to squeeze anything else out of this?

admin

March 2nd, 2014 at 12:49

Have you tried increasing the power target limit by modifying the video bios, you can find a guide here:
http://cryptomining-blog.com/1014-how-to-increase-the-geforce-gtx-750-ti-power-target-limit/

Mark

March 4th, 2014 at 04:25

@admin :
You used a powered PCIe riser, which is powered via a molex connector, right ?
And from what I understand, you measured the draw from that molex connector ?

If so, you are neglecting the power sent from the motherboard PCIe to the graphic card via the flat ribbon cable. That could explain your very optimistic result.

You could try to measure the wall power outlet draw from 6 such cards, you’ll get way less interesting results. Sadly :(

nVidia would be the best competitor to Scrypt Asic if they could reach such a high KH/s per watt !

admin

March 4th, 2014 at 10:30

Mark, no, actually you are missing something… look at the picure above!

We did not use powered PCI-E riser with ribbon cable, but a one that uses USB 3.0 cable and that one does not have any of the power pins connected to the motherboard… all of the power to the video card is supplied through the molex connector powering the extender. The USB 3.0 cable used is only for the data lanes. So this is the full actual power usage of the video card only.

Mark

March 4th, 2014 at 20:43

Are you sure that the USB 3.0 cable doesn’t drive a bit of power ? By design, it can bring 100 Watts, so the cable is no limiter to what to motherboard can send through its PCI Express ports.
A motherboard can deliver power through PCIe even without additionnal molex.

A way to concur your findings would be to measure the total power draw from a 6 x 750 Ti Scrypt rig. Hope you are right!

admin

March 4th, 2014 at 22:33

Mark, if you are using a USB 3.0 cable as it is meant to be for a USB device you normally transfer 5V over it to power the device connected with the cable. This however is not the case with the use of the USB 3.0 cable here with the PCI-E risers, here it is solely used to transfer data – the PCI-E data lanes are routed through the cable all of the power goes to the card through the molex connector with the board where the PCI-E slot for the video card is located. The USB 3.0 cable is a convenient solution for longer distance as it is designed for high bandwidth transfers being shielded and all with enough data lines on compact design, unlike the ribbon cables that are used for the cheaper PCI-E extenders that do not allow for very long distance due to signal degradation.

Mark

March 5th, 2014 at 19:35

Thanks for the detailed information. As I said, I hope you are right, and you seems very confident that you are.
I just don’t understand why every other person measuring their power draw with Kill-A-Watt and similar instrument find around 60W, and not 30-37W as you do.

May it be possible that the reference card is more power-efficient than custom OEM design with additionnal PCIe 6 pin connector?

Mark

March 5th, 2014 at 19:41

Another question : did you measure the power consumption and hashrate in both PCIe 16x and PCIe 1x ?

admin

March 5th, 2014 at 20:14

Kill-A-Watt measures the power at the power outlet of the whole system and measuring the system at idle and then with the video card mining and subtracting the numbers does not give you accurate results for the card consumption only (it also takes into account the efficiency of the PSU). When you start mining with the card it also increases the overall power consumption of the whole system (processor, motherboard, memory…), that is why you are getting around 60W and not more like 40W for example. Do have in mind that the Power Target limiter of these cards is set to 38.5W and that goes not only for the reference board, so the built-in power limiting functionality of the card will not allow it to go past this power consumption anyway… unless you modify the BIOS to increase the limit.

decagrog

March 6th, 2014 at 10:08

It’s not a viable way to measure the power usage, your riser is giving power only to the 12Volt line, but you missing the 3.3V power (on pins 8-9-10) and those pins pass for the usb connector…by specific the 3.3V lines are rated to 3A, so there are at least other 10Watt of consumption.

admin

March 6th, 2014 at 10:57

decagrog, in fact you are only partially right. The card does use 3.3V and by PCI-E specs this can go up to 3A. However this 3.3V line is not going through the USB cable at all, as we’ve said the USB 3.0 cable is used only for the data lanes. All of the power to the video card goes through the 12V line provided by the 4-pin molex power connector that is plugged in the PCI-E USB Extender…

The board of the extender where you place the video card has a single AZ1084 voltage regulator that does take 12V input and outputs 3.3V in order to supply the video card with the required 3.3V along with the 12V power. There are no like extra up to 10W on top of what we have measured because of the 3.3V line, the power consumption of this line is included in what we have measured on the 12V input line.

decagrog

March 6th, 2014 at 11:30

@admin
Well in that case since data lines power are negligible you are really measuring the effective power usage.
Still 37watt seems really low to me but I’ll give a few test myself as soon I receive my Gainward 750 TI. :D
By the way, the riser with the onboard 3.3V power regulator seems pretty nice, do you have a link to them?

admin

March 6th, 2014 at 12:50

decagrog, we are gong to be doing some more tests with different cards as well, including ones with external PCI-E power connector soon. It seems that pretty much all PCI-E Extenders that use USB 3.0 cables follow this design with a 3.3V power regulator onboard, we have picked up ours from eBay. Just search for “PCI-E USB 3.0 Extender” or “PCI-E USB 3.0 Riser”

decagrog

March 6th, 2014 at 13:19

thanks, I’ll check those riser!

BrianB

March 8th, 2014 at 02:57

I just bought a 750 Ti after reading this, and I would greatly appreciate you run more tests, I’m also in the group of “there’s no way”….

MetalPhreak

March 15th, 2014 at 17:01

Asrock PRO BTC board + 1x4GB DDR3 memory stick + Intel Celeron CPU + 6x Gigabyte GTX750Ti OC cards + In Win 80 Plus Bronze PSU

Idle: 65-70W
Full Mining Load: 450W

Just under 300kh/s per card, using pci-e x1 USB powered risers. No TDP BIOS mod at all.

So assuming my PSU is 85% efficient (240V supply), 380W / 6 Cards * 0.85 = 54W of power per card.

Not sure if these gigabyte cards have the power limit raised with the factory bios. 38.5W TDP makes sense if you account for some efficiency losses in the DC-DC power circuits on the card itself.

Anthony D Diehlman

March 17th, 2014 at 15:56

Where did u get the watt meter from? What’s the name of it. I have been searching for one for couple weeks now.

admin

March 17th, 2014 at 16:02

Anthony, it is a SKYRC Watt Meter.

Mantle

April 3rd, 2014 at 04:46

I have read that the use of the USB risers can have a negative effect on hash rates due to the fact that there are not enough lanes of data and that the asrock risers that use sata cables to transfer the data work better. Have you thought about doing any testing in this matter?

admin

April 3rd, 2014 at 10:32

Mantle, yes, there is a slight drop in performance when you use risers, but that not only covers USB cable risers, but any riser that is 1x and not a full 16x… the lower performance comes from the fact that you are not using 16 PCI-Lanes, but only a single one for data transfer.

Roy360

June 8th, 2014 at 02:08

same motherboard, Corsair 750W Gold PSU, using cudaminer, however why is my power draw in the 600s? With my original Ultra power supply, I was drawing close to 700W.

What kernel are you guys using?

G3220 overclocked and unvolted to 800Mhz, 0.6V
1333mhz undervolted to 1.5V

Roy360

June 8th, 2014 at 02:11

Leave a Reply

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

top