At 100% the draw goes up by 4-5W.
 
0% Idle

 
100% - Idle. The screenshot shows a spike at 15% but the PC is idle with all other apps closed.
 

 

talon951

KingEngineRevUp

My Hypothesis

I'm going to guess what is going on here. Because the both of you are drawing the same average voltage across the run, this leads me to believe the same thing I found here: https://forums.evga.com/I-did-some-testing-and-I-believe-I-understand-the-discrepancies-of-power-draw-between-the-3080-3080-Ti-and-3090-m3439433.aspx

Although you are drawing more power, that doesn't mean that power is actually going to rendering and rasterization. The extra power you're drawing is going to your fans, RGB, if someone installed a hybrid kit the pump, etc. 

That would explain why gsrcrxsi and you both have similar average voltages. 

This means the XC3 has 320W (or less) for the GPU to use to do rendering and rasterization and the other 30W is saved for peripherals and who knows what else.

 

The bit that you are missing is that the card will use more power as it gets hotter because of increase electrical resistance.  One card is air cooled, the other water.  It's not a big change, but you're talking about small percentages anyway.  I could see this last night while I was doing the clamp meter tests.  The current draw would increase slightly as the card heated up.  
 
The fan wattage comparison is what I've seen before and what I stated in these threads previously.  About 5-10w going from 0% to 100%.  Tested it on both my Zotac 3090 and 3080ti FTW3 when they were air cooled.

.. That logic makes sense, I wish there was a safe way to test that.

How about this?  I turned my fans down some so temps would go a bit higher for a bigger delta.  Looked like about 15w.  Went form low 430's to mid/high 440's.
 

ZheGoose
I just got my 3080 Ti XC3 gaming a few days ago and my card also has the low power consumption issue. The card usually hovers around 315W - 330W during gaming. My card also has this weird issue, where it cannot hit the advertised 1665mhz boost clock in Metro Exodus Enhanced Edition at 3440x1440p at ultra settings. It will stay a few boost bins below the advertised 1665mhz boost clock. This is with the card at stock with just the power target at 105% and the temperature limit set to 91.
 
Here is an Imgur link to a short video showing how the card is behaving in-game (sorry for the shaky camera lol):  
 
I also ran the Timespy benchmark so I could contribute my results to this thread. Here is the link to my run at stock settings with the power slider and temperature limit set to the max: 3dm/64373499
 
I also have my GPU-Z log for the Timespy run but I'm unsure on how I can share it. 
 
Edit: It looks like I can't share links. Is there any way I could share my Timespy results and the video showing how the card is behaving in Metro Exodus Enhanced edition?

What's the power usage during time spy's second graphics test? For the metro video, just add a few spaces in the url

Would it be possible for you to download OCCT and run the graphics stress test in it? I'm reliably able to hit 350W in that on my card.

ZheGoose

speedysloth

ZheGoose
I just got my 3080 Ti XC3 gaming a few days ago and my card also has the low power consumption issue. The card usually hovers around 315W - 330W during gaming. My card also has this weird issue, where it cannot hit the advertised 1665mhz boost clock in Metro Exodus Enhanced Edition at 3440x1440p at ultra settings. It will stay a few boost bins below the advertised 1665mhz boost clock. This is with the card at stock with just the power target at 105% and the temperature limit set to 91.

Here is an Imgur link to a short video showing how the card is behaving in-game (sorry for the shaky camera lol):  

I also ran the Timespy benchmark so I could contribute my results to this thread. Here is the link to my run at stock settings with the power slider and temperature limit set to the max: 3dm/64373499

I also have my GPU-Z log for the Timespy run but I'm unsure on how I can share it. 

Edit: It looks like I can't share links. Is there any way I could share my Timespy results and the video showing how the card is behaving in Metro Exodus Enhanced edition?

What's the power usage during time spy's second graphics test? For the metro video, just add a few spaces in the url

The card is mostly staying in the 310W-320W range during the test. The lowest the card dips to is 299W for a brief second and the absolute max it hit was 346W, also for a brief second. I added the spaces to the links for the Timespy results and the Metro Exodus video.

Do a TimeSpy run with your fans at 100%. I bet you youll hit speedys power draws.

Feel free to log the results with GPU-Z too for us

ZheGoose
I ran OCCT and with the power slider at 105% and the temperature limit at 91C in PX1 and the card will pull between 359W-361W at around 1500-1530 Mhz. So it looks like my card has the ability to pull in more than 350W but only does so with this test. 

Run time spy with fans at 100%.

KingEngineRevUp

speedysloth
 
For sure. Let me know if this works
 
https://www.swisstransfer...4dd6-b904-d725f9b37d76

Your Results compared to gsrcrxsi

 
Interesting, your results are as you say they are:

• Fan 1-3: 70%  
• Average Power: 345W
• Average Voltage: 0.894v

gsrcrxsi on the beta bios, he had a +125 on his core:

• Fan 1-3: 0% (he's using a hydro copper)
• Average Power: 320W
• Average Voltage: 0.901v

The difference in your voltages are just 0.7% which is within a margin. 
 

The difference between your set-up and gsrcrxsi

 
The XC3 probably uses similar fans to the FTW3, I can't find them so we'll have to use these for now:
 
https://www.gpufanreplacement.com/collections/evga-graphics-card-fan-replacements/products/evga-rtx-3080-ftw3-cooling-fan 
 

 
These fans are 0.55A and they're 12V. At max, one would push 6.6W, so all three together would be 20W. But you had them running at 70% so I'll estimate 14W from your card was used to run the fans. I'm going to assume you have the side, back and main RGBs running too? 
 
If gsrcrxsi had a power draw of 320W average and he had fans and RGB running it's very well possible his average power draw would be up there, like at 340W. But your average voltages are the same, even though you are drawing more power. 

What does this mean? 

Because the both of you are drawing the same average voltage across the run, this leads me to believe the same thing I found here: https://forums.evga.com/I-did-some-testing-and-I-believe-I-understand-the-discrepancies-of-power-draw-between-the-3080-3080-Ti-and-3090-m3439433.aspx
 
Although you are drawing more power, that doesn't mean that power is actually going to rendering and rasterization. The extra power you're drawing is going to your fans, RGB, etc. 

Help me verify your fan power draw

Do me a favor and do the following:
 

1. Make sure your system is idling with nothing running and using your GPU
2. Open GPU-Z
3. Set fans to 0%
4. Set board power draw to average
5. Click reset
6. After 20 seconds, what is your board power draw? 
7. Set fans to 70% (like in your test) and hit reset, don't forget to hit reset
8. After 20 seconds, what is your board power draw?
9. Set fans to 100% and hit reset, don't forget to hit reset
10. After 20 seconds, what is your board power draw? 

My Hypothesis

I'm going to guess what is going on here. Because the both of you are drawing the same average voltage across the run, this leads me to believe the same thing I found here: https://forums.evga.com/I-did-some-testing-and-I-believe-I-understand-the-discrepancies-of-power-draw-between-the-3080-3080-Ti-and-3090-m3439433.aspx
 
Although you are drawing more power, that doesn't mean that power is actually going to rendering and rasterization. The extra power you're drawing is going to your fans, RGB, if someone installed a hybrid kit the pump, etc. 
 
That would explain why gsrcrxsi and you both have similar average voltages. 
 
This means the XC3 has 320W (or less) for the GPU to use to do rendering and rasterization and the other 30W is saved for peripherals and who knows what else.

But what about Koopa's Ventus? 

Koopa has a card that is designed differently. I wouldn't know how it draws its power without testing it and testing it against a water cooled Ventus. A simple test would be for Koopa to do the same fan test as you and see if the power draw goes up when he uses his fans. My guess? 

• It's possible that MSI doesn't calculate the fan power draw into the board power draw
• The fans might draw a X amount of W, but it won't show up on the board power draw probably
• If the above is true, that means the MSI card gets the full 350W of power to be used for rendering and rasterization
• This would explain why Koopa has a higher average voltage than you guys, his average voltage in the run was 0.937V
• Because he has access to 0.037V higher than you guys, his card gets to hold onto higher clocks on the voltage curve

 

What now?

If my hypothesis is true, this might be bad news. 

• EVGA might argue back that performance is as intended, the card can reach 350W TDP when all of its PWM and RGB is all used
• EVGA might say that they only promised a boost of 1665 Mhz and the card delivers that

This is not looking good for you XC3 owners but maybe things can change with a BIOs or firmware update still? We'll have to wait and see.

I think you’re suffering from confirmation bias. You’re looking for the answer that explains your preconceived theory.

What you’re missing is that I’ve had two identical hydro copper models in my possession. And they acted wildly different. Up to 30W difference in power draw between the two cards under the same exact situation/load. RGB and fans (which I don’t have) can’t explain this. The most conclusive trend I’ve shown is the link between memory controller load and power draw at an inverse relationship. Also. Try to compare krams HC card to my HC card. Makes little sense to try to compare his air cooled card against my HC card with so many differences when HC:HC is a much better apples to apples comparison.

Fans DO contribute to the board power draw. That’s why it’s called “board” power draw. It’s global. It’s covered everything. Fans and RGB get their power from the PCIe slot.

Boost clocks are not 1665, they are 1725 for the cards that kram and I have. And the majority of the time, under the load I’m running most of the time, my card is unable to sustain it.

ZheGoose
Here are the results with timespy with the fans at 100%, 105% PL, and 91C temp limit. https:// www. 3dmark. com/ 3dm /64377889
 
In the GPU-Z log for test 2, the GPU stays mostly in the 325W to 335W range with the highest power pull at 340W for a brief second, and the lowest pull was 315W also for a brief second. 

I'll need your log from this to check your voltages and how much they're dropping to compare to speedy and gsrcrxsi. Sorry if it's asking for too much but if your average voltages are around their range then your cards across-the-board are functioning the same at least.

talon951
Keep in mind my test showing the total power increases with temp only happena/works because the card isn't reaching the board power limit in the first place. If it rode the limit all the time you would so no increase in power with temp.

every benchmark and load test I've run has the card hitting the power limit. GPUz reporting perfcap reason = PWR. this is a very clear indication that the card is under power limit.
 
that's the point of this whole thread. that there is SOME power limit in place, but it's not due to total board power as it should be. clocks are being pulled back due to some other power limit (most likely suspect for the XC3 cards being the PCIe slot since it's riding 66W for nearly everyone that has one), the lower clocks reduce overall power draw, so we end up seeing these artificial limits displayed in the total board power. I believe that some or all parts of the memory subsystem are taking power via the PCIe slot, so more load = more of the PCIe slot budget being used = earlier clock limiting = lower total power draw total. the card only knows how to limit core clocks, it will never pull back on the memory clocks.
 
think of it like a rev limiter on a car limiting the total MPH. the car isnt limiting based on MPH, it's limiting based on revs, which in turn caps you at some MPH value. allow more revs, get more MPH as a result.

ZheGoose
How can I send you the logs? They are txt files. 

Before you start your run, on the bottom left in GPU-Z sensors, click "log file."

quote=gsrcrxsi]

talon951
Keep in mind my test showing the total power increases with temp only happena/works because the card isn't reaching the board power limit in the first place. If it rode the limit all the time you would so no increase in power with temp.

every benchmark and load test I've run has the card hitting the power limit. GPUz reporting perfcap reason = PWR. this is a very clear indication that the card is under power limit.
 
that's the point of this whole thread. that there is SOME power limit in place, but it's not due to total board power as it should be. clocks are being pulled back due to some other power limit (most likely suspect for the XC3 cards being the PCIe slot since it's riding 66W for nearly everyone that has one), the lower clocks reduce overall power draw, so we end up seeing these artificial limits displayed in the total board power. I believe that some or all parts of the memory subsystem are taking power via the PCIe slot, so more load = more of the PCIe slot budget being used = earlier clock limiting = lower total power draw total. the card only knows how to limit core clocks, it will never pull back on the memory clocks.
 
think of it like a rev limiter on a car limiting the total MPH. the car isnt limiting based on MPH, it's limiting based on revs, which in turn caps you at some MPH value. allow more revs, get more MPH as a result.


We know there's a limit, the point I'm trying to make is that limit makes you hit a perf cap at 300-320W. The remainder of the cards draw is dedicated to peripherals.

Your card sucks, it doesn't work like the Ventus which uses more of that power to do rendering and rasterization.

345W XC3 < 345W Ventus

We have two cards and data so far where both pull close to the same power YET one sustains higher average clocks and voltages, 0.037V more on average.

That's my point. I don't know how that's being"confirmation" bias when I'm literally looking at numbers and data to showed me this.

I did that for my 2 runs. The log files are already created and sitting on my desktop from the Time spy runs. They are .txt files. How can I send them or post them here?

KingEngineRevUp

gsrcrxsi

KingEngineRevUp
So feel free to explain why your card and his card average the same voltages, and therefore would be arou the same voltage curve and clock ranges yet the both of you draw different average power.

silicon quality and leakage
 
so you think my RGB must be the reason that my first XC3 HC only pulled 270W under furmark, while my second XC3 HC pulled 300W? exact same system, exact same software/settings. only difference was one card to another, same sku, different serial.

Where are your logs for both of those cards? What were the average voltages for them and average clocks?

I'm not stating anything as a fact yet btw, I said it was a hypothesis.

Sent you the logs. 

gsrcrxsi

KingEngineRevUp
The remainder of the cards draw is dedicated to peripherals.

This idea is frankly laughable. the HC card has no peripherals. RGB on vs off is like 2W lol.
 
and the fact that I CAN get it to 100% TDP with certain tests (DX12 RT) proves this false. there is no dedicated power "reserved" for peripherals. anything pulling power from the slot gets fair access to the power allotment, but the card will cap it to a set power limit, and it only knows how to limit power in one way, limiting clock speeds. so the fans and RGB can cut into the power budget, but taking them away doesn't prevent you from accessing the extra. it's not setup that way.

Then what you're saying could be true. Your card still sucks *** in the end. 
 

gsrcrxsi
edited my post a little for more context. I said it will pull back on clock speeds, but it also can pull back on core voltage. you can take the same exact card in the same exact system, not even two different cards, and run it at higher or lower voltage, with the same exact core clocks, just from applying different power limits. just overclock more on the lower power limit one, reach the same clocks, but on less voltage. this is a common undervolting technique. A lot of the Windows guys like to play with the V/F curves or other voltage specific tools, but in Linux, the first method is the only way to try to get this kind of control since voltage control in Linux is otherwise not possible. I use it all the time to run my 2080Tis at the same clocks as 250W, but under a power limit of 225W. and resulting performance is the same, just with less power used.
 
you ever consider that some people aren't hitting certain voltages as a result of the card trying to power limit? not the other way around.
 
you're trying to apply your theories from your FTW3 cards to the XC3 cards, and it's probably not applicable. not hitting 450W is a LOT different than not hitting 350W. from what limited data I've looked at with FTW3 cards, you guys aren't hitting the PCIe slot limit like we are, so the cause is probably not related.

So far from the logs I've looked at, everyone's voltage in TimeSpy graphic test2 has been the same, average clocks have varied but that's most likely do to boost bins, the lower temps the more boost.

You're probably right, leaked probably has something to do with it between your two other cards.