Hello guys, I have made some important discoveries with my FTW3 3080 Ti Hybrid and I want to share them with you. This probably applies to all the 30 series cards, specially from EVGA. If someone you know can draw a bit more power than you it can be various factors. One thing that will really reveal your performance compared to someone else is your "average voltage" in a synthetic benchmark. I'M NOT SAYING THIS EXPLAINS WHY CARDS LIKE THE 3090 CANT DRAW 500W. I'm saying this will explain why you might be drawing a different amount of power than another person even though you own the same cards.

What happened and why did I do these test. My FTW3 3080 Ti should have a board power draw of 450W max TDP but doesn't go anywhere near that. It averaged 412W in a TimeSpy Extreme graphic test 2 run. Another user averaged 445W. This made me want to investigate if I was losing performance compared to his power draw.

Power draw isn't everything and it's not something you guys should compare because I have discovered that a portion of power is saved for and dedicated to the RGB, fans and extra accessories you can attach to these cards.

What you guys need to do is compare one another's runs, graph it in excel and compare AVERAGE VOLTAGE.

Just because you are looking at "board power draw" doesn't tell you how much power is actually being used to render graphics.

If you are truly drawing less power than someone else, then that means you're voltages should be less than theirs. Power limits lower voltages. Measuring your voltage is the only way to know how your rendering capabilities are being affected by your GPU power draw.

I recommend running TimeSpy Extreme Test #2. Log the run with GPU-Z, extract the log, get it up into excel and get the "average voltage."

Compare your average voltage to someone else's. If they're truly drawing 30-50W higher than you, then their voltages should be higher than yours due to throttling from power limits.

Below are my findings. Thanks to Faux123 who shared his GPU-Z run of Timespy Extreme

Observations

• My setup differs from Faux123 who is running both of his PWM 120mm fans off of his GPU, I am not.
• For my setup, I have my fans drawing power directly from the PSU and controlled by my motherboard
• Faux123 results:
  • Fan 1: 100% (4.5W)
  • Fan 2: 100% (15.6W)
  • Average Power Draw 444W
  • Average voltage 0.969V
• My results:
  • Fan 1: 33% (0.4W)
  • Fan 2: 0% (running off of motherboard)
  • Average Power Draw 412W
  • Average voltage 0.972V
• Running Fan 1 at 33% consumed 0.4W
• Running Fan 1 at 100% consumed 4.5W

Edit: I was asked to do the same offsets as Faux123 (+150/+1000), the average voltage are within a margin (0.2%) of one another and can be seen here https://forums.evga.com/FindPost/3439610 
 
I also added Stock 100 vs 113 so you can see the difference in power draw vs. average voltage.

• Faux123 results:
  • Fan 1: 100% (4.5W)
  • Fan 2: 100% (15.6W)
  • Average Power Draw 444W
  • Average voltage 0.969V
  • Average core clock: 1934
  • Memory: +1000
• My Results
  • Fan 1: 33% (0.5W)
  • Fan 2: 0% 
  • Average Power Draw 400W
  • Average voltage 0.967V
  • Average clock 1983
  • Memory: +1000

 
I want to share my PL 100 vs 113

• 100
  • Fan 1: 33% (0.4W)
  • Fan 2: 0% (running off of motherboard)
  • Average Power Draw 390W
  • Average voltage 1V
  • Average Clock: 1887

• 113
  • Fan 1: 33% (0.4W)
  • Fan 2: 0% (running off of motherboard)
  • Average Power Draw 412W
  • Average voltage 1.055V
  • Average Clock: 1934

End Edit

Conclusions

The difference in power draw is due to the fans. If I ran the fans off of the GPU and ran both fans at 100%, I would have approximately 20W of more power draw. What is evident is our voltages. If we were both power limited to the GPU itself and if my card was really drawing 36W less then I would have a lower average voltage. But our voltages are close to the same. 
 
What this means is that that the RTX 3080 Ti consumes around 400W for rendering. The remaining 30-50W is saved for other things such as:

• Shroud fan for hybrid
• AIO 120mm fans
• Hybrid pump for AIO
• GPU fan for non-hybrid, 
• Case fans from PWM port on card
• RGB light bar
• External RGB strips

It's even easier just to look at the average MHz clock speed during the run. If they are blasting a much better average MHz clock speed, then they are getting more power.

I put a waterblock on my XC3 and the max power draw went from 280w to 250w while folding/mining.....better performance without the waterblock and with the fans pulling power from the board LOL. Definitely seems to be something with memory load, although I'm not sure how I lost wattage by removing the fans and adding a waterblock.

You're going to have to test back to back on the same card.  My 3080 FTW3 Hybrid has spent 99% of it's life with it's Rad fans powered by an external fan hub, with PWM control from the card, case switch, or MB.  445w average consistently on Furmark with 2X MSAA and +90 OC.  
 
My card has fairly even power balancing and can run moderately high 4K gaming loads at 1.075-1.1V and 2,100mhz with 400-425w.
 
I will certainly agree you need to get the voltage up with the 3080 to draw over 400w.  The 3090 and 3080 Ti can easily get over 400w at 1V.

talon951
Well my fans are only 1.56w each according to Arctic. So I wouldn't expect to see much change. I'm using the mobo mainly because I like using Argus Monitor for fan control.

https://www.arctic.de/en/...P120/ACFAN00167A#specs

0.13a x 12v

Hey Talon, the fans on the Hybrid are 0.65A at 12V

ObscureEmpyre
I thought this was known already since the power for the fans and LEDs has to come from somewhere. Anyway, I thought Kram's is a hydro copper version of the 3080 Ti. The XC3 Ultra, correct? My FTW3 came with a stock air cooler, but I slapped a water block on it. There aren't any fans connected, and I disabled all LEDs through PX1. My card is able to pull the full 450W, but only using Kombustor. That is literally the only program to have maxed out my card's power draw. A game like Cyperpunk 2077 will get close at around 430W, and 3DMark benchmarks are usually 400-425W. Kombustor, however, doesn't touch overclocks for some reason. Doesn't matter if it's +15 or +150, the max clock when running Kombostor for me is 1965 MHz, which is the card's natural highest boost clock with the power slider set to max in PX1. Regardless, what I'm trying to get at is that fans or LEDs shouldn't be a factor with power when it comes to Kram's card, and I doubt the LEDs are pulling that many watts. I'd think his card should be pulling max power or at least close to it if it's on water.

I have the same thing as you, I will hit 445W max in some loads and cases but I won't sustain it.

But techpowerup hit a spike of 530W on their test. I'm surprised they didn't blow a fuse.

Talon will find this part interesting because they use external equipment to measure the power draw. Similar to Talon's findings, the draws they saw seemed to be quite higher than what GPU-Z reported.



You mind doing the same test as I? Run TimeSpy Extreme with +120, +1000ish on the memory. What is your average power draw and voltage?

run timespy with a 0 memory offset. you'll see an increase in average core clock speed. though usually a lower score since the memory speed is beneficial to the score.

ObscureEmpyre

KingEngineRevUp

ObscureEmpyre
I thought this was known already since the power for the fans and LEDs has to come from somewhere. Anyway, I thought Kram's is a hydro copper version of the 3080 Ti. The XC3 Ultra, correct? My FTW3 came with a stock air cooler, but I slapped a water block on it. There aren't any fans connected, and I disabled all LEDs through PX1. My card is able to pull the full 450W, but only using Kombustor. That is literally the only program to have maxed out my card's power draw. A game like Cyperpunk 2077 will get close at around 430W, and 3DMark benchmarks are usually 400-425W. Kombustor, however, doesn't touch overclocks for some reason. Doesn't matter if it's +15 or +150, the max clock when running Kombostor for me is 1965 MHz, which is the card's natural highest boost clock with the power slider set to max in PX1. Regardless, what I'm trying to get at is that fans or LEDs shouldn't be a factor with power when it comes to Kram's card, and I doubt the LEDs are pulling that many watts. I'd think his card should be pulling max power or at least close to it if it's on water.

I have the same thing as you, I will hit 445W max in some loads and cases but I won't sustain it.

But techpowerup hit a spike of 530W on their test. I'm surprised they didn't blow a fuse.

Talon will find this part interesting because they use external equipment to measure the power draw. Similar to Talon's findings, the draws they saw seemed to be quite higher than what GPU-Z reported.

Yeah, those are transient spikes. Many high-powered electronics experience them, and they’re partly why CPU over clocks can be tricky if setting the LLC too high.

Yes and fuses have a required duration for them to blow. I'm guessing they use fuses that require at least a few seconds above their rating before blowing.

I found some fuses that at least look like the ones on the EVGA PCBs. Specs showed quite long durations particularly for a current only a few amps above the rating.

I was researching them because I'm pretty sure I exceeded 20 amps once or twice and was curious why the 2nd 8 pin fuse didn't let go. Probably the answer.

Swash87
Depends if it was drawing equally or if it was unbalanced though. Many cards draw near even power across the 3 pci-e's and higher wattage from the slot. 
If you factor in all 4 power sources, properly balanced it'll give you 525w of total power draw within Spec. So whilst eVGA could have "reserved" some headroom for accessories, theres over 100w in the power spec to play with on a 3080 and about 25w on a 3090 (500w bios).

Regardless of all of that, the 3080's only pull about 45-50w from the PCI-E Slot, wheres correctly balanced cards across the brands (in eVGA where balance isn't affecting the specific card) draw upward of 75w (the spec).
Just comes down to poor, cheaper component choice on eVGA's behalf (on whats supposed to be their higher tier offerings), couple all the larger tolerances from all the components with a statically set analog voltage controller (look at buildzoid's video ,if they had have just gone 1 step up to the digital version of the same controller which just about everyone uses on anything other than the bottom of the line cards, they could have just updated it and fixed all this, but they saved a buck and shot them selves in the foot with it), theres no surprise there's such a variance in what cards can hit (ie some 3080's can't get past 380w, others can't get past 400w, others hit 420w and even 450w) when you pick components with higher percentage "variance tolerance" in their construction

Fuse wise, they're always set higher than the Spec. ie PCI-E Slot spec if 6.25amps (75w @ 12v), and its fused at 10amps, same as the 8pins are 12.5amps fused at 20amps, along with them probably being a slow blow fuse, theres plenty of leadway there, they're there to stop the board going up in smoke if theres a major component failure (like a dead shorted vrm or something) as opposed to a load spike drawing a bit more power

Both our cards draw unequal

After reading all of your post, it will beneficial if all run exactly the same template of overclocking, the load of the memory also plays a factor.
Has anyone bother Ocing the PCIEx bus in the bios?
Please include that as well.

gsrcrxsi
overclocking the PCIe bus notoriously causes issues. I don't see how that would help anything though.

I understand that but the motherboard you guys have do they overclock the PCIe Bus automatically, when auto turbo is enabled?🤔

talon951
I mentioned this in one of the XC3 threads when someone commented that their power draw dropped when installing a water block. I saw a drop in total power when I went to the hybrid cooler and put the rad fans on the motherboard.  Whatever limit the cards hit is not changed by the fans (fan power comes from a different power rail/source on the board), so total power drops. 
 
20-30w seems like a lot though for fans.  Two 120mm fans don't pull that much power unless maybe you had a couple of 3k+ rpm fans or 4 of them in push/pull.  I think I saw about a 10w drop moving from the AIO to the motherboard header.

Edit: also, even using voltage to compare only works if offsets are the same as increasing clocks with offsets takes power which will drop voltage

I have done the test with the same offsets as Faux123 now. They are within a margin of one another in terms of average voltage. 
 

• Faux123 results:
  • Fan 1: 100% (4.5W)
  • Fan 2: 100% (15.6W)
  • Average Power Draw 444W
  • Average voltage 0.969V
  • Average core clock: 1934
  • Memory: +1000
• My Results
  • Fan 1: 33% (0.5W)
  • Fan 2: 0% 
  • Average Power Draw 400W
  • Average voltage 0.967V
  • Average clock 1983
  • Memory: +1000

 
I want to share my PL 100 vs 113

• 100
  • Fan 1: 33% (0.4W)
  • Fan 2: 0% (running off of motherboard)
  • Average Power Draw 390W
  • Average voltage 1V
  • Average Clock: 1887

• 113
  • Fan 1: 33% (0.4W)
  • Fan 2: 0% (running off of motherboard)
  • Average Power Draw 412W
  • Average voltage 1.055V
  • Average Clock: 1934

 
 

Buildzoid's analysis was valid at the time, and spot on. For sure EVGA would have done their users better to use the uP9512. But none of that is relevant to this particular situation.
 
The 3080 Ti doesn't use the uP9511, it uses the Onsemi NCP8610:
EVGA GeForce RTX 3080 Ti FTW3 Ultra Review - Circuit Board Analysis | TechPowerUp
 
As does the revised v1.0 3090 FTW3
(don't have a link, but its been reported various times around the forum)
 
And at least some non-Ti 3080 FTW's (although they are still labelled v0.1)
EVGA RTX 3080 XC3 en FTW3 Review - PCB-analyse - Tweakers
2004196198.jpeg (5292×3814) (tweakers.net)
 
The Onsemi NCP8610 can be ran in either analog or digital mode (and seemingly has I2C), but no idea if EVGA is using it in digital mode. I'd be surprised if they did, since they were switching to it from an analog uP9511.
 
(They also switched the power stages from AOZ to Onsemi)

MartinUK
 

ObscureEmpyre

I want to mention that this isn't just a observation for the FTW3. The ASUS Strix, MSI Suprim X and Palit Gamerock Oc  users I have met have experienced the same thing. Their cards kind of float around 400W, not 450W and 440W. This is the Strix below. 

 
 

Fuzzy833
So as a 3090 XC3 Ultra Hybrid user, would I see any tangible gaming performance increase by switching my 2 X 120 AIO rad fans from the GPU to the motherboard? Or are we talking very minor differences, only really noticeable in benchmarks? Thanks

I don't know unless you test. If I were to guess, very minor if not none for a FTW3 3090. I had observed similar things when I had a FTW3 3080, my power draw on the XOC BIOs was around 425W, but if I had been running the fans it would have been 450W. 
 
If you did test it, it would be nice to see. Do a test exactly with your same OC settings with the fans on and off the card. What's the average power draw? What's the voltage? Average clocks? 
 
If you had time to test this out, it would be very interesting to see if the XC3 is behaving just like the FTW3, a portion of power is dedicated to other things besides rendering.