Can you imagine if it is a defective product how many cards they'd have to replace... Or the question is would they have to? I honestly don't think they'd have to but if it is due to a defective card issue and they would replace upon request with the updated product it would go a long way in making people again super happy with evga cs.

arestavo

tobincake1471
 
My 2014* SN 3090 FTW3 Ultra seems to top out at 491W during Time Spy, according to GPU-Z's Board Power Draw MAX value on the XOC VBIOS.  What is regarded as the definitive and most reliable way to test (which tool, reading, etc)?

It's average power draw, not max. EDIT: or current power draw if you can't get the average without it dropping too much from the before/after benchmark readings.
 
The folks with the chinese boards are able to get 480 to 510W average board power draw in tests like Port Royal, Time Spy Extreme, and games that put a high load on the card at higher resolutions (Doom Eternal from EVGA's Jacob Freeman himself). There is one outlier that does not from what I've seen in this thread.
 
The folks with the taiwan manufactured boards are able to get anywhere from 400 to 450W average board power draw in those same tests or games / resolutions. There are two outliers that are able to get 480W+ average that I've seen in this thread.
 
Using the 500W XOC VBIOS, of course. And typically monitored with GPU-Z.

Thanks a lot for this helpful post.  I just ran Time Spy Extreme and my avg board power draw was 458 (logged to a file from GPU-Z -> Excel -> calculated the average). 

I just refreshed the XOC bios and now my card is maxing out at 420w ish..... something is definitely going on and imma play with a some more and see what happens 
 
had to uninstalled X1 and re-install it
 
but with a 170mhz and 300mhz OC port run.... I'm 440w average with a 14290 score
 
can't seem to get this card to run over that 440-450w average 

vlcapata
I just refreshed the XOC bios and now my card is maxing out at 420w ish..... something is definitely going on and imma play with a some more and see what happens 
 
had to uninstalled X1 and re-install it
 
but with a 170mhz and 300mhz OC port run.... I'm 440w average with a 14290 score
 
can't seem to get this card to run over that 440-450w average 

The XC3 VBIOS will get you to 14.7/14.8K very likely. Again, I wouldn't use it for gaming (too much heat, fans not working correctly) - but for benching it will really show what you card can do.

Post approved, sorry for any delay

This PR run was done with +1000 mem +190 core.
I found the more you overclock this card the lower it wants to draw which contradicts the responses im getting from support. 
 

joshjd93
One thing that I have noticed in all of this, is the X1 shows the % TDP normalized, while every other monitoring software seems to show just the % TDP. These two values show in HWINFO. IF you watch the %TDP normalized, the card is always hitting and staying at whatever power limit I set regardless of the bios I use. (Normal, OC, OC XOC) I am wondering if the card is just watching the wrong TDP value and thats why it is going all over the place? The % TDP is usually less under load and is what is actually tied to power usage from what I can tell. This % TDP normalized is another value that the card reports, but isn't the TDP other manufacturers seem to use for power limits, but EVGA seems to have set this value as what the card watches when trying to stay within the power envelope.

I explained this earlier.
 
TDP "Normalized" (if the TDP slider is >100%) is simply the normalized TDP of the individual power rails with respect to the power limit of that rail.  If TDP Normalized is within 2% of TDP%, then it means all of the power rails are in perfect balance with each other, and you will *NOT* hit an individual rail's power cap before hitting the "Board" power limit.  TDP% is all the 8 pins and PCIE Slot power added together versus the max board power limit.
 
If TDP Normalized is much higher than TDP%, it means that an individual power rail is out of sync with the board power limit, and that an individual power rail may reach its own power limit (Triggering PWR) before you reach the board TDP limit.

Kylearan

joshjd93
One thing that I have noticed in all of this, is the X1 shows the % TDP normalized, while every other monitoring software seems to show just the % TDP. These two values show in HWINFO. IF you watch the %TDP normalized, the card is always hitting and staying at whatever power limit I set regardless of the bios I use. (Normal, OC, OC XOC) I am wondering if the card is just watching the wrong TDP value and thats why it is going all over the place? The % TDP is usually less under load and is what is actually tied to power usage from what I can tell. This % TDP normalized is another value that the card reports, but isn't the TDP other manufacturers seem to use for power limits, but EVGA seems to have set this value as what the card watches when trying to stay within the power envelope.

I explained this earlier.
 
TDP "Normalized" (if the TDP slider is >100%) is simply the normalized TDP of the individual power rails with respect to the power limit of that rail.  If TDP Normalized is within 2% of TDP%, then it means all of the power rails are in perfect balance with each other, and you will *NOT* hit an individual rail's power cap before hitting the "Board" power limit.  TDP% is all the 8 pins and PCIE Slot power added together versus the max board power limit.
 
If TDP Normalized is much higher than TDP%, it means that an individual power rail is out of sync with the board power limit, and that an individual power rail may reach its own power limit (Triggering PWR) before you reach the board TDP limit.

So only solution is to find which rail does it and mod the rail, aka pcie mod. But in the end if u want all together high pwr limit u have to shunt mod everything, this is my deduction I got from your explanation.

slovak_killer

Kylearan

joshjd93
One thing that I have noticed in all of this, is the X1 shows the % TDP normalized, while every other monitoring software seems to show just the % TDP. These two values show in HWINFO. IF you watch the %TDP normalized, the card is always hitting and staying at whatever power limit I set regardless of the bios I use. (Normal, OC, OC XOC) I am wondering if the card is just watching the wrong TDP value and thats why it is going all over the place? The % TDP is usually less under load and is what is actually tied to power usage from what I can tell. This % TDP normalized is another value that the card reports, but isn't the TDP other manufacturers seem to use for power limits, but EVGA seems to have set this value as what the card watches when trying to stay within the power envelope.

I explained this earlier.
 
TDP "Normalized" (if the TDP slider is >100%) is simply the normalized TDP of the individual power rails with respect to the power limit of that rail.  If TDP Normalized is within 2% of TDP%, then it means all of the power rails are in perfect balance with each other, and you will *NOT* hit an individual rail's power cap before hitting the "Board" power limit.  TDP% is all the 8 pins and PCIE Slot power added together versus the max board power limit.
 
If TDP Normalized is much higher than TDP%, it means that an individual power rail is out of sync with the board power limit, and that an individual power rail may reach its own power limit (Triggering PWR) before you reach the board TDP limit.

So only solution is to find which rail does it and mod the rail, aka pcie mod. But in the end if u want all together high pwr limit u have to shunt mod everything, this is my deduction I got from your explanation.

You know I'm Falkentyne, right?
But yes, basically that's it.
Usually on the eVGA boards it's PCIE slot.  However there's other stuff possibly going on as well...because PCIE slot limit should NOT change from bios version to bios version (i mean it can but that's...weird).
 
1) On shunt modded boards, PCIE Slot limit can have a strange effect on MVDDC power draw.  While there is no problem on a stock board (MVDDC may be around 50-95W depending on brand of board / AIB), modding all the shunts except the PCIE slot shunt has caused some people's MVDDC wattages to skyrocket to absurd levels.  Like 258W!  With PCIE Slot at its power limit cap of 79W, and the 8 pin rails (and of course, Board Power Limit) far below their limits.  When PCIE was shunted, MVDDC dropped way down.  All this despite the MVDDC being shunted.
 
https://www.overclock.net/threads/rtx-3090-founders-edition-working-shunt-mod.1773780/page-4#post-28656411
 
2) On boards where one 8 pin shunt is modded extremely poorly and the other modded decently, this has caused the *SRC* shunt to skyrocket in power draw.  Like if 8 pin #1 is reaching 175W (Triggering an 8 pin power limit) and 8 pin #2 is only at 140W, the SRC winds up reporting like 150W or something.  Once the 8 pins are modded correctly, for example, 8 pin #1 showing 154W to GPU-Z and 8 pin #2 showing 136W to GPU-Z, at 541W total board power draw, PCIE slot showing 53.4W (98.4% of TDP---board power base TDP=350W, board power current TDP=343W in GPUZ--this is multiplied by 1.6x to get my real power draw; Normalized is 104.6% with my TDP slider at 114%---this 6% delta between TDP and normalized, with both being below my TDP% slider, is an acceptable result!)----with this correctly done mod, the SRC now only reports 89.3W!  Almost 40% lower.
 
A stock unmodded FE card at 114% TDP (400W) seems to get SRC about 103W in Superposition.  My last heaven test that drew 541W max (no power limit reached) had 89W SRC and 104.6% normalized TDP and 98.4% of base TDP (343W GPUZ board power vs 350W base at 100% TDP).  My TDP slider was at 114%.  So maybe there's some relationship there between the stock value for SRC and what SRC is on a mod.
 
3) it's unknown if MVDDC or SRC have some power limit attached to themselves or what it is, but considering anethema had MVDDC at 258W...and I've already demonstrated that other power rails or lack of balancing can affect MVDDC and SRC (PP Input Source, or Power Plane), and they both have shunts.
 
4) There seems to be another thing that can trigger a power limit even with no rails anywhere close to their max limits.  Timespy Graphic Test #2 can trigger this in two locations for a few seconds, close to each other.

So i just ran couple port royal benchmarks first with 450 watt bios i hit max board power 455 watts with oc (160/1000)
Then i flashed 500 xoc bios and i got max 501 watts with oc (160/1000)


I have a 2014 black edition cards

slovak_killer
 
Sorry, nick Falkentyne doesnt ring a bell for me, again sorry Im rly bad with names, maybe if u can help me remember :)
Thanks for these insights, ive noted everything down, what u said seems to lineup with what Elmor told me.
Ordered EVC, when it comes im gonna mod the pcie (and rest of the shunts), nvvdd (FB vcore mod), pex (elmor suggested lower voltage on pex) 

Hi,
if I may have permission to ask, what did Elmor tell you exactly?  Some juicy power limit stuff?  Any details that might help expand my knowledge?

slovak_killer

Kylearan

slovak_killer

Kylearan

joshjd93
One thing that I have noticed in all of this, is the X1 shows the % TDP normalized, while every other monitoring software seems to show just the % TDP. These two values show in HWINFO. IF you watch the %TDP normalized, the card is always hitting and staying at whatever power limit I set regardless of the bios I use. (Normal, OC, OC XOC) I am wondering if the card is just watching the wrong TDP value and thats why it is going all over the place? The % TDP is usually less under load and is what is actually tied to power usage from what I can tell. This % TDP normalized is another value that the card reports, but isn't the TDP other manufacturers seem to use for power limits, but EVGA seems to have set this value as what the card watches when trying to stay within the power envelope.

I explained this earlier.
 
TDP "Normalized" (if the TDP slider is >100%) is simply the normalized TDP of the individual power rails with respect to the power limit of that rail.  If TDP Normalized is within 2% of TDP%, then it means all of the power rails are in perfect balance with each other, and you will *NOT* hit an individual rail's power cap before hitting the "Board" power limit.  TDP% is all the 8 pins and PCIE Slot power added together versus the max board power limit.
 
If TDP Normalized is much higher than TDP%, it means that an individual power rail is out of sync with the board power limit, and that an individual power rail may reach its own power limit (Triggering PWR) before you reach the board TDP limit.

So only solution is to find which rail does it and mod the rail, aka pcie mod. But in the end if u want all together high pwr limit u have to shunt mod everything, this is my deduction I got from your explanation.

You know I'm Falkentyne, right?
But yes, basically that's it.
Usually on the eVGA boards it's PCIE slot.  However there's other stuff possibly going on as well...because PCIE slot limit should NOT change from bios version to bios version (i mean it can but that's...weird).
 
1) On shunt modded boards, PCIE Slot limit can have a strange effect on MVDDC power draw.  While there is no problem on a stock board (MVDDC may be around 50-95W depending on brand of board / AIB), modding all the shunts except the PCIE slot shunt has caused some people's MVDDC wattages to skyrocket to absurd levels.  Like 258W!  With PCIE Slot at its power limit cap of 79W, and the 8 pin rails (and of course, Board Power Limit) far below their limits.  When PCIE was shunted, MVDDC dropped way down.  All this despite the MVDDC being shunted.
 
https://www.overclock.net/threads/rtx-3090-founders-edition-working-shunt-mod.1773780/page-4#post-28656411
 
2) On boards where one 8 pin shunt is modded extremely poorly and the other modded decently, this has caused the *SRC* shunt to skyrocket in power draw.  Like if 8 pin #1 is reaching 175W (Triggering an 8 pin power limit) and 8 pin #2 is only at 140W, the SRC winds up reporting like 150W or something.  Once the 8 pins are modded correctly, for example, 8 pin #1 showing 154W to GPU-Z and 8 pin #2 showing 136W to GPU-Z, at 541W total board power draw, PCIE slot showing 53.4W (98.4% of TDP---board power base TDP=350W, board power current TDP=343W in GPUZ--this is multiplied by 1.6x to get my real power draw; Normalized is 104.6% with my TDP slider at 114%---this 6% delta between TDP and normalized, with both being below my TDP% slider, is an acceptable result!)----with this correctly done mod, the SRC now only reports 89.3W!  Almost 40% lower.
 
A stock unmodded FE card at 114% TDP (400W) seems to get SRC about 103W in Superposition.  My last heaven test that drew 541W max (no power limit reached) had 89W SRC and 104.6% normalized TDP and 98.4% of base TDP (343W GPUZ board power vs 350W base at 100% TDP).  My TDP slider was at 114%.  So maybe there's some relationship there between the stock value for SRC and what SRC is on a mod.
 
3) it's unknown if MVDDC or SRC have some power limit attached to themselves or what it is, but considering anethema had MVDDC at 258W...and I've already demonstrated that other power rails or lack of balancing can affect MVDDC and SRC (PP Input Source, or Power Plane), and they both have shunts.
 
4) There seems to be another thing that can trigger a power limit even with no rails anywhere close to their max limits.  Timespy Graphic Test #2 can trigger this in two locations for a few seconds, close to each other.

Sorry, nick Falkentyne doesnt ring a bell for me, again sorry Im rly bad with names, maybe if u can help me remember :)
Thanks for these insights, ive noted everything down, what u said seems to lineup with what Elmor told me.
Ordered EVC, when it comes im gonna mod the pcie (and rest of the shunts), nvvdd (FB vcore mod), pex (elmor suggested lower voltage on pex) 

from the src physical perspective, it shows continuity to the shunt at the very top and the shunt at the top left (within the 4 group) not sure the reasoning... (i can't recall if it also shared with the pcie shunt or not, but the pcie rail and the src didn't complete, so not sure...
as well, power pin 3 shares the 2 shunts on the right side of the group of 4 along with the src, as well other pairings for the resistors for power pins 1 and 2
 
doesn't really change much for us at this point, but later on, when inclined, maybe i'll try mixing up resistances across all shunts and see what happens... any guesses? probably 2d clock lol
 

Any updates?

Well 14302 is the best I can do on stock OC bios: https://www.3dmark.com/3dm/54208938? 
Core +135
Mem +1000
Power 107%
Taiwan red lips
 
It pulls a max of 460 W as reported by GPU-Z. 
 
Anything higher on core is unstable. For TimeSpy or gaming I have to bring it down to +100. My TimeSpy graphics score sucks: 19600. But I am unsure if my CPU is holding me back there (even though CPU is not supposed to affect graphics score). 

lemuth80

Zgapzy
This PR run was done with +1000 mem +190 core.
I found the more you overclock this card the lower it wants to draw which contradicts the responses im getting from support. 
 

How you staying so cold???

Open case, cold ambient temps.
I figure if i can't get this card to run at 500w in these unrealistic conditions on air then it wouldn't do it for anybody running it in a closed case.