talon951
It's been shown by multiple people now that the card actually does reach 450w or even a bit more. GPUZ/HWINFO underreport the total power draw. This has been covered over and over.

And I've shown multiple times including just a few posts back in this thread that the difference in benchmark scores between the stock bios and the 1kw Galax is very small. 300-400 pts in PR.

i dont think thats true for everybody. my 3080ti ftw3 is showing 400 watt at 100% PL but it will not go above 108% PL in 3dmark. 
If you would be right, it would tell me that 108% is 112% in  Reality and even with 100% PL im over 400 watt. 
 
But on the other hand i found out that i can pretty easy hit 450 watt with this card showing in Gpu-z if i use OCCT stresstest.
This shows me also that this card is heavy throttling by over 50 watt if it hits 450. So i guess it does reach 450 also in benchmarks but we dont
see it as the intervall is to short.


Some numbers from my occt run default x1 setup with 450 W PL for 25 sec with a intervall of 0.1sec in GPU-Z so i got 250 samples. compaired to a curve optimized setting where i wouldnt stay under 450 W with a PL 112% ( i tried to find a setup with 440 watt average so i would be save below 450 Watt) (std=standard deviation)


 
 PL 450 / 450PL 450/440Max469,1 watt435,7 wattMin396,1 watt392,6 wattAverage446,7 watt427,8 wattStandard Deviation82,5663,31 
My theory is, that with this powerhorses its best to stay far away from powerlimit, as best es u can. U see standard deviation is much higher if u hit PL. At least you cut peaks al lot.  U see also u can hit 446 watt average readout from gpu-z with some heavy workload. I guess this holds ftw3 card back from hitting 450wat in "real World" benchmarks or games.  the second setting is not the best possible you can get closer to 450 PL but its long way to find the right setup for every case ( benchmark  game)
Thats what i thought before this test. you wont have a single setup for every usecase with this CUDA heavy cards.

ra773

talon951
It's been shown by multiple people now that the card actually does reach 450w or even a bit more. GPUZ/HWINFO underreport the total power draw. This has been covered over and over.

And I've shown multiple times including just a few posts back in this thread that the difference in benchmark scores between the stock bios and the 1kw Galax is very small. 300-400 pts in PR.

i dont think thats true for everybody. my 3080ti ftw3 is showing 400 watt at 100% PL but it will not go above 108% PL in 3dmark. 
If you would be right, it would tell me that 108% is 112% in  Reality and even with 100% PL im over 400 watt. 
 
But on the other hand i found out that i can pretty easy hit 450 watt with this card showing in Gpu-z if i use OCCT stresstest.
This shows me also that this card is heavy throttling by over 50 watt if it hits 450. So i guess it does reach 450 also in benchmarks but we dont
see it as the intervall is to short.

As I said, the power is reported low (by about 30w). So any power reading you see needs to have 30w or so added to get the true power draw. Yes this means if you see 450w reported, it's actually 480w. I verified this with a clamp meter to measure amps at each 8pin connector. The middle 8pin runs about 30w higher than reported. (yes meaning power balance is even worse than it appears)

I finally slapped the hybrid kit on my card... And I'm back in the top 100 again! My previous best on the stock air cooler was 15066. 15405 so far today and it's getting chillier outside 😂.

https://www.3dmark.com/pr/1417587

This was my previous best TS on air as well. Haven't gotten to messing with that yet. 21110

https://www.3dmark.com/spy/25266717

It says Galax right on the run report... And I wasnt too far off using the regular EVGA bios. https://www.3dmark.com/pr/1415602
Stop whining.

That's a sneaky ninja edit.

Anyway, I know how electrical components work so I am okay running the Galax bios for the occasional bench session. I'm sorry you didn't pay attention in class. You probably wouldn't hit those numbers anyway since it seems you got a poor binned card.


Bye.

What sort of tricks can I do to improve my score? I'm still trying to dial in my OC but are there any "tricks" I can do to improve things by a few points? I'm just running the benchmark normally with AB and HWinfo running. Does closing programs effect the results? I'm on air so I'm not expecting fantastic scores. Any help appreciated. Thanks!
Port Royal
https://www.3dmark.com/pr/1416266
Time Spy
https://www.3dmark.com/spy/25787842
 

Eek'ed out a bit more on the stock bios. Couldn't get temps enough lower to make a big difference though,

15464
https://www.3dmark.com/pr/1430927

I did manage to break 23k graphics in TS though,

23019 graphics,

https://www.3dmark.com/spy/26080108

This was easier than I thought. All I had to do was open my office window in the morning to let that cold air in and my scores went up several hundred points! Cracked my goal of 14600 so now I'm going to go for 15k.
 
https://www.3dmark.com/3dm/71849780

 got my 3080 ti pretty high and still ran kinda cool lol

talon951
Probably my last effort for this winter with the Galax bios. Personal bests for these,

15810
https://www.3dmark.com/pr/1460554

23183 gfx reBar off (reBar on was 233xx, but this gives my highest combined)
https://www.3dmark.com/spy/26568532

12052 gfx reBar off
https://www.3dmark.com/spy/26568732

Definitely is a Silcone winner you have there

No, only higher power limit. Yea I'm still using the hybrid, just sitting next to an open window. A good block would be 10C or more lower.

kraade
just wondering if you crank up the fans to precool the card as much as possible before the run?

Oh yea, 4 ML120's running full speed along with the hybrid fan.  It gains like 10C through the run as the coolant warms up.  Just such a small rad and volume of coolant.  And the hybrid block delta doesn't appear to be particularly good either.  Probably 20C or more at the power levels in the runs above.  From what I've seen comparing to others too, I don't think my mount can really get a lot better either.  Poor little thing just wasn't meant to dump 600w.  LOL

03whitegsr
Yeah, that's pretty good silicon I think. Mine seems to be decent and I don't think I could get it over ~2230MHz on the voltage limit on light loads. I don't think I really tried to push peak frequency though when I had it pulling cold air as the stock 450W only got me to ~2160MHz anyway.

I think it's just decent silicon, nothing that amazing. Like I think I talked about before (one of these threads), it's all about low temps, enough power to hold 1100mv, and a bunch of little tricks and tweaks to bump up the score.

talon951
I think it's just decent silicon, nothing that amazing. Like I think I talked about before (one of these threads), it's all about low temps, enough power to hold 1100mv, and a bunch of little tricks and tweaks to bump up the score.

Ah, you get rid of the repeat runs and the N/A temperatures (sub-ambient) and you're #3 on Port Royal for 3080Ti. Pretty sure that's better than "decent" silicon.

On a run you posted a while back, you are like 250-300 points higher to a run I have with nearly identical average clock and memory speeds, your tricks are definitely working. Nice job.

talon951

03whitegsr

talon951
I think it's just decent silicon, nothing that amazing. Like I think I talked about before (one of these threads), it's all about low temps, enough power to hold 1100mv, and a bunch of little tricks and tweaks to bump up the score.

Ah, you get rid of the repeat runs and the N/A temperatures (sub-ambient) and you're #3 on Port Royal for 3080Ti. Pretty sure that's better than "decent" silicon.

On a run you posted a while back, you are like 250-300 points higher to a run I have with nearly identical average clock and memory speeds, your tricks are definitely working. Nice job.

Keep in mind that the average clock 3DMark shows is not effective clock like you see in HWINFO. How you manipulate the VF curve can change the difference between reported clock and effective clock. So it's possible I had higher effective clocks than you and effective clock is what really determines the performance you get.

I assume you are forcing rebar on? That's 200-300 pts right there.

so you're running with rebar off to get higher scores? 

talon951
No forcing it on. It's good for 200-300 pts.

ohhhh thats in time spy. does it also juice up TSE? nice runs btw! I think you're in the top 3 for non sub ambient.