Very interesting that MSI bios worked.  I have given up trying to get more out of my card and just undervolted it to 1950 at 0.9v, with memory at +1000 and my temps are in the mid 60's.

Apaul82
Very interesting that MSI bios worked.  I have given up trying to get more out of my card and just undervolted it to 1950 at 0.9v, with memory at +1000 and my temps are in the mid 60's.

Yeah, its very interesting. It appears the card is at a minimum drawing up to the 430w of the MSI BIOS, but more likely is drawing at a peak somewhere between 450w and 500w. My only concern without proper testing equipment and having to try and rely on what my HX1000i reports is if the BIOS has just "broken"  the power limit to allow the card to draw whatever it wants, how does it balance this?
 
I don't know enough about PCBs to know if it will just naturally pull "safe" limits from each pin and the slot if it was hypothetically tricked into thinking it has no global limit. Because what I do know is each power pin can technically run above the power widely stated as their "safe limit". IIRC I believe people have been able to pull upwards of 200w through a single pin. Long-term if something like that is occurring I don't know how safe it is and what consequences there could be.
 
A 3 pin card should be able to handle 500w just fine, but the problem is if my card is pulling something like that I have no way to know how it gets to 500w, just the raw output from my power supply saying it's hit a 700w max spike. With a CPU pulling around 120~130w max, this leave something like a 570w balance to figure out between the rest of the components.
 
Which is why I really wish EVGA would look at this card and its BIOS again and figure out why their official BIOS is tapping out at 400w but this MSI BIOS seems to somehow remove that restriction.

If power readings are incorrect (and low), it will continue to pull more power until either a different limit is hit whether that be an 8pin, an internal limit, or the voltage limit.

All you need is a decent clamp meter and some 8pin cable extensions that have separate wires (that uses looms). That's what I did with my 3080ti when running the Galax 1kw bios with totally borked readings.

talon951
If power readings are incorrect (and low), it will continue to pull more power until either a different limit is hit whether that be an 8pin, an internal limit, or the voltage limit.

All you need is a decent clamp meter and some 8pin cable extensions that have separate wires (that uses looms). That's what I did with my 3080ti when running the Galax 1kw bios with totally borked readings.

That makes sense, explains why I might be hitting like 500w but a few interesting things to note

Both Asus and Gigabyte 3080 LHR bios for 450w don't work. While the power readings break like the MSI bios, it appears pin 3 becomes totally non-responsive. This appears to turn the card into a 2 pin even although both are 3 pin BIOS, resulting in no performance gain. Potentially even a minor performance loss coming it at under 400w max. Only the MSI bios works like I've discovered.

Overall the question now raised is what is wrong with the EVGA BIOS? Why are quite a number of cards unable to go above 400w on this official BIOS? This MSI bios, at least for me, as no one else seems to have tested it yet, allows my card to pull what appears to be upwards of 100w more.

EVGA simply won't talk about the 400w ceiling on many of these cards resulting in many thinking it was a hardware fault/PCB issue/poor power balancing controller. This throws a spanner into the works. Unless the power controller is still crap and as you pointed out all this MSI BIOS is doing is completely circumventing it. Obviously EVGA are unlikely to release a BIOS with effectively no power limit and state the card will just draw whatever it can. With no way to monitor how much power the card is drawing through software, I wouldn't say it's dangerous, but it's broken in terms of an official release from a manufacturer.

To a layman like me I still can't understand why only the MSI BIOS works, but I guess it's just something in the way they've configured their BIOS not present in the ASUS and Gigabyte ones. Maybe their 3 pin implementation is closer to how EVGA have done theirs?

Audioboxer

Cool GTX

Audioboxer

Memory starts crashing Time Spy at 1100, so 1000~1050 seems to be the best I can do.

Core clock starts crashing at +165, so I presume if I want to attempt to push more there I would need to play around with the voltage curve. I did that with my 2080Ti (used a curve at 2100 1.093v) but for now I think I'll just stick with +150/+1000~1050.

GPU score above improving likely needs core going higher, not memory.

I'm surprised at how well these 3080s watercool, but then again this EK Block also has a nice nickel backplate with thermal pads for some additional passive cooling. My 2080Ti Corsair waterblock just had a standard backplate with no cooling properties.

(I abbreviated your post)
 
Test with even lower Memory OC to see if performance increases   ... crashes are an outright failure of OC.  These new cards will use "error correction" .... so performance can suffer - before crashing occurs ..(edit) If you push the RAM too hard
 
NVIDIA Boost has the last word in your boards final numbers, been that way for several generations.
 
As long as your MHz meet the advertised minimum it is Fine; after that it is the luck of the silicon lottery
 
Getting Max performance before hitting the Watt limit is a Good thing in my book
 
EDIT:  added = (edit) If you push the RAM too hard - for clarity

Thanks, I will keep that in mind! It's definitely the core causing the crashing, I've even tried with the memory slider at 0 lol.
 
Speaking of the core, Port Royal will happily complete with an unstable core https://www.3dmark.com/3dm/73317219? That's +200 lol
 
Time Spy is pretty much locked at +150 for me https://www.3dmark.com/3dm/73317601? I got a better score than earlier now though, GPU was up to like 19762.
 
 Max power draw with Time Spy
 

 
400w

My Timespy Graphics score is similar to yours but your Port Royal score is a lot better than what I got.
 
I got curious about my 3080 FTW3 card and ran TS and PR to see if mine was going over 400W.  I think the most I got was like 407 for a split second but generally stayed at 400W.  This is a recent RMA KR that I got back last week.

I am having similar issues with my 3080 Ti, this is my fourth card... They at least sent me a BNIB 3080 Ti, but it's having this problem. Old card before had no problem pulling 450W.

fugly16

Audioboxer

Cool GTX

Audioboxer

Memory starts crashing Time Spy at 1100, so 1000~1050 seems to be the best I can do.

Core clock starts crashing at +165, so I presume if I want to attempt to push more there I would need to play around with the voltage curve. I did that with my 2080Ti (used a curve at 2100 1.093v) but for now I think I'll just stick with +150/+1000~1050.

GPU score above improving likely needs core going higher, not memory.

I'm surprised at how well these 3080s watercool, but then again this EK Block also has a nice nickel backplate with thermal pads for some additional passive cooling. My 2080Ti Corsair waterblock just had a standard backplate with no cooling properties.

(I abbreviated your post)
 
Test with even lower Memory OC to see if performance increases   ... crashes are an outright failure of OC.  These new cards will use "error correction" .... so performance can suffer - before crashing occurs ..(edit) If you push the RAM too hard
 
NVIDIA Boost has the last word in your boards final numbers, been that way for several generations.
 
As long as your MHz meet the advertised minimum it is Fine; after that it is the luck of the silicon lottery
 
Getting Max performance before hitting the Watt limit is a Good thing in my book
 
EDIT:  added = (edit) If you push the RAM too hard - for clarity

Thanks, I will keep that in mind! It's definitely the core causing the crashing, I've even tried with the memory slider at 0 lol.
 
Speaking of the core, Port Royal will happily complete with an unstable core https://www.3dmark.com/3dm/73317219? That's +200 lol
 
Time Spy is pretty much locked at +150 for me https://www.3dmark.com/3dm/73317601? I got a better score than earlier now though, GPU was up to like 19762.
 
 Max power draw with Time Spy
 

 
400w

My Timespy Graphics score is similar to yours but your Port Royal score is a lot better than what I got.
 
I got curious about my 3080 FTW3 card and ran TS and PR to see if mine was going over 400W.  I think the most I got was like 407 for a split second but generally stayed at 400W.  This is a recent RMA KR that I got back last week.

If you flash the MSI 3080 BIOS I linked to it will pull upwards of 500w, and possibly even spikes exceeding that. It's difficult to measure when the software reporting is broken, but my crude measurements based on my power supply draw seem to indicate that. As well as watching what my water temp and card temp get to.
 
Worth it? Well, if you have the cooling capacity and aren't happy EVGA released a broken 450w BIOS, why not? However, instead of playing around with 2200mhz at 1.1v with my new power overhead, I've settled for locking in at 2100mhz 1.050v. That can be maintained in Metro Exodus EE even with RTing, so seems like the best balance for me between power, thermals and not hitting power limit constantly in one of the toughest gaming loads.
 
2200mhz at 1.1v for example is power limited in Metro, games without RTing or less intensive games can maintain it (even most of early Timespy can), but it's really entering "bragging rights" only territory versus a good balance for daily running.

Really strange that a BIOS from a different company with a lower (430w) advertised power limit allowed your card to push past 400w. Have you ever tried the Evga Hybrid bios? I really hope that changing to a 16pin connector will alleviate these problems from future generations. Pretty ridiculous that you're having to jump through so many hoops. This is something that should have been handled a year+ ago.

Yup.. and stupid thing is on the Evga 3090 Ftw3 ultra using their xoc 500 watt bios, I cant hit more then 460watts.. And best part is it will hold power limit from 424watts and up and tank the clock.

Though.. if I use the kingpin 520watt bios I can run over 500 watts! Barely ever does it report power limit hit and stable clocks.

bzabrown419
Really strange that a BIOS from a different company with a lower (430w) advertised power limit allowed your card to push past 400w. Have you ever tried the Evga Hybrid bios? I really hope that changing to a 16pin connector will alleviate these problems from future generations. Pretty ridiculous that you're having to jump through so many hoops. This is something that should have been handled a year+ ago.

I think going with 3rd party BIOS might disable your safeguards on the GPU, unless that's already happening with the EVGA versions.

AHowes
Yup.. and stupid thing is on the Evga 3090 Ftw3 ultra using their xoc 500 watt bios, I cant hit more then 460watts.. And best part is it will hold power limit from 424watts and up and tank the clock.

Though.. if I use the kingpin 520watt bios I can run over 500 watts! Barely ever does it report power limit hit and stable clocks.

That's really odd.  Were others able to hit 500w+ with the same GPU you're using?