Question Asymmetric power draw on 3090 GPU connectors

[Marakai]

This may just be a question for clarification and confirmation, but I wanted to make sure there isn't an issue I need to dive into:

I have a brand new system with a MSI GeForce RTX 3090 Gaming X Trio 24GB . As a result I'm still running through benchmarks and stress tests to gauge performance and stability. In online comparisons my performance is good but the numbers always seem to be a bit below where they should be for a comparable system.

I tried some tweaking with Afterburner, but that was unsuccessful with being told that it couldn't stabilise due to power limitations.

I ran various GPU benchmarks (Superposition, Time Spy, Port Royal) while watching stats in GPU-Z and HWINFO. What jumped out at me was that the #3 8-pin connector was pulling significantly less power than #1 and #2. Just glancing at it right now on my idle machine, it's #1 59.3W, #2 45.8W and #3 11W. During benchmarks, #1 and #2 will jump to around 120W, whereas #3 stays behind at around 58-60W.

It should be noted that the *overall* power draw then is measured at around 350W - which is precisely what's documented for the card.

So, now I'm wondering if that notably lower draw on the 3rd connector is an indicator of a problem or actually shows that MSI is limiting the overall draw somehow?

Further info, as considered pertinent:

- 5800X CPU on MSI Unify X570
- Corsair RM850x PSU, each connector to GPU is a separate lead, no sharing

 

[MrTeal]

This may just be a question for clarification and confirmation, but I wanted to make sure there isn't an issue I need to dive into:

I have a brand new system with a MSI GeForce RTX 3090 Gaming X Trio 24GB . As a result I'm still running through benchmarks and stress tests to gauge performance and stability. In online comparisons my performance is good but the numbers always seem to be a bit below where they should be for a comparable system.

I tried some tweaking with Afterburner, but that was unsuccessful with being told that it couldn't stabilise due to power limitations.

I ran various GPU benchmarks (Superposition, Time Spy, Port Royal) while watching stats in GPU-Z and HWINFO. What jumped out at me was that the #3 8-pin connector was pulling significantly less power than #1 and #2. Just glancing at it right now on my idle machine, it's #1 59.3W, #2 45.8W and #3 11W. During benchmarks, #1 and #2 will jump to around 120W, whereas #3 stays behind at around 58-60W.

It should be noted that the *overall* power draw then is measured at around 350W - which is precisely what's documented for the card.

So, now I'm wondering if that notably lower draw on the 3rd connector is an indicator of a problem or actually shows that MSI is limiting the overall draw somehow?

Further info, as considered pertinent:

- 5800X CPU on MSI Unify X570
- Corsair RM850x PSU, each connector to GPU is a separate lead, no sharing

My 3080 FTW3 does that as well, it's normal. It just has to do with how the card is designed, but it's not indicative of a problem.

 

[Marakai]

My 3080 FTW3 does that as well, it's normal. It just has to do with how the card is designed, but it's not indicative of a problem.

Thank you for that. Though it may mean I need to dive a bit deeper to find the cause of my just-ever-so-annoyingly-lower benchmark results - and why Afterburner won't stabilise even for the slightest OC increase.

 

[alcoholbob]

It's normal, there are Youtube videos on this. The power limit on this card is pretty similar to dual 8-pin cards so the 3rd connector is hardly used.

 

[JoeRambo]

I ran OCCT 8 GPU stress test, and it seems that at least on my system all inputs are being used ~ equally:



So maybe it is total card power limit thing, if limited to 350W, the card won't pull much over 3rd connector. Card is MSI 3090 suprim x.

 

[Justinus]

Now that per connector power monitoring is available, I've been watching a few cards. It seems none of them load balance evenly, they bias toward specific connectors until those connectors reach their individual power limit then it begins drawing more power from the other connector(s).

Seems like it's just how they are.

Even my 980ti and 1080ti behave the same way. It's probably always been this way and we just never could observe the behavior before we were able to view individual connector power draw.

 

[Marakai]

Thanks for the replies. I've now been watching this closely, with running HWINFO and GPU-Z on another monitor. It seems that there is a hard power draw limit of around 350-ish Watts. In fact above-mentioned Afterburner tuning failed to stabilise due to power issues! MSI advertises the card as "pre-OCed" - but the numbers don't bear that out.

Looking at the numbers, however I now also noticed that the temperatures are possibly what leads to throttling: I'm a bit worried that they seem to get uncomfortably and constantly close to 90C. Just looking at the numbers right now on my idle system, they GPU is at 50, but memory and hotspot at 62-ish. One thing I rememer: when I first got the card and it started to get under load, there was almost a smell of heated lubricant that I could detect. Not burning, just warm oil, sort of thing. I haven't noticed any leaks or such, but I hope that wasn't something really bad.

The case is a Meshify2 and I have a front-mounted Alpenfoehn Gletscherwasser 360 AIO, which is doing an amazing job on the CPU, which even multi-threaded stress tests can't push much beyond 50C.

Otherwise there's 140mm fans, 2 on top, 1 towards rear. Yet, with all that, the GPU seems to be a furnace.

Could that be the issue?

 

[alcoholbob]

The card is hot because the default BIOS focuses on acoustics. AFAIK it or the Suprim are the quietest 3090s around. You can download MSI dragon center to switch fan profiles to a more aggressive one, or I think MSI offers a performance BIOS with higher fan rpms although it doesn't raise the TDP AFAIK so it's probably not recommended to go so far just for a fan profile.

 

[Marakai]

I was going to give the Low Temperature BIOS a try, but was a bit concerned that no matter where I searched, I could only find version 94.02.26.x, whereas when I tried to apply it, it detected a higher version 94.02.42.x or thereabouts already installed.

From https://www.msi.com/Graphics-Card/support/GeForce-RTX-3090-GAMING-X-TRIO-24G/ :

BIOS

Title

Low Noise BIOS (Default)

Version

94.02.26.80.C8

Release Date

2020-10-16

File Size

and:


Title

Low Temperature BIOS

Version

94.02.26.08.52

Release Date

2020-10-16

File Size

4.28 MB

 

[alcoholbob]

I was going to give the Low Temperature BIOS a try, but was a bit concerned that no matter where I searched, I could only find version 94.02.26.x, whereas when I tried to apply it, it detected a higher version 94.02.42.x or thereabouts already installed.

From https://www.msi.com/Graphics-Card/support/GeForce-RTX-3090-GAMING-X-TRIO-24G/ :

BIOS

Title

Low Noise BIOS (Default)

Version

94.02.26.80.C8

Release Date

2020-10-16

File Size

and:


Title

Low Temperature BIOS

Version

94.02.26.08.52

Release Date

2020-10-16

File Size

4.28 MB

Well if temps are a problem and you want to keep the same noise profile you can always undervolt. I get ~96% of stock performance at 260W TDP after running OC Scanner on MSI Afterburner and capping the power target at around 70%. Keeps temps nice and cool, I don't have a problem staying in the low 70s even at default fan curve.

 

[Marakai]

Well if temps are a problem and you want to keep the same noise profile you can always undervolt. I get ~96% of stock performance at 260W TDP after running OC Scanner on MSI Afterburner and capping the power target at around 70%. Keeps temps nice and cool, I don't have a problem staying in the low 70s even at default fan curve.

Actually, I'd love to give the low temp profile BIOS a try, but am wondering why the card currently has a version higher than what's on all their download sites. 😢