RTX 3090 / 3080 (3x8 pin AIB) with Corsair SF750 question

[talos91]

To my fellow SF750 owners, if anyone is getting an AIB 3x8 pin RTX 3080 / 3090, are you going to run the 2 x 8 pins from one rail and use the 2nd rail for the last 8 pin? Is this safe? Nvidia recommends using separate cable connections for each 8 pin connector, i was wondering if this was the same for Board partner versions.

 

 

I noticed the SF750 has 2x GPU cables and 2X CPU 8 pin cables which kinda make no sense for an SFX PSU. I don't understand why they would include 2 x 8pin CPU cables. I don't think i've seen an ITX board require more than 1 x 8 pin connector. Is it safe to connect the extra CPU 8 pin to the 3rd 8 pin connector on the 3080/3090?

 

 

Sorry for all the questions and bad structure, i'm just wondering what i need to do to prepare for my upcoming card. Thanks!

[Jurrunio]

16 minutes ago, talos91 said:

Is this safe? Nvidia recommends using separate cable connections for each 8 pin connector, i was wondering if this was the same for Board partner versions.

Nvidia follows PCIe power specs (unlike AMD which violates it from time to time) so this is just a "in theory better" thing, not a must.

 

17 minutes ago, talos91 said:

Is it safe to connect the extra CPU 8 pin to the 3rd 8 pin connector on the 3080/3090?

No, EPS12V for the CPU has two 12V pins in which are ground for PCIe 8pin, you will cause a short if you managed to do that.

[talos91]

1 minute ago, Jurrunio said:

Nvidia follows PCIe power specs (unlike AMD which violates it from time to time) so this is just a "in theory better" thing, not a must.

 

No, EPS12V for the CPU has two 12V pins in which are ground for PCIe 8pin, you will cause a short if you managed to do that.

Okay cool, thanks for clearing that up. So in summary, just use the 2 cables from the PSU to the GPU and use 3 of the 4 (8pin) GPU connections and i should be safe right? Thanks for the advice. 

[Jurrunio]

7 minutes ago, talos91 said:

Okay cool, thanks for clearing that up. So in summary, just use the 2 cables from the PSU to the GPU and use 3 of the 4 (8pin) GPU connections and i should be safe right? Thanks for the advice. 

Yep

[BTGbullseye]

3 hours ago, Jurrunio said:

Nvidia follows PCIe power specs (unlike AMD which violates it from time to time)

Source? (I'm interested)

[Jurrunio]

12 minutes ago, BTGbullseye said:

Source? (I'm interested)

RX 480 breaking PCIe slots on older boards? R9 295x2 literally drawing well over 400w with only two PCIe 8pin?

[BTGbullseye]

4 minutes ago, Jurrunio said:

RX 480 breaking PCIe slots on older boards? R9 295x2 literally drawing well over 400w with only two PCIe 8pin?

Again, a link please so I can read up on it?

[Jurrunio]

15 minutes ago, BTGbullseye said:

Again, a link please so I can read up on it?

https://www.google.com/amp/s/www.tomshardware.com/amp/reviews/graphics-card-power-supply-balance,3979-5.html

https://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-10,4616-9.html

 

Nvidias have shunts to monitor power draw from the connectors and the slot itself, but AMDs only monitor the GPU die.

[BTGbullseye]

2 hours ago, Jurrunio said:

Nvidias have shunts to monitor power draw from the connectors and the slot itself, but AMDs only monitor the GPU die.

The 5700XT has a sensor specific to total card power draw. (ASIC Power) It's not just Nvidia with that sensor.

2 hours ago, Jurrunio said:

https://www.google.com/amp/s/www.tomshardware.com/amp/reviews/graphics-card-power-supply-balance,3979-5.html

https://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-10,4616-9.html

Thank you for the links.

[Jurrunio]

2 hours ago, BTGbullseye said:

The 5700XT has a sensor specific to total card power draw. (ASIC Power) It's not just Nvidia with that sensor.

It is true that Navi gets memory power monitoring despite it not being in the GPU package (like Vega) but ASIC power in Hwinfo (previously Chip power) is NOT the full power draw of the card.

https://www.hwinfo.com/forum/threads/gpu-power-confusion.2930/

It's partly.monitored and the rest are calculated based on load.

and if you break down the name, Application-Specific Integrated Circuit has nothing to do with the whole card, the VRM, PCB traces, fans and more are not a type of ASIC.

[BTGbullseye]

10 hours ago, Jurrunio said:

https://www.hwinfo.com/forum/threads/gpu-power-confusion.2930/

May 2017, 2 years before Navi. Navi uses the ASIC sensor for full GPU power draw reporting.

[Jurrunio]

18 minutes ago, BTGbullseye said:

May 2017, 2 years before Navi. Navi uses the ASIC sensor for full GPU power draw reporting.

That's unfortunately the most up to date post I could find on hwinfo forums related to this. Any links I can read about it then? Cuz if you look at the BIOS, Navi cards always draw more than what the BIOS limit is set to. For example, Tom's HW found that the RX5700 reference card draws about 175w in games and 180w in Furmark, but at the same time the BIOS for the RX 5700 reference card has the power limit of only 150W. Why does the card not follow the limit written into the BIOS?

 

and since we're at it, can you show me where the power monitoring chip is on the PCB? To monitor the power draw of the whole card, this sensor must be an external thing in order to take VRM efficiency and PCB losses into consideration.

[BTGbullseye]

7 minutes ago, Jurrunio said:

That's unfortunately the most up to date post I could find on hwinfo forums related to this. Any links I can read about it then? Cuz if you look at the BIOS, Navi cards always draw more than what the BIOS limit is set to. For example, Tom's HW found that the RX5700 reference card draws about 175w in games and 180w in Furmark, but at the same time the BIOS for the RX 5700 reference card has the power limit of only 150W. Why does the card not follow the limit written into the BIOS?

 

and since we're at it, can you show me where the power monitoring chip is on the PCB? To monitor the power draw of the whole card, this sensor must be an external thing in order to take VRM efficiency and PCB losses into consideration.

I'm going off my own testing of having a power monitor at the wall, and comparing to the ASIC sensor in a way that is as unaffected by other system power draw as possible. Accounting for my PSU efficiency curve, the ASIC sensor comes within margin of error for total from-the-wall power draw accuracy.

[Jurrunio]

13 minutes ago, BTGbullseye said:

I'm going off my own testing of having a power monitor at the wall, and comparing to the ASIC sensor in a way that is as unaffected by other system power draw as possible. Accounting for my PSU efficiency curve, the ASIC sensor comes within margin of error for total from-the-wall power draw accuracy.

doesnt sound credible I must say. Also you cant rule out that it's only partially based on calculations and not fully measured.

[BTGbullseye]

1 hour ago, Jurrunio said:

doesnt sound credible I must say.

It's entirely up to you to believe or disbelieve, but seeing as you have no evidence to the contrary from the last 3 years, you're far more likely to be off than I am.

1 hour ago, Jurrunio said:

Also you cant rule out that it's only partially based on calculations and not fully measured.

Every rounded number for every sensor is based on a calculation, even when measured. Especially if they combine multiple sensors worth of information.

[Jurrunio]

29 minutes ago, BTGbullseye said:

but seeing as you have no evidence to the contrary from the last 3 years

I'm certain the tom's hardware one is within a year old

 

I actually started a thread for this in Hwinfo forums, we'll see.

[BTGbullseye]

8 minutes ago, Jurrunio said:

I'm certain the tom's hardware one is within a year old

And it confirms with what I'm saying though. ASIC sensor shows 180w at stock, power draw is at 180w. With the +50% power limit, it draws 225w according to the ASIC sensor, and within margin of error the same amount from the wall. (per my own tests)

[Jurrunio]

3 minutes ago, BTGbullseye said:

And it confirms with what I'm saying though. ASIC sensor shows 180w at stock, power draw is at 180w. With the +50% power limit, it draws 225w according to the ASIC sensor, and within margin of error the same amount from the wall. (per my own tests)

Then why doesnt the power limit value in BIOS apply to ASIC reading directly? Why use core power instead when it's possible to monitor the whole card?

 

btw if ASIC power is just the new name for chip power in the past, then this problem existed since the start of GCN

[BTGbullseye]

57 minutes ago, Jurrunio said:

Then why doesnt the power limit value in BIOS apply to ASIC reading directly?

I don't know what you're asking here. A power limit change shouldn't interact with the power draw sensing capabilities.

59 minutes ago, Jurrunio said:

Why use core power instead when it's possible to monitor the whole card?

Because the power going to the core is the most variable, and the most relevant to the functionality of the card. VRAM power rarely if ever changes, but core power fluctuates based on load.

1 hour ago, Jurrunio said:

btw if ASIC power is just the new name for chip power in the past, then this problem existed since the start of GCN

ASIC power is not chip power on the 5700 XT, it's total board power draw. (at least from my tests) This means power draw from the 8-pin, 6-pin, and PCIe slot added together. Chip power is a different sensor entirely. (the "Core Power" sensor)

[Jurrunio]

13 minutes ago, BTGbullseye said:

I don't know what you're asking here. A power limit change shouldn't interact with the power draw sensing capabilities.

power limit works according to power draw number, how you get the power draw number matters a lot.

 

14 minutes ago, BTGbullseye said:

Because the power going to the core is the most variable, and the most relevant to the functionality of the card. VRAM power rarely if ever changes, but core power fluctuates based on load.

1. Thinking secondary power rails always draw similar levels of power is why Furmark blew up so many old cards. Because individual power limit for each rail is too expensive, AMD and Nvidia end up reducing frequency significantly when the driver detects Furmark running in the short term and in the long run, beefed up secondary rails so they would survive Furmark

 

2. Not using whole card power draw to trigger power limit means the card can't directly protect PCIe slots and PCIe cables from drawing more power than specified

 

[Jurrunio]

Reply linked to this:

https://www.hwinfo.com/forum/threads/which-gpu-power-option-is-the-total-gpu-power.6676/#post-26569

 

so yeah, it's not a proper sensor to the level of Nvidia. Close, not quite.

[BTGbullseye]

7 hours ago, Jurrunio said:

Reply linked to this:

https://www.hwinfo.com/forum/threads/which-gpu-power-option-is-the-total-gpu-power.6676/#post-26569

 

so yeah, it's not a proper sensor to the level of Nvidia. Close, not quite.

That's not what was stated. 

 

"This is not quite clear, but "GPU ASIC Power" should be the closest to total GPU power, yet it might not include all rails."

 

There isn't enough info to prove one way or the other, and it may or may not provide accurate readings for draw from all power sources. Since it does match up with my 5700 XT testing for it to be accurate, it's safe to assume that the 5700 XT ASIC sensor is a total card power draw sensor that covers everything.

[jasonc_01]

On 9/6/2020 at 7:40 PM, talos91 said:

To my fellow SF750 owners, if anyone is getting an AIB 3x8 pin RTX 3080 / 3090, are you going to run the 2 x 8 pins from one rail and use the 2nd rail for the last 8 pin? Is this safe? Nvidia recommends using separate cable connections for each 8 pin connector, i was wondering if this was the same for Board partner versions.

 

 

I noticed the SF750 has 2x GPU cables and 2X CPU 8 pin cables which kinda make no sense for an SFX PSU. I don't understand why they would include 2 x 8pin CPU cables. I don't think i've seen an ITX board require more than 1 x 8 pin connector. Is it safe to connect the extra CPU 8 pin to the 3rd 8 pin connector on the 3080/3090?

 

 

Sorry for all the questions and bad structure, i'm just wondering what i need to do to prepare for my upcoming card. Thanks!

Why would you want a 3080/90 with 3 x 8 pin PCIe connector setup anyways with an sf750?

 

 

 

 

[Jurrunio]

6 hours ago, BTGbullseye said:

That's not what was stated. 

 

"This is not quite clear, but "GPU ASIC Power" should be the closest to total GPU power, yet it might not include all rails."

 

There isn't enough info to prove one way or the other, and it may or may not provide accurate readings for draw from all power sources. Since it does match up with my 5700 XT testing for it to be accurate, it's safe to assume that the 5700 XT ASIC sensor is a total card power draw sensor that covers everything.

https://www.hwinfo.com/forum/threads/amd-navi-card-power-monitoring.6683/#post-26591

this is the full answer

 

so it's meant to cover everything, but it cannot prevent overspec power draws down each source like Nvidia cards can

[BTGbullseye]

7 hours ago, Jurrunio said:

https://www.hwinfo.com/forum/threads/amd-navi-card-power-monitoring.6683/#post-26591

this is the full answer

 

so it's meant to cover everything, but it cannot prevent overspec power draws down each source like Nvidia cards can

"(and this is not just the case of AMD, but all other vendors)"

 

Nvidia is the exact same.