MP 1,1-5,1 Mac Pro 5,1 (2010) dual GPU power limits

[CS20]

Hello!

I have a Sapphire HD7950 (flashed with Mac bios) currently installed in my Mac Pro 5,1. I would like to install a secondary Nvidia GPU for CUDA accelerated apps, I am considering either a Quadro K4000 or K4200. I am using the two 6 pin headers provided on the motherboard to power my AMD card. After reading a few other forums, the Mac should provide 75w for each PCIe slot and 75w for each 6 pin header which equals to 225w total. The AMD card draws 200w which is within the limits, if I install a K4200 (which draws 108W) and power it using a 2x 6 pin splitter on one of the headers, would it draw too much power on that header? Considering that the PCIe slot should supply the Quadro card with 75w and would only pull 33w from the 6 pin splitter, or should I play it safe and use a K4000 (which only draws 80w).

I would appreciate any advice on which Nvidia card should I use.

 

[MIKX]

AMD cards do not work well installed with NVIDIa cards.

Either two AMD GPUs or two Nvidia GPUs would be safer.

 

[tsialex]

Hello!

I have a Sapphire HD7950 (flashed with Mac bios) currently installed in my Mac Pro 5,1. I would like to install a secondary Nvidia GPU for CUDA accelerated apps, I am considering either a Quadro K4000 or K4200. I am using the two 6 pin headers provided on the motherboard to power my AMD card. After reading a few other forums, the Mac should provide 75w for each PCIe slot and 75w for each 6 pin header which equals to 225w total. The AMD card draws 200w which is within the limits, if I install a K4200 (which draws 108W) and power it using a 2x 6 pin splitter on one of the headers, would it draw too much power on that header? Considering that the PCIe slot should supply the Quadro card with 75w and would only pull 33w from the 6 pin splitter, or should I play it safe and use a K4000 (which only draws 80w).

I would appreciate any advice on which Nvidia card should I use.

Your understanding of power consumption of the GPU is incorrect. While the slot can feed the GPU up to 75W, only one modern GPU uses it fully, the reference AMD RX 480, almost all other GPUs since then use just around 30 to 45W from the PCIe slot and the PCIe power connectors feed the rest. Mac Pro PCIe AUX A and B are rated by Apple for 75W, with power protection shutdown happening around 110 to 120W. The way you intend to power the second GPU will cause power protection shutdowns, use the SATA connectors to feed the second one or, better yet, do a Pixla's mod.

Another thing to think, CUDA only works with NVIDIA WEB drivers and NVIDIA only offer it up to High Sierra - a macOS release that will have support terminated as soon as Apple release 10.16 around September/October this year, Apple only supports current release and the two previous.

 

[CS20]

I never read about the PCIe slots only providing half the specified power on older cards, I learn something new everyday.
Aren't Kepler based Nvidia GPUs natively supported in MacOS? That was the reason why I was going with a K series Quadro card. If I replace my HD7950 with a RX480, will my method of powering the cards work? But if CUDA only works with Nvidia drivers then there is no point of running the card then.

 

[tsialex]

I never read about the PCIe slots only providing half the specified power on older cards, I learn something new everyday.
Aren't Kepler based Nvidia GPUs natively supported in MacOS? That was the reason why I was going with a K series Quadro card. If I replace my HD7950 with a RX480, will my method of powering the cards work? But if CUDA only works with Nvidia drivers then there is no point of running the card then.

You still don't understand, the PCIe slot can feed 75W, but the card only uses 30 to 40W. It's the card power plane that control the power requirements, not the PCIe slot.

Apple native NVIDIA drivers don't support CUDA, only the NVIDIA WEB drivers.

 

[CS20]

Thank you for clarifying PCIe power limits, I had no idea that the card itself was responsible for how much power it can pull through the slot. Since High Sierra will be losing support soon, then there is no point in buying the card if the Nvidia drivers is only supported up to that version of MacOS. I appreciate your help, Thank you.

 

[Enthuser]

Your understanding of power consumption of the GPU is incorrect. While the slot can feed the GPU up to 75W, only one modern GPU uses it fully, the reference AMD RX 480, almost all other GPUs since then use just around 30 to 45W from the PCIe slot and the PCIe power connectors feed the rest. Mac Pro PCIe AUX A and B are rated by Apple for 75W, with power protection shutdown happening around 110 to 120W. The way you intend to power the second GPU will cause power protection shutdowns, use the SATA connectors to feed the second one or, better yet, do a Pixla's mod.

Another thing to think, CUDA only works with NVIDIA WEB drivers and NVIDIA only offer it up to High Sierra - a macOS release that will have support terminated as soon as Apple release 10.16 around September/October this year, Apple only supports current release and the two previous.

Hi, I have bought an ASUS GeForce GTX680 DirectCU2 which has both 6 pin and 8 pin pcie sockets I was going to flash it to Mac but am concerned on the power requirements from my MP 5,1 2010 running Catalina. Can you help with a workaround please or should I replace it? Thanks in advance. Mike

 

[MIKX]

You can power the GTX680 successfully using an EVGa PowerLink

This thread shows how.

EVGA PowerLink with AMD GPUs

Just got the EVGA PowerLink working with my Sapphire Dual-X 7970 3gb. IMPORTANT : The power input sockets on the PowerLink are 2 X 8 Pin female and cannot be changed. The EVGA PowerLink ensures that exact power demanded by a GPU is supplied. That said, one must still supply enough power to...

forums.macrumors.com

( shows my Sapphire Dual-X 3gb - 8 pin & 6 pin -powered by the PowerLink )
( The PowerLink was originally designed to power NVidia GPus )

Amazon ( there are cheaper offers on Amazon )

EVGA PowerLink, Support All NVIDIA Founders Edition & All EVGA GeForce RTX 2080 Ti/2080/2070*/2060*/Super*/GTX 1660 Ti*/1660*/1650/1080 Ti/1080/1070 Ti/1070/1060 0600-Pl-2816-Lr

The EVGA powerlink is designed to provide both a more stable power source and reduce ripple and noise, compared to connecting your power supply directly to the graphics card. The EVGA powerlink has another function: to make your PC look even more awesome. This adapter allows you to reroute the po...

www.amazon.com

 

[tsialex]

Hi, I have bought an ASUS GeForce GTX680 DirectCU2 which has both 6 pin and 8 pin pcie sockets I was going to flash it to Mac but am concerned on the power requirements from my MP 5,1 2010 running Catalina. Can you help with a workaround please or should I replace it? Thanks in advance. Mike

I don’t know this GPU but overclocked GTX 680 can shutdown your Mac Pro when feed just with the backplane PCIe AUX connectors, running benchmarks or heavy games.

Since it’s 8+6pin, if you can use an eVGA PowerLink with it, you probably won’t have power protection shutdowns.

If PowerLink won’t fit your GPU layout, Pixla’s mod is your last resort.

 

[MIKX]

With an extra 8 pin to 8 pin cable to extend the PowerLink's 'reach' it can be mounted in ' bridge mode' on top of the PCI area fan.

It 'should' fit on the GTX680 - if not . .then bridge mode WILL work.

 

[jscipione]

Your understanding of power consumption of the GPU is incorrect. While the slot can feed the GPU up to 75W, only one modern GPU uses it fully, the reference AMD RX 480, almost all other GPUs since then use just around 30 to 45W from the PCIe slot and the PCIe power connectors feed the rest. Mac Pro PCIe AUX A and B are rated by Apple for 75W, with power protection shutdown happening around 110 to 120W. The way you intend to power the second GPU will cause power protection shutdowns, use the SATA connectors to feed the second one or, better yet, do a Pixla's mod.

How do you know that the reference RX480 pulls 75W from the slot and what about the reference RX 580?

 

[tsialex]

How do you know that the reference RX480 pulls 75W from the slot and what about the reference RX 580?

AMD made a power plane design error, RX 480 initially was using more than the specified 75W, sometimes up to 90W, and sites stress testing reference RX 480 started to have burned PCIe slots - this was an enormous shi_storm back then and AMD rapidly released tweaked drivers changing the slot power consumption back to the specified 75W.

All initial reference AMD RX 480 since then are driver limited to the specified 75W, but Dell released a tweaked reference RX 480 with just one PCIe 6-pin connector that use more than the 75W from the PCIe slot, since Dell motherboards are engineered to provide more power from the PCIe slot. Not satisfied, Dell did even worse and modified the firmware to show Polaris 10 GPUs (Polaris 10 XT/Ellesmere is the RX 480) as RX 580 (Polaris 20). Those Dell cards are know to damage Mac Pro backplanes and several users here now have damaged slots.

All AMD GPUs since the RX 480 fiasco use between 30 to 45W from the PCIe slot and the rest are feed by the PSU via PCIe power connectors.

 

[thecrownguy]

What about have the two cards in the system but only use the Nvidia on Windows and then the AMD on Catalina?

 

[tsialex]

What about have the two cards in the system but only use the Nvidia on Windows and then the AMD on Catalina?

If you have Pixla's mod for powering the GPUs and if you disable the PCIe card with OpenCore…

A bad idea is still a bad a idea even if you write it differently.