Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Korie Cull

macrumors newbie
Original poster
Apr 26, 2020
7
2
Guys, I've been searching everywhere for answers about how many standard full-size GPU's can fit in the current Mac Pro.

I was looking to buy 4 Radeon Pro WX9100 as seen in this picture below but I'm not sure if the Mac can fit 4 with the current PCI setup.

58670_01_amd-radeon-pro-wx-9100s-require-1250w-psus.jpg


If I can't get 4 Radeon Pro WX 9100 inside the Mac, my next plan is to buy 4 Radeon Pro WX 7100 which are single card slots. Can someone verify if these would be ok?

Any help would be greatly appreciated!
 
Tech PowerUp says the TDP of the WX9100 is 230W with 1x 6-pin + 1x 8-pin power connector (75W + 150W = 225W).

The Mac Pro Technology Overview document says up to four 225W graphics cards can be connected (150W from 8 pin aux power + 75W from the slot). The four slots are double wide so the WX9100 should fit. 900W total. Seems like there might not be enough for four WX9100 except:
1) If using two MPX modules, it's 475W for the MPX slot, 75W for the PCIe slot = 1100W total for both MPX slots.
2) The Mac Pro power supply is 1400W.
So maybe it might work if the Mac Pro doesn't mind pulling more than 150W from each 8 pin aux connector...

The WX7100 has TDP 130W so there should be no problem with those (one 6 pin connector). You could install 7 of them but I don't know if macOS will use more than 4.
 
  • Like
Reactions: Korie Cull
^^^^But the NcMP 7,1 only has five power connectors for non Apple video cards. four 8 pin and one six pin.

Lou
 
  • Like
Reactions: Korie Cull
@joevt @flowrider Yeah the power connectors is what I'm trying to figure out. What I don't want to do is buy the MP and 4 GPU's and not be able to power them or even fit them all in. I need at least 4 GPU's for the upcoming launch of Octane X for Mac. This is a software based purely on GPU power utilising the Metal API, so I need as many supported GPU's as possible. I'd love to put 4 RADEON VII's in but I just don't think that's possible which is really disappointing. :(

 
@joevt @flowrider Yeah the power connectors is what I'm trying to figure out. What I don't want to do is buy the MP and 4 GPU's and not be able to power them or even fit them all in. I need at least 4 GPU's for the upcoming launch of Octane X for Mac. This is a software based purely on GPU power utilising the Metal API, so I need as many supported GPU's as possible. I'd love to put 4 RADEON VII's in but I just don't think that's possible which is really disappointing. :(

Two Radeon Pro Vega II Duos?
 
  • Like
Reactions: Adult80HD
You can't because there aren't enough power outputs, you can mix Apple Mpx modules with AMD cards.
The gpus you can use are:
N°5 installing 2x Vega 2 Duo Mpx modules + 1 pci WX9100 or W5700
N°4 installing 2x Vega 2 Duo Mpx modules
N°3 Installing 1X Vega W5700X Mpx module + 2 pci WX9100 or W5700
N°2 Installing 1X Vega W5700X Mpx module + 1 pci WX9100 or W5700
N°1 Installing 1X Vega W5700X Mpx module
 

Attachments

  • mac_5X.png
    mac_5X.png
    1.9 MB · Views: 713
Last edited:
You can't because there aren't enough power outputs
You would just need to use 8 pin to dual 8 pin splitters. The question is, how much above 150W can each 8 pin connector allow (up to 225W)?

What would be nice is a MPX to PCIe dual 8 pin power adapter (triple 8 pin actually, since the MPX slot should be able to do 475W). The power pin of the MPX slot is a huge/wide pin - just insert a PCB in that part of the slot with a bunch of wires sticking out for the 12V side and the ground side.
 
This would be more close to what I need on the left and what I have seen online for the size on the right. Why isn't this possible... 😭
 

Attachments

  • korie_cull_mac_pro_7.1_4_gpu_setup.jpg
    korie_cull_mac_pro_7.1_4_gpu_setup.jpg
    573.8 KB · Views: 1,245
  • Like
Reactions: Hps1 and Schismz
This would be more close to what I need on the left and what I have seen online for the size on the right. Why isn't this possible... 😭
The right has an 580X Pro which is removed from the left.
Otherwise, I don't see a problem.

Another note: the camera for the right side is much closer to the Mac Pro than the camera on the left side - notice that the left side has no 3D perspective.
 
You would just need to use 8 pin to dual 8 pin splitters. The question is, how much above 150W can each 8 pin connector allow (up to 225W)?

each 6+6 pin to 8+8 pin cable deliver 150W+150W = 300W

on the motherboard there are the following power slots:

n°2 6+6 pin power outputs for a total of 300W + 300W available for 2 non Apple gpus.
n°1 6 pin power output for 75W available (hard disks, audio/video cards, etc.)

Then all the available power for pci cards is 300+300+75 = 675W

you can split a single 8 pin cable but you will get 75+75W

The Radeon Pro WX 9100 has a 230W TBP
230 x 4 = 920W

Remember the Mac Pro 7.1 power supply can deliver a maximum of 1.400W. It can't power 4 full lenght gpus except for the Vega II Duo Mpx modules.

Also the Mac Pro 7.1 case is too small and the front axial fans are designed for Mpx passive radiators.

I have to tweak the fans speed to cool my two RTX2080Ti (bootcamp)
 
  • Like
Reactions: OkiRun
^^^^I would not be afraid of taxing the PS. Ik'd be more afraid of damaging a trace due to the extra draw.

Lou
 
Here's the real riddle you will need to solve before installing four (4!) AMD GPUs in a new cMP 7,1....

How can an AMD (Sapphire RX Vega 64 Liquid Cooled) GPU with an official TDP of 345W casually draw more than 500W when running the Windows 7 "Windows Experience Index" testing benchmark INSIDE a Parallels Virtual Machine?

Skeptics will say intermittent (or 'spike') power draw exceeding the rated/stated TDP of 345W by a full 50% is novel and a bit surprising but ultimately quite acceptable.

To those people I say...... "Hold my beer!"

When I ran the "Windows Experience Index" test and took 10 screenshots over a 60 second period - I was highly ALARMED by the SUSTAINED power draw during the graphics evaluation segment.


Screen Shot 2020-04-12 at 5.45.04 pm.png
Screenshot 1: The lowest value was 570W

Screen Shot 2020-04-12 at 5.44.45 pm.png
Screenshot 2: The highest value was 675W (..BOOM!)

Screen Shot 2020-04-12 at 5.45.39 pm.png
Screenshot 3: The random value at 30 seconds was 601W

CONCLUSION.. modern AMD GPUs provide/publish unreliable 'TPD' information in their product datasheets.

Expressed another way, [and in my experience] modern AMD GPUs draw power from the PSU as though the TDP parameter was left empty (null) in the BIOS before the boards were shipped.

Extremely uneven power draw (and the damage done) is a key theme that senior contributors on this website have described (in particular @tsialex ) in an effort to elevate technical awareness of this issue. I can personally confirm that these problems are NOT exaggerated - they are genuinely concerning and as you can see in my screenshots - even the most prodigious contributors to this forum may have underestimated the problem.

Circling back..... just imagine installing four (4!) AMD GPUs in a new cMP 7,1.... OUCH!

NOTE: My GPU test was 'fueled' by 2 PSUs - 1 internal (8-pins) and 1 external (8-pins)
 
Here's the real riddle you will need to solve before installing four (4!) AMD GPUs in a new cMP 7,1....

How can an AMD (Sapphire RX Vega 64 Liquid Cooled) GPU with an official TDP of 345W casually draw more than 500W when running the Windows 7 "Windows Experience Index" testing benchmark INSIDE a Parallels Virtual Machine?

Skeptics will say intermittent (or 'spike') power draw exceeding the rated/stated TDP of 345W by a full 50% is novel and a bit surprising but ultimately quite acceptable.

To those people I say...... "Hold my beer!"

When I ran the "Windows Experience Index" test and took 10 screenshots over a 60 second period - I was highly ALARMED by the SUSTAINED power draw during the graphics evaluation segment.


View attachment 910829
Screenshot 1: The lowest value was 570W

View attachment 910831
Screenshot 2: The highest value was 675W (..BOOM!)

View attachment 910832
Screenshot 3: The random value at 30 seconds was 601W

CONCLUSION.. modern AMD GPUs provide/publish unreliable 'TPD' information in their product datasheets.

Expressed another way, [and in my experience] modern AMD GPUs draw power from the PSU as though the TDP parameter was left empty (null) in the BIOS before the boards were shipped.

Extremely uneven power draw (and the damage done) is a key theme that senior contributors on this website have described (in particular @tsialex ) in an effort to elevate technical awareness of this issue. I can personally confirm that these problems are NOT exaggerated - they are genuinely concerning and as you can see in my screenshots - even the most prodigious contributors to this forum may have underestimated the problem.

Circling back..... just imagine installing four (4!) AMD GPUs in a new cMP 7,1.... OUCH!

NOTE: My GPU test was 'fueled' by 2 PSUs - 1 internal (8-pins) and 1 external (8-pins)
Spikes, the problem is always the instantaneous spikes. All modern GPUs do it and it's not a problem when a GPU is connected directly to the PSU, but it's a problem with Mac Pros and the GPUs being feed from the backplane.

Btw, TDP is correct, the problem is that TDP is the wrong metric here. Apple GPUs don't do it at all, just PC GPUs, btw.
 
Thanks Alex - is this likely happening with NVIDIA GPUs as well..? iStat Menus gives us beautiful diagnostics on AMD GPUs (so now I can see something that alarms me) but not for non-Mac edition NVIDIA GPUs - if I understand you correctly - the absence of apple-blessed NVDA kexts may mean this issue is just as bad with Nvidia GPUs. Cheers - useful insight for a theory I've been working on but it's bed time for me :)
 
Thanks Alex - is this likely happening with NVIDIA GPUs as well..? iStat Menus gives us beautiful diagnostics on AMD GPUs (so now I can see something that alarms me) but not for non-Mac edition NVIDIA GPUs - if I understand you correctly - the absence of apple-blessed NVDA kexts may mean this issue is just as bad with Nvidia GPUs. Cheers - useful insight for a theory I've been working on but it's bed time for me :)
NVIDIA GPUs does the same, just not as aggressively as with AMD GPUs. Tom's hardware has been warning about this for a long time, they use the best scientific power consumption measurement that I know of. It's a real pity that now all tom's reviews span for 40 pages and it's hell to get any info out of it.
 
  • Like
Reactions: Krevnik
I'd love to put 4 RADEON VII's in but I just don't think that's possible which is really disappointing. :(

I've got a pair of Radeon VIIs running in my current 5,1 build and am planning to add a third. The current two are running on a downvolt kext to 204v, and have been running beautifully. Even (as is happening a lot while I'm stuck at home in lockdown) playing for hours on maxed out 4k games, or fiddling around with raw video in Davinci Resolve.

If you downvolted the Radeon VIIs, you should (in theory) be able to run four of them in the 7,1 using the four onboard 150w 8-pin cables (each split into dual 8 pin connectors), plus the 75w from the PCIe board. That'll give you 225w per card - so you could even run all four at slightly less severe downvolt.
 
I've got a pair of Radeon VIIs running in my current 5,1 build and am planning to add a third. The current two are running on a downvolt kext to 204v, and have been running beautifully. Even (as is happening a lot while I'm stuck at home in lockdown) playing for hours on maxed out 4k games, or fiddling around with raw video in Davinci Resolve.

If you downvolted the Radeon VIIs, you should (in theory) be able to run four of them in the 7,1 using the four onboard 150w 8-pin cables (each split into dual 8 pin connectors), plus the 75w from the PCIe board. That'll give you 225w per card - so you could even run all four at slightly less severe downvolt.
Downvolting does absolutely nothing for spikes, VII has spikes of 330.35W.

Screen Shot 2020-05-01 at 02.31.54.png
 
  • Like
Reactions: tommy chen
I've currently got a Vega II mono and W5700X MPX modules in my 7,1. I'm curious if there would be enough room to throw a 2080ti in slot 5 above the two MPX modules? Might consider this option for bootcamp gaming at some point. Is there enough juice left in the PSU and enough space in case for this?
 
  • Like
Reactions: impulse462
I've currently got a Vega II mono and W5700X MPX modules in my 7,1. I'm curious if there would be enough room to throw a 2080ti in slot 5 above the two MPX modules? Might consider this option for bootcamp gaming at some point. Is there enough juice left in the PSU and enough space in case for this?
I’d be interested to know this as well.
 
MacPro 2019, 12 Core, 96 GB RAM, Radeon Pro 580X, Radeon RX Vega 64 8 GB

Total Power Consumption (nearly idle): 160-200 W, iStat reports for 580X 37 W, for Radeon RX Vega 53 W
Unigene Valley or Luxmark: iStat reports continuous power draw (with no spikes) for Vega 64 of more than 600 W

Measurement of the total power consumption of MacPro on the wall outlet: 460 - max. 500 W

I think iStat reporting is misleading. The numbers postet by tsialex (#19) fill the gap between 160-200 and approx. 500 W quite well.
 
I have my NcMP running through a UPS with a readout. I have an 8 core, 96GB Ram, 7 internal SSDs with an RX5700XT. The power draw on my APC is currently reading 240 W. Thats, with my monitor and my Klipsch speakers and a few other smaller items plugged into it. It rarely goes over 300 W.

Lou
 
  • Like
Reactions: OkiRun
I really want someone to just buy 4 WX9100's and just put them in to see what happens 😂. There's a guy on YT who messaged me. He's just brought 2 W5700X from Apple to see what space is left inside and to see how much power they use. I'm going to wait and see the results.

 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.