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

MacScout

macrumors newbie
Original poster
Nov 21, 2016
19
4
Made a video showing how I modified my power supply to install the Radeon VII:
 
  • Like
Reactions: mikas

eicca

Suspended
Oct 23, 2014
1,773
3,604
Where is everyone getting their Radeon VIIs? Of all things that I can’t seem to be able to purchase anywhere, Radeon VIIs are apparently hard to find.
 
  • Like
Reactions: Eschers

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Thank You for the video.

I grabbed two of those just recently. First one for 699€ from abroad, and then accidentally waiting for it to be delivered, found one for 360€ domestic. I think prices might be coming down now, while people are waiting for RTX 30xx avalability and maybe Big Navi stuff too, and thus getting rid of their older equipment.

Just did a test with 2x 6-pin power to make sure that they do work. I have to make Pixlas mod now to go ahead and stress test them too. I've got a couple of Vega 56s (8GB) now, and they are just fine. But I do need that 16GB of VII for my bigger models to fit in the VRAM (Twin Motion).

I bought some cheap clips already, but I am not quite sure if I will use those or am I gonna find some more professional looking ones. Any links for good clips? Or should I just go ahead and solder them to wires, or maybe straight to PSU like some users have done.

Or maybe try all of those choices, hah. I've got three Mac Pro 5,1s at home now.

My cheapo clips in pics with some GPU options:
1602663682194.png

1602663715104.png

1602664319255.png

With these the added extra wiring will attach parallel to PSU original wires, like 90° different direction from those wires in MacScouts video.

I plan to try 2 VIIs simultaneously too, and I thought to just do that with splitting cables. I am not sure if I needed to make a double Pixlas instead. I take double Pixlas as using double the amount of clips connected straight to PSU wires.

Why I am not sure about this is because I doubt if the connectors, cables and clips will all handle the wattage going thru them.

There are a lot of testimonies about Pixlas and the methods to do it, and I think I have read them all. So thank you again for this additional piece of information.
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Ok, I decided to buy/order different kinds of crimps. I want to match the AWG numbers right for the wires and tabs (12AWG and 18AWG). That is what you seem to have done. At least I believe so, based on those color codings from your video? So I ordered T-tap wire connectors of 3 different sizes (including in the range of them are AWG12 and AWG18).

1602758911729.png
1602758911729.png 1602758884255.png
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
It's been long due, but finally took the time. Changed my plans at least a couple of times with parts i'd use for this. But it's in working order now, my first Pixlas modded cMP (4,1->5,1).

2 x VII's are showing me 100k+ LuxBall numbers (LuxBall ver. 3.1). Now I'll get going with the cinema 4D and twinmotion, ArchiCAD and some Rhinoceros viewports. Maybe I'll see better navigation in Solibri too.

First I thought I'd push it to three of VII's, but I don't think the power supply could have done that. And I would have lost a PCIe NVMe drive too, see pics (I did not power up with three VIIs ever). So now I'm testing 2x VII's: one VII with Pixlas mod enabled power feed, and the other one with 2xmini6pins+2xSATApower and a powerlink in between here to balance the power. Works just fine.

sorry about the pics flood. I'm a little bit excited about this. Finally had/took the time for this.
1609085666686.png

1609085714152.png

1609085769098.png


1609085920371.png
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
For those still hesitating, I want to say. Take your time, and it will go allright. I did use the knife too, as to make sure the clip will reach contact. I peeled the skin, and placed the clip carefully in line with the peel. You want to have appropriate tools available too, otherwise it's gonna fail I am afraid. please see another pic flood still.

I made a black mark to all those ground wires, just to make it failsafe for me.

1609090184585.png

1609090238919.png

1609090486284.png
 
Last edited:

h9826790

macrumors P6
Apr 3, 2014
16,656
8,584
Hong Kong
It's been long due, but finally took the time. Changed my plans at least a couple of times with parts i'd use for this. But it's in working order now, my first Pixlas modded cMP (4,1->5,1).

2 x VII's are showing me 100k+ LuxBall numbers (LuxBall ver. 3.1). Now I'll get going with the cinema 4D and twinmotion, ArchiCAD and some Rhinoceros viewports. Maybe I'll see better navigation in Solibri too.

First I thought I'd push it to three of VII's, but I don't think the power supply could have done that. And I would have lost a PCIe NVMe drive too, see pics (I did not power up with three VIIs ever). So now I'm testing 2x VII's: one VII with Pixlas mod enabled power feed, and the other one with 2xmini6pins+2xSATApower and a powerlink in between here to balance the power. Works just fine.

sorry about the pics flood. I'm a little bit excited about this. Finally had/took the time for this.
View attachment 1701970
View attachment 1701972
View attachment 1701973

View attachment 1701974
3x Radeon VII may work actually if you downvolt it properly.

Let's say you use my default 204W kext (link), then 3x204 = 612W

The cMP itself (including RAM, SSD, etc) let's say need 150W

And we assume the CPU always draw 130W

Then total still under 900W, may be OK for a 980W PSU.
 
  • Like
Reactions: mikas and Eschers

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Thanks for reminding me about that. I had read about that, but totally forgot it. You have very well documented it all in there. I'll check the link carefully.

I do not really _need_ 3x VII in my cMP, but I am still tempted to temporarily test it that way. Just for the sake of it.

I think i'll manage with only 1x VII with my daily workloads. Cinema 4D R21 Prorender did not give me a lot of promise yesterday. It could be the single X5680 cMP being too slow to prepare the Prorender scene for those VIIs. I don't know, I'll test more later.

TwinMotion does not support but only one GPU at this time unfortunately. I would need to decide a more capable GPU renderer. And we all know the state of those on MacOS / Macs today, especailly now that there is Apple Silicon on the table too. It's all too vague right now to make any big decisions.

GPU rendering on the new Mac Pro. News, updates, chat.
 

h9826790

macrumors P6
Apr 3, 2014
16,656
8,584
Hong Kong
Thanks for reminding me about that. I had read about that, but totally forgot it. You have very well documented it all in there. I'll check the link carefully.

I do not really _need_ 3x VII in my cMP, but I am still tempted to temporarily test it that way. Just for the sake of it.

I think i'll manage with only 1x VII with my daily workloads. Cinema 4D R21 Prorender did not give me a lot of promise yesterday. It could be the single X5680 cMP being too slow to prepare the Prorender scene for those VIIs. I don't know, I'll test more later.

TwinMotion does not support but only one GPU at this time unfortunately. I would need to decide a more capable GPU renderer. And we all know the state of those on MacOS / Macs today, especailly now that there is Apple Silicon on the table too. It's all too vague right now to make any big decisions.

GPU rendering on the new Mac Pro. News, updates, chat.
It’s too bad that those M1 Mac doesn’t support eGPU. Otherwise, we can enjoy the highest CPU single core performance + very high GPU power.

According to Apple, all Mac will use Apple silicon in two years. Let’s see what will they release for the Mac Pro. And if that still has PCIe slot, and support dGPU.
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Exactly. If they did support eGPU I might have bought a Mac Mini allready (16GB/1TB).
That's despite the fact that not all of my programs are supported on Apple Silicon. And there is a chance they never will be either. The 7,1 is not worth it for me with all that money I'd need to throw at it. Newest OSes are not behaving very well with the software I am using, Mojave seems to be the best with them.

And because of all this uncertainty everything is hanging in the air with Macs for me now. The anticipated Mac Pro mini / Mac Pro 8,1 / xMac or whatever it's gonna be will tell wheather I have to leave MacOS for good or not.

So I'll just keep working with my cMPs for now. At least I got some more graphics power now.
:cool:
 

Eschers

macrumors member
Oct 27, 2015
86
35
ch
Where is everyone getting their Radeon VIIs? Of all things that I can’t seem to be able to purchase anywhere, Radeon VIIs are apparently hard to find.
I guess its the best GPU one can buy for the cMP 4,1(5,1) since it is natively supported in the latest available OS for this machine (Mojave 10.14.6)

ps: had one myself, but one day it did not display anything and had to send it to rma, but they did not replace/repair it and instead refunded.. I am also looking on ebay and other sited for a reasonably priced used one since its hard to find a new one.
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Ok, I was happy with my Vega56. But ArchiCAD 3D navigation viewport basically flyes with one Radeon VII in cMP. I think I have to start modding another PSU for Pixlas. Fortunately I've got one 4,1 taken apart right now. I'll mod that readily available PSU, and then change it to my main work machine. So my actual work pause is minimal. Afterwards I can mod the last one too.

It's good to have some spare machines available.

Cinema 4D + ProRender viewport is faster with 2x VII than RX580 or Vega56, a lot, I have to admit it. But not as fast as some other GPU realtime renderers I have seen. VII's are both used as can be seen from screenshot. I tried this with older C4D (R21), so maybe that's why. And it crashed a couple of times too when preparing the scene. That's pobably not the fault of the VII, I think that phase is done in CPU. The trees and vegetation might stil be too complex for ProRender, at least for this version that I tried it with. Maybe I should upgrade, don't know.

Simple Clay model of a building works ok. Not so with trees and vegetation. I tried couple of Maxon bundled architectural scenes.

C4DR21_w_2xVII.png
 
Last edited:
  • Like
Reactions: h9826790

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
I hope it's still ok to use this thread you started MacScout?

I'm having this feel you all know. You are all set up and thrilled and ready to go - and then, something is missing.
Dang, you have to pull of, pull out, halt or whatever.

Well I was so ready to try that aforementioned "Dual Pixlas", but I am missing two PCIe pins. See pictures.
1615479242591.png

Oh my, this has to wait still for a couple of days or a week. ModDiy sent me open ended 6+2 pin, 6 pin cable. I'm quite sure I had ordered 6+2, 6+2.

I'm so disappointed that I'll give it a break now. I do have necessary means to bypass this, but let's just wait for a minute.

Phuuh. I am waiting for 150000 LuxBalls from this. And maybe sometime later test the new Octane software too.
ps I had to mod the PCIe bracket of that upmost card, so it would fit in to the Mac Pro.

1615479550391.png

It will be easy to pry back, not that hard of a metal plate, not this one.
1615479680241.png

I could have removed the bracket all together, but one of the screws was under the card assembly. It would have required the disassembly of the hole card. I did not want to do that this time, that's why I gently pryed the bracket into something like 70-80 degrees angle. Fits real nice. Even the PCIe securing plate fits in and tightens allright.

So no benchmarks nor scores this time. Just relieving my dissappointment sharing it all with you.. :)
 
Last edited:

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Ok, a little late like everytime it seems, but now they are finally all lit up in a cMP 2009 single, with Dual Pixlas and with 2xmini 6-pin + 2xsata power through EVGA powerlink.

I did install CustomPowerPlay.kext (204W limit) by Martin Lo, but it did inform me aboutsome errors, which do not necessarily mean anything. SIP was and is disabled. I already ran 2x VIIs at 600W+ overall system power consumption (=100k+ LuxBall)

Now it's time for 3x VIIs with LuxBall.

I've got HW monitor launched. I've got system monitor graphics card history launched.
And I've got one red emergency button here too in the near reach. I am not gonna go and get my fire extinguisher though.

1617634014732.png

1617633979281.png

1617633960777.png

Just checking every cable one more time..
 
  • Like
Reactions: MoerBoer

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
Well I don't think I get into top battles in Luxmark charts with this, but that's an 12 year old computer with over 2 year old GPUs in it.
1617634844841.png

Watts did stay just under 900W through all the test. A 886,6W was highest I could observe. HW monitors frequency to record watts did not catch that one. I am pretty sure though it was under 900W all the time.
1617635044776.png

Now I need to do some calculations if PowerPlay actually was in effect or not. What do you think about it? I am sure somebody has tested with this allready?
 
Last edited:

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
It seems I draw 245W per VII with LuxMark. So I have to conclude the CustomPowerPlayTable-kext is not in effect at the moment.
System draws somewhere in between 100 to 150 idle (from low to peak Watts).
150+3x245=885W
PPT (PowerPlayTable) would not be working at the moment, wouldn't you think the same? I would not want to stress 3x VIIs with something even more demanding before I get that PPT.kext to appear to be effective. I need to try the procedure again. Maybe I need to try to understand what's going on in there too. The extension system, and the table, and the injecting of extensions.

I'll put the thing to sleep for a while. And me too.
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
After some sleep I had better eyes and a tiny bit sharper mind too I think. The kext was indeed installed in to the wrong library folder. Phuuh, what an amateur..

Corrected the situation and went through necessary procedures again. Now it seems to be working right according to my measurements.

705,6W seems now to be the ceiling with my 3x VIIs with LuxMark.

Lets approximate power usage during the test:
System= 150W
3x VII= 555W
That would make only 185W per VII

According to HW monitors Long Term Extreme Values per one hour, to me it seems quite a legit number.
Some of you might have better insights to this, if so please share?

Got just a little bit better luxball result too (1,6%, pretty much non significant). Now with these numbers I think I would be brave enough to test something that would draw more power than what LuxMark usually does. Games bechmarks, do they measuren multiple GPUs simultaneously (Unigine for example). I don't think so.

There is still Cinema 4D ProRender, but it's the old version (C4D R21) and it's a little bit in between becoming production quality product. To say it nicely.

1617677125658.png

1617677351308.png
 
  • Like
Reactions: h9826790

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
A few tests still before tearing this thing apart again. Some people might find this interesting, at least I hope so.
Cinema 4D R21 with ProRender. A huge speed-up with 3xVII to 1xVII in preview times. With final renders I got some mixed results which I can't explain - for better and worst.

Sedan refreshes in something like 2,5 seconds to look like that left picture. With one VII it seemed to settle at 5-6 secs. It's the viewport in which you edit your model, the work encironment. It does look really good, doesn't it.

Final render at right in attached screenshot took about 1:15 (75 sec) to accomplish itself. You can observe the differences betweemn the two easily, like transparency and the subtlety of shadows and reflections. But the preview too truly is a workable representation of that car object. It takes 2-2,5 seconds to finalize the preview at default settongs. You don't need to wait that long, you can re-navigate or move around at anytime of course, and refresh starts immediately again.

1617733539024.png


But I've been studying the internet. Some grayscale gorillas say this is not a notable render engine, but there are much better alternatives to it - numerous even.

I think I will use these VIIs better serving alone in separate machines right now. I am not going to invest in GPU rendering software while Apples/Macs/Tims status and intentions in this field has no visible or given roadmap. I just had to get a short peek in there.
 
  • Like
Reactions: h9826790

KeesMacPro

macrumors 65816
Nov 7, 2019
1,453
595
I've been following your progress with great interest from a technical point of view.
I'm not really into graphics , i work with audio on a MP4,1>5,1 so i dont know the possibilities of multiple GPUs.
Apart from rendering , are there more advantages with a multiple GPU setup?
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
This has been educating and fun project for me. I'm not into video or audio that much, I am an architect, and I am very much into 3D through my profession. I do produce videos and graphics too at least in a way, though they appear mostly as a part of a design and building process. As helpers to achieve and realize a built building for real life use.

I think video guys can use multiple GPUs easily for color correction and video editing and maybe other tasks related to video producing industry, like encoding or decoding (DaVinci Resolve Studio for example can use multiple GPUs).

Scientific simulations and calculations, prognostications. Particle and fluid calculation accelerations and other swarmlike behavior modelling with real time animations. Those are fascinating areas, and I would like to dwell in there too if I only had any time left for it.

Maybe a huge video wall installation with 16 combined monitors in it would need multiple GPUs. But you wouldn't need VIIs for that, there are better low power alternatives with more ports to drive multiple screens.

And of course, if I dare to say, there is this mining industry out there too. About that I do know hardly anything. Apart that I could probably double my investment reselling the three VIIs right now.

If I installed OC+Big Sur I could try Octane for Mac. It should be accelerated almost linearly by the number of GPUs I think. I haven't decided yet about that.
 

mikas

macrumors 6502a
Sep 14, 2017
898
648
Finland
One more thing, a pleasant surprise; Affinity Suite.
At least Affinity Photo does take advantage of all three VIIs (Metal based photo editing app).
About every effect is real time and instantenous, I can drag the sliders and most on them respond in realtime with no lag whatsoever. I use Affinity Designer too a little, and that should be multi-GPU accelerated too. Probably their Publisher app is also, why not.

Allright!
See screenshot. Dragging the effect slider and all 3xVIIs activate in GPU history. Did I say no lags - not with any effect I tried. Multiple ones applied and clicked on and off - still fast. Image: 5120x2880 14,75MP, RGBA/16, with approx. tens of layers and groups I think.
1617894378436.png
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.