Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Thanks guys for participating to this discussion in a hijacked thread by me (sorry MacScout). I am having second thoughts about taking this build apart too soon. Maybe there is something to test still out there.

OT:
Best part is the price of that aforementioned software. I am not affiliated anyway in their business. Optional and not in any means a paid link: Affinity suite apps 50% off
 
  • Like
Reactions: Stex
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.
View attachment 1755282
This is a very good news to cMP. It makes the cMP a great photo editing machine again.
 
One more another thing. This 3xVII thing is fast. Previously I have accidentally started C4D R19 a couple of times when testing ProRender, because it was in my Dock as an icon without any indication of release version. Got that now.

C4D R21 w ProRender:
Trying again with C4D R21, and I see much better AMD ProRender stability. And that thing is quite fast. I set the iteration for preview to 500 (a lot) and for final render with better overall settings at 700 (not so much). Please see screenshot, and I attached a screen recording too so you can see and comment the redraw speed of the engine with my rig. And maybe the level of finesse of that final income (PNG link). This engine should be in the category of unbiased, I think. Or at least I have read so.

As an architect I would first think about the desire for reality, because the final realized building of our designs can't be "biased", can they. Not easily I say, not without a filter or with a scrub line or anything like that. It would be built in and live in the world we all know and share. That equals more to unbiased renderings to me as architect. At least it does to me. I do love fantasy and arts and everything mind boggling there is in imagination, but that's not the reality for my work - unfortunately maybe.

C4D w RedShift:
Maxon bought or merged with RedShift. ProRender seems like a stepchild now.

Redshift would be a biased renderer. It looks great though. For Redshift tryout I would need to install Big Sur on this ancient Mac Pro of mine; feasible with OC maybe, I have thought about that a couple of times now. But Mojave is the preferred OS for my main software as of today still. I know, it's odd. But that's the way it is because of Apple, and their practices as of lately, the recent years at least.

OTOY Octane:
It's kind of a shame with Octane for Mac. It should be the greatest, the best, and everything you would ever need for a GPU renderer. It's either always almost there, or it is just unstable enough to prohibit actual work. Nastly said, sorry, I have no concrete experience with this engine yet.

For me ProRender would be a match made in Heaven. If only my software would support it, ArchiCAD. One of my chosen softwares do support it, Rhinoceros does. Unfortunately Rhino is just the supporting act to me, though. Still, that's probably the subject of my next study. Could I somehow use Rhino with ProRender for my rendering results in parallel with ArchiCAD, my main program. Or, maybe I could convert to Rhino, and ArchiCAD would become my supporting program - I don't think so, just a quick thought.

There are a lot of direct linking going on nowadays software wise. I'm gonna look at it, but I don't have my hopes high with that path.

I'm sorry, back to the subject. Preview (left, 920x688, 500 iterations, about 15 seconds). Final render (right, 700 iterarions, 1200x1600, exactly 4 min and 6 secs)

Please see attached;
C4D ProRender (Scene by Yan Ge) screenshot +

.MOV video link (DropBox).

ProRender_tests_101.jpg
 
Last edited:
ProRender multi GPU benefits don't seem to be quite worth it. It's not near linear improvement, more like logarithmically decreasing benefits per added GPU.

1618557305444.png

1xVII: 6:57 =417 sec
2xVII: 5:05 =305 sec
3xVII: 4:22 =262 sec

I save 112 seconds =27% of the original time with 2xVII over 1xVII.
I save 155 seconds =37% of the original time with 3xVII over 1xVII.

I don't know what would be the optimal setup, things might change if I reorder a little bit. My main GPU, the one with a monitor attached to it, is now the one in 4x PCIe 2.0 slot. I believe this is the default one if I choose to use one GPU only. I deduced that in an analog way, with my fore finger. The VII was much warmer to touch than the other too while rendering.

If I changed that to x16 slot, I might see a little bit different results. Maybe I will try. Seems like 1xGPU is much much more efficient than multiple ones (per watt, per used time). End result is still the same of course. Maybe the other GPU renderers out there are better at this.

But overall I like this setup, though one GPU would be enough it seems. I can preview effects almost real time, like DOF (depth of field), DOF on/off effect takes something like 20-25 seconds to preview. In final render there seems to be no difference in final render time.
1618557785872.png

This has spread out maybe too far from original subject, sorry about that. Still, Pixlas mod is behind all of this, making this possible for an ancient 12 yo Mac Pro.
 
This has been educating and fun project for me.
It has been educating for many of us as well! Thanks for such detailed documentation and for sharing the process!

Your endeavor made me wonder if I would get benefits for multi-GPU accelerated software if I were to install both my RX580 (non-flashed) and AMD 7970 (MVC flashed)...?

Or do the GPUs need to have the same model to fully benefit from such multi-GPU setup?
 
You are a GD HERO!!! You have won the "Ultimate GPU Config" trophy and conclusively reached the ceiling of GPU performance in the cMP. Glorious to behold.

When did Apple permit (re-authorise) multiple-GPU configs greater than 2..? This capability was removed from OS X between 10.8.5 and 10.9.2

Can you really use 3 or more GPUs again? If so - which macOS release brought it back.?
 
Thank You.
I guess one might still be able to use a PCIe extender box to get more GPU power with a cMP. I am not the one willing to go through that road though.
😅

I am on Mojave still, because of my softwares working the best with it.
I can't remember about changes in Mac OS support for multiple GPUs. I recall that they broke different brands at the same time in a Mac at High Sierra. I think you can still run different cards from AMD simultaneously, but did not try that this time. I might try that soon, I'm gonna repurpose the 3 VIIs a little bit differently I think. My main work computer, a dual X5680 needs a GPU or two to replace a Vega56 in it.

BTW, I am in the middle of thinking performing another Dual Pixlas mod to my spare PSU. Why not make it a dual now that it's out and easy to work with.

I've got a Rhinoceros 3D license, and have just read that there is a ProRender plug-in for Mac Rhino too. That was really great news for me, I'm certainly a fan of Rhino + Grasshooper tool. Now with Prorender it might up to another level for me. Can't wait to try it with some VIIs, I am so excited again.
 
I've got a CUBIX and a NETSTOR (Dynapower) PCIe expansion box. They're awesome.. but for me the modding challenge has always been about what can be accomplished within the confines of the cMP tower.

Trophies for hitting the CPU config ceiling and the RAM ceiling were handed out within 3 years of the 5,1 release BUT the PCIe architecture meant that a GPU ceiling and a disk transfer speed ceiling would not be tested until 2021 - 11 years after it was first introduced. A true testament to the innovative, 'spare no cost' approach of Apple Engineering and Design team from this time period.

Confirmation that you are using (testing) this 3x Radeon VII config using Mojave (the last officially supported macOS release for 5,1) and not Open Core makes this the cMP ultimate GPU trophy winner in my mind.

Here is a link [ AMAZING NEWS - 2.5 Width GPUs will never again interfere with SLOT 2 ] to the post I made when I found a clean approach to GPU installation in Slot 4 - not that I'd suggest making any modifications to a GPU that is now fetching close to USD 1500 in the second-hand market(!)

I know you contemplated a Unigine 4.0 test - did you ever give it a go?
 
Did not try Unigine. I don't think it's multi-GPU capable test, or is it?

That PCIe bracket change in your link is a clever trick indeed. Thought about that too, but with VII design you would need to disassemble the whole card before you could reach the screws necessary for bracket change. So I decided to gently bend it instead. I can bend it easily back to original form, only the black paint is slightly cracked, so visually a little bit annoying, but that's it.

I think I'm gonna put the VIIs in my main dual proc Mac Pro now that I have the power play kexts working, or then again, maybe only two of them. That might be enough speed for ProRender use for me. I will test it either way of course and then decide. I will loose the NVMe in exchange if I decide to go with 3x GPU, but it's not that much a big loose for my kind of daily work.

Time will tell what Mx series will offer for me as a possible new Mac Pro variant. Until then I will go with this ancient rig I think. I just hope it lasts that long with me.
 
Last edited:
not that I'd suggest making any modifications to a GPU that is now fetching close to USD 1500 in the second-hand market(!)
Not necessarily : there are a couple of companies that sell 1-slot brackets e.g. Watercool, EKWB, Aquacomputer.
 
Last edited:
It's not so sure there is any Trophy to behold for long. 6800X, 6900X AMD's have been out for some time now, MacOS 11.4 is being tested and there seems to be drivers for those cards. There should be x86/x64 drivers available for 5,1 too maybe if everything goes ok and our clever guys and testers somehow can iron it out.

I do think 2x 6800X or 2x 6900X would beat 3xVII in a cMP. Easily. That's what 2x 3090 does. And if you could put 3x of those in to a cMP.. well that would be a treat.

Anyway, I'll split this (maybe a little bit questionably too soon announced maybe, but thanks anyway. I did feel good for a moment after working so long with this).

Trophy to pieces, to all of you. The hardware guys. The software guys. The Pixla himself! Who ever that is actually (AndreeOnline in MR forums was at least one of the very first I guess).

So here, take your own all of You. (I just happened to have the three VIIs available at the moment).

🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆

We'll see if anything comes out of those newer AMDs in a cMP. It's possible, but the incentive to do that might have gone allready for all of us cMP enthusiasts.
 
Last edited:
  • Like
Reactions: KeesMacPro
Trophy to pieces, to all of you. The hardware guys. The software guys. The Pixla himself!


the Pixlas Mod is (half) dead, long live the Mikas Mod!


:D

PS: Jokes aside, I will indeed start calling the double Pixla the "Mikas Mod" from now on!
 
Thank You so very much. And again thanks for your participation all of you.

I am pretty sure this all has been done allready somewhere, way before my own tricks. And do not forget it's a community work. I would not have achieved this without the help of more than a handful of forum member supporters and not without their prior work. They are the true heroes. I won't list those guys here. You all know who they are.

However, I do have a plan to do a Trixlas, just maybe some day I'll do it.
;)
 
  • Like
Reactions: KeesMacPro
It's not so sure there is any Trophy to behold for long. 6800X, 6900X AMD's have been out for some time now, MacOS 11.4 is being tested and there seems to be drivers for those cards. There should be x86/x64 drivers available for 5,1 too maybe if everything goes ok and our clever guys and testers somehow can iron it out.

I do think 2x 6800X or 2x 6900X would beat 3xVII in a cMP. Easily. That's what 2x 3090 does. And if you could put 3x of those in to a cMP.. well that would be a treat.

Anyway, I'll split this (maybe a little bit questionably too soon announced maybe, but thanks anyway. I did feel good for a moment after working so long with this).

Trophy to pieces, to all of you. The hardware guys. The software guys. The Pixla himself! Who ever that is actually (AndreeOnline in MR forums was at least one of the very first I guess).

So here, take your own all of You. (I just happened to have the three VIIs available at the moment).

🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆🏆

We'll see if anything comes out of those newer AMDs in a cMP. It's possible, but the incentive to do that might have gone allready for all of us cMP enthusiasts.

for 6900 take care for a good return policy. I put a reference one in a 5,1 and the machine refused to even post.

 
  • Like
Reactions: mikas
Thanks for the link.
Maybe it's worth to try later again if/when 11.4 with 6000 series drivers is up and running with cMP?
For now it all seems too flaky with 11.4/cMP as we all have read here in the Mac Pro forums.

At the moment for me there would be a chance to grab one.
Availability: Right Now
1620715398987.png

Did not complete this order though. Too salty for me.
 
So you think that an official released driver for 68xx/69xx from apple won't help it?
edit. a purely academic question. Not a fuel for my future testings, I hope so.
 
no, the Mac has not started, no chime. It's got stuck very early in the firmware phase.

Long before any OS or Driver has been loaded. I asume it's the Bios as it got a start loop when adding a GPU with UGA (Mac Bootscreen)

Just had this only GPU - as the box won't come up I had no chance to investigate further. My customer tested the GPU ok in a PC later.
 
So, hot morning, woke up too early as always. What to do?
Work? Nope, not this early, don't think.

Had a couple of cables, arrived too late for premiere, so for a spare, or to use it?

I did a Pixlas mod today at 5 am.
And upgraded My Aqua Kryo EVO with EVO 1TB to plain Aqua Kryo with Corsair 400 2TB.
And a Samsung EVO SSD 2TB to back it all up.

No more 3xVII for me. There was no use for it enough to justify it. I am now splitting it all to all my workstations to make good use of those all.

About Pixlas: The red ones do crimp the best at least for MODDIY cables.
1623435548023.png
 
Almost 2,5 years later I had to revisit this. Cinebench 2024 was released, and it is based on RedShift GPU render engine of Maxon (a Nemetchek company). My daily software (ArchiCAD) includes RedShift too, though not in it's latest incarnation, but still, anyway. It's a promise to be fullfilled eventually.

So I had to take on Open Core and my trusty cMP once more still. After some couple of nights reading, studying and trying, I figured out the open core way. And I figured out how to edit plists. And injected DeviceProperties. And found out PCIe paths. And what not.

Way out of my league software wise, but here I am. My humble GPU cruncher is back.
3xVII_monster.jpg

3xVII gave me eventually, after all the hooplas and huuplas, a decent score for Apple platform (didn't count any nVidia scores for now). Radeon VII is spoofed as Vega II by Open Core in here.
1696180175487.png

And the same old LuxBall stands still, downvolted and Monterey, stil over 150k.
1696180242087.png

You got to give it to an Apple cMP. It has lasted for a long time, just hanging in there tightly. With a little help of course. Community help. Gotta love you guys who still keep digging, and then some more, finding all the possibilities there possibly are to be found with this old cMP HW and SW.
🥳
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.