Pixlas 4.1 Mac Pro mod

[two-mac-jack]

Congratulations you made it!

Suggestions:

Since you need two GPUs, and I see you made a lot of modifications, you may want to solder a pair of 12V + GND from the PSU. Trust me, it's a lot easier, saving your cables and safer. I have been using this way to drive two 1080 ti (plus an extra 12V power in the optical drive space for 3rd). I haven't encounter any shot off. Strictly speaking, in fact, electronically taping is more or less for signal not for power source as much loading as two Titan X GPU (12V 30A) would draw, and I think two 280X can be just the same.

Another suggestion is that you don't have to dangling your PCIE riser like that for your PCIE drive and USB 3.0. Put your 2 gpu to slot 1, and slot 4. That's not only better looking but also be more stable. Puting GPU to slot 4 is not hard at all. Change your spinning HDDs to SSDs and take SSD's metal case off then you have all you want. Any serious bulk files save in a NAS or USB 3.0 drive. My two cents...

Thanks for the suggestions!

I'm comfortable with soldering, like working on high-voltage tube amps. But the attach points I've seen on the PSU looked small and unprotected, and the threading of the wires out of the PSU cage seemed tricky, so I opted for the tapping approach.

I moved the 2nd GPU to slot 4, and it's a much better solution -- thank you! All the PCIe cards mount securely now and are held in place by the standard two-screw bracket/bar; no more hanging cards in space or propping them up with cardboard shims. And the Amfeltec is back in x16 slot 2 for best perf, and in case I want to do RAID striping.

At first I couldn't get the double-wide GPU card to seat properly in slot 4, then I discovered there is a metal tab sticking up where slot 5 would go:

I used pliers to bend that tab up and fold it down to the left side (part of it actually came off), and that fixed it so the card would sit flat:

A full-length double-wide card in slot 4 partially obstructs drive bays 2, 3, and 4, so I moved one high-capacity HDD to bay 1, and the other to the optical bay. I moved my Windows 10 SSD from the optical bay to drive bay 2. I was able to fit the SSD in a sled adapter above the 2nd GPU in drive bay 2 -- although it appears to be pushing the GPU card down slightly. If I let the SSD hang by itself, unattached to the sled, it leaves a little more room. Side view:

Looking up into the drive bays:

Rear view, showing that all 2nd GPU ports are accessible:

 

[pierrox]

I've been keeping an eye on this thread for a while. So when I decided to take the plunge and upgrade my GPU, I also ordered the cables and parts to replicate it.

I went for a GTX 980 Ti since it's tested and proved, and the benchmarks are still pretty good compared to a 1080. Also, I have some legacy apps only working in 10.10.x, so I wanted to have something compatible.

I got some side crumplers: they go on the wire, then you plug the power to them. Two advantages in my mind: since the power extension has smaller wires, you can mix and match to get it all right. Second, those are flat enough and as you connect the extension after the crumplers are in place, I could remove the PSU and work on a bench instead of inside the case.

The OCD in me went as far as pairing the +12v and gnd...

I made a mistake by having the connectors too close to each other. I should have left more space between them, which would have allowed to bend the original wires a bit to make the whole thing flatter.
But hey, once the back plate is in place, it's just invisible!

And with the card in:

Pretty happy with this mod. Ok, not as thorough as a proper soldering job, but pretty close. Took about an hour, and that's because I spent time checking each crumpler as I installed it, then rebooted several times to make sure I had 12v where it was supposed to happen.

 

[tuchubby]

Just finished this with all the info provided in this thread (big thanks to theitsage from the RX 480 thread). I believe moddiy created their own product specifically for this because it seems to pair up perfectly in terms of cable length. I took the splice route because I don't have a soldering iron that would heat up those points on the PSU. With a few suggestions in this thread, I also used t-tap splice connections. Because the gauges are different, please keep in mind to use correct connector sizes. I purchased a t-tap "kit" for a little over $7 on amazon that came with 10 sets (the splice and the quick disconnect) in 3 sizes:

- Cable nominal cross section: 0.5 to 1.5 mm² (Red)
- Cable nominal cross section: 1.5 to 2.5 mm² (Blue)
- Cable nominal cross-section: 2.5 to 4mm² (Yellow)

A simple google search told me that the yellow set will fit on the mac pro's PSU wires, and the blue set will fit on the pci-e cable that is getting tapped in. Shown below, I fitted yellow t-taps using locking pliers - you really should be using locking pliers since the gauge and insulation is pretty thick to clip on these t-tap connectors. To easily identify 12V vs GND for the quick disconnects, I simply tapped 3 12V to the left and 5 GND to the right (I could have spaced it out better). PSU removal is highly recommended. This is what it looks like when all connected, don't mind the dust and dog hair:

And this is what it looks like finished:

 

[pierrox]

I have an issue with my 980Ti. It's connected with the Pixlas mod as stated above, and when I start a render in a gpu intensive app (Compressor or Davinci), it does shut down with light still on on the front of the Mac. After a moment it reboots, reopens the apps and I can hit render again and this time it goes on for hours no problem. It's weird as it does that only the first time I use the gpu crunching app after a morning boot.
A surge peak from the card? But why only once and then the machine works no problem?

 

[h9826790]

I have an issue with my 980Ti. It's connected with the Pixlas mod as stated above, and when I start a render in a gpu intensive app (Compressor or Davinci), it does shut down with light still on on the front of the Mac. After a moment it reboots, reopens the apps and I can hit render again and this time it goes on for hours no problem. It's weird as it does that only the first time I use the gpu crunching app after a morning boot.
A surge peak from the card? But why only once and then the machine works no problem?

If only happen once. It's hard to tell if you hit a software bug, or really Pixlas mod related.

 

[Theophany]

So I just went to start on this mod, got to the part where I was ready to put the T-taps onto the main PSU line and holy heck the taps were impossible to close by hand. Did you guys have to use pliers to get the taps to snap shut?

 

[deppest]

So I just went to start on this mod, got to the part where I was ready to put the T-taps onto the main PSU line and holy heck the taps were impossible to close by hand. Did you guys have to use pliers to get the taps to snap shut?

Yes, you will need pliers

 

[Theophany]

Cool, just wanted to check before going into this.

Am I right in thinking that two 8-pins powered by the main PSU 12v cable will be enough to sustain a 980ti at full tilt? Or should I power one 8-pin from the aux PCI power?

 

[Theophany]

Also, is it worth removing the PSU to make it easier to splice the cables? There's not a lot of room to work with when trying to do the mod in situ. Seems like it would be easier to clamp the t-taps outside of the chassis, then simply slide on the new cable once it's back in the chassis.

 

[pierrox]

Also, is it worth removing the PSU to make it easier to splice the cables? There's not a lot of room to work with when trying to do the mod in situ. Seems like it would be easier to clamp the t-taps outside of the chassis, then simply slide on the new cable once it's back in the chassis.

Yes I did that, and it was much easier to apply the required force with the pliers. You'll have to do a test to make sure the crimps can still go behind the partition between the PSU and the optical drives bay as it's fixed to the chassis.

 

[Theophany]

Excellent, thanks. I'll bear the partition in mind!

 

[rueyloon]

I have an issue with my 980Ti. It's connected with the Pixlas mod as stated above, and when I start a render in a gpu intensive app (Compressor or Davinci), it does shut down with light still on on the front of the Mac. After a moment it reboots, reopens the apps and I can hit render again and this time it goes on for hours no problem. It's weird as it does that only the first time I use the gpu crunching app after a morning boot.
A surge peak from the card? But why only once and then the machine works no problem?

I didn't do this mod, but I run my power supply externally. When I bought the first underpowered one, this is exactly the same problem I got. It went away after I upgraded to a more powerful power supply. This may suggest that you are pulling too much power and the surge protection was activated.

 

[MisterAndrew]

I did the mod on my 5,1 to run the Vega 56. It works great so far! I took the PSU out and I took it apart because I was planning on soldering the wires to the back of the board, but I couldn't get the solder to melt like Macinsquatch even at 399 degrees celsius, so I soldered to the wires. I used a new 8-pin connector with pins from ModDIY and I used 1.5 mm² wire, which is about twice the diameter of 18 awg. I covered them with braided sleeving so it looks factory and matches the optical drive cables. Since I used only black wire, I temporarily taped red tape on the +12v wires and yellow on the sense wires (that I doubled up with 2 ground wires) so I could tell them apart after I fished it through. I crimped the pins onto the wires and put them in the connector after I had them fished through the hole in the back of the Mac where the optical cables go through. Then I connected that to a y-splitter adapter to make it 2 connectors for the card.

 

[hamishb]

Question for AndreeOnline,

Are you booting from the 1TB SSUBX? I can't find this info anywhere or if these chip are AHCI compatible or not?

Thanks!
-Chris

What card did you use for the SSUBX? Is it in a sled in the drive bay, or on a PCIe card?

 

[MisterAndrew]

What card did you use for the SSUBX? Is it in a sled in the drive bay, or on a PCIe card?

Probably like this one. It's a PCIe adapter. This is what I'm using. Works great. http://r.ebay.com/9vNTuj

 

[AndreeOnline]

What card did you use for the SSUBX? Is it in a sled in the drive bay, or on a PCIe card?

Like MisterAndrew writes above, it's that cheap little adapter. It only serves the purpose of remapping the pins from the SSD to the PCIe slot. It was $12 when I bought it though. I feel ripped off now that it's down to $4

 

[m3rob]

Hi AndreeOnline,

I am curious what graphics card(s) you are running at the moment. Did you find a good solution for your workflow? I am running FCP X and Resolve and would love to upgrade from my GTX 680.

Thanks for updating us with your findings and great modifications on your rig!

Rob

 

[h9826790]

Hi AndreeOnline,

I am curious what graphics card(s) you are running at the moment. Did you find a good solution for your workflow? I am running FCP X and Resolve and would love to upgrade from my GTX 680.

Thanks for updating us with your findings and great modifications on your rig!

Rob

For FCPX, HD7950 is a much better choice than GTX680.

 

[AndreeOnline]

I am curious what graphics card(s) you are running at the moment. Did you find a good solution for your workflow? I am running FCP X and Resolve and would love to upgrade from my GTX 680.

Hi, there is no one solution that is best for both FCPX and Resolve, since Resolve is faster with a strong CUDA card and FCPX is better with OpenCL and Nvidia cards in OpenCL mode in FCPX don’t deliver the same bang for buck as they do in other apps/modes.

It’s not that OpenCL is broken in Resolve but CUDA is faster. I had a 1080Ti but returned it because of some general instability and some benchmarks that didn’t run properly. I deemed it not quite ripe for a workstation and went to my fallback card wich is a RX480, while waiting to check Vega. Now that Vega is out I first thought I might stay OpenCL and go with a Vega 64 internally. I would get good performance in both FCPX and Resolve. But I would leave quite a bit of functionality on the table in terms of some 3D work I do (Maxwell and Cycles Render + TurbulenceFD) — they all can use CUDA along with CPU based rendering.

I didn’t have instability in FCPX and Resolve with my 1080 Ti, so I might go back to that accepting that some synthetic stuff and maybe some games (I’m not a gamer) won’t run quite right.

I thought for a long time that I would use a dual price/performance card setup, but I wasn’t ready to give up USB3 or PCIe SSD. And now those gfx cards aren’t that cheap anymore due to mining. The way it looks now I think I’ll return to the 1080 Ti.

With my RX480 I get all the performance I need in FCPX and Resolve is OK for basic stuff in 2k, including a few nodes. A stronger card would allow for realtime with more involved grades and for 50/60fps. In my experience FCPX is a pretty forgiving app that gives you good performance even on middle of the road gfx cards (AMD).

But the real pro tip is to check what you really need and not to over extend spending too much money on a rig more meant for the user you ‘want to be’ (and that’s not meant specifically for you, but more a reminder for us all).

 

[m3rob]

Hi, there is no one solution that is best for both FCPX and Resolve, since Resolve is faster with a strong CUDA card and FCPX is better with OpenCL and Nvidia cards in OpenCL mode in FCPX don’t deliver the same bang for buck as they do in other apps/modes.

It’s not that OpenCL is broken in Resolve but CUDA is faster. I had a 1080Ti but returned it because of some general instability and some benchmarks that didn’t run properly. I deemed it not quite ripe for a workstation and went to my fallback card wich is a RX480, while waiting to check Vega. Now that Vega is out I first thought I might stay OpenCL and go with a Vega 64 internally. I would get good performance in both FCPX and Resolve. But I would leave quite a bit of functionality on the table in terms of some 3D work I do (Maxwell and Cycles Render + TurbulenceFD) — they all can use CUDA along with CPU based rendering.

I didn’t have instability in FCPX and Resolve with my 1080 Ti, so I might go back to that accepting that some synthetic stuff and maybe some games (I’m not a gamer) won’t run quite right.

I thought for a long time that I would use a dual price/performance card setup, but I wasn’t ready to give up USB3 or PCIe SSD. And now those gfx cards aren’t that cheap anymore due to mining. The way it looks now I think I’ll return to the 1080 Ti.

With my RX480 I get all the performance I need in FCPX and Resolve is OK for basic stuff in 2k, including a few nodes. A stronger card would allow for realtime with more involved grades and for 50/60fps. In my experience FCPX is a pretty forgiving app that gives you good performance even on middle of the road gfx cards (AMD).

But the real pro tip is to check what you really need and not to over extend spending too much money on a rig more meant for the user you ‘want to be’ (and that’s not meant specifically for you, but more a reminder for us all).

AndreeOnline! Awesome! Thanks for the update! You are right about the pro tip on how much money to sink into this rig. I am still getting by with my gtx680 Mac edition. I render a lot of raw cinema dngs from my FS700 and Odyssey 7Q, but I can just let it render overnight. I may check out the RX580 Pulse that everyone talks about if they become available new again. Thanks again!

 

[h9826790]

L

AndreeOnline! Awesome! Thanks for the update! You are right about the pro tip on how much money to sink into this rig. I am still getting by with my gtx680 Mac edition. I render a lot of raw cinema dngs from my FS700 and Odyssey 7Q, but I can just let it render overnight. I may check out the RX580 Pulse that everyone talks about if they become available new again. Thanks again!

The PULSE was supposed at the same price level as the other RX580, but now clearly not the case. I am now more tends to get the Nitro+, it will work anyway, and I have 2 ROM to play around. I can even flash the PULSE ROM onto it (or mod it to cheat the MacOS believe it’s the PULSE). At the end, the PULSE only provide correct identification in MacOS at this moment, no other actual benefit. It’s still the safest RX580 to use, but just no real world benefit.

 

[VanneDC]

Nice work Andree, I keep coming back to your thread so see what your getting up to next.

very cool.

 

[Harry322]

Hi there - I am planning an upgrade to my 5,1 Mac Pros at the end of the year in anticipation for new GPUs (Titan Xp?). This mod looks like a great way to power said GPUs, but I lack the confidence to solder / re-wire myself.

This may be the wrong place to ask, but is there a Los Angeles based MacRumors user who I could hire to perform the Pixlas mod on my systems? PM me if possible.

 

[@WillCoffin]

Hi, there is no one solution that is best for both FCPX and Resolve, since Resolve is faster with a strong CUDA card and FCPX is better with OpenCL and Nvidia cards in OpenCL mode in FCPX don’t deliver the same bang for buck as they do in other apps/modes.

It’s not that OpenCL is broken in Resolve but CUDA is faster. I had a 1080Ti but returned it because of some general instability and some benchmarks that didn’t run properly. I deemed it not quite ripe for a workstation and went to my fallback card wich is a RX480, while waiting to check Vega. Now that Vega is out I first thought I might stay OpenCL and go with a Vega 64 internally. I would get good performance in both FCPX and Resolve. But I would leave quite a bit of functionality on the table in terms of some 3D work I do (Maxwell and Cycles Render + TurbulenceFD) — they all can use CUDA along with CPU based rendering.

I didn’t have instability in FCPX and Resolve with my 1080 Ti, so I might go back to that accepting that some synthetic stuff and maybe some games (I’m not a gamer) won’t run quite right.

I thought for a long time that I would use a dual price/performance card setup, but I wasn’t ready to give up USB3 or PCIe SSD. And now those gfx cards aren’t that cheap anymore due to mining. The way it looks now I think I’ll return to the 1080 Ti.

With my RX480 I get all the performance I need in FCPX and Resolve is OK for basic stuff in 2k, including a few nodes. A stronger card would allow for realtime with more involved grades and for 50/60fps. In my experience FCPX is a pretty forgiving app that gives you good performance even on middle of the road gfx cards (AMD).

But the real pro tip is to check what you really need and not to over extend spending too much money on a rig more meant for the user you ‘want to be’ (and that’s not meant specifically for you, but more a reminder for us all).

Thanks for sharing so much of your journey on this mod. I am looking to build a 5,1 Editing/CAD/VR rig dual booting Windows/MacOS. Main uses are: FCPX/Motion, Premiere/AfterEffects/Photoshop, Resolve (MacOS) | Solidworks and SteamVR (Windows 10). My current plan is going with a 1080Ti. Have you (or anyone else on here) had issues with powering that card from the OEM power supply under heavy load? Other places suggest adding an external PSU or a SATA->PCIE cable for the card to prevent shutdowns but the 5.25" bay supplemental PSU's don't seem to be in production anymore. Soldering doesn't scare me, but I don't want to mod the power supply just to find out that it won't provide enough juice.

This is the article that got me scared:
http://www.macvidcards.com/blog/the-pesky-power-issue-with-pascal-1080ti-and-titan

The SATA->PCIE power cable seems like more of a hack than a stable solution.

Any help is greatly appreciated!!

 

[h9826790]

Thanks for sharing so much of your journey on this mod. I am looking to build a 5,1 Editing/CAD/VR rig dual booting Windows/MacOS. Main uses are: FCPX/Motion, Premiere/AfterEffects/Photoshop, Resolve (MacOS) | Solidworks and SteamVR (Windows 10). My current plan is going with a 1080Ti. Have you (or anyone else on here) had issues with powering that card from the OEM power supply under heavy load? Other places suggest adding an external PSU or a SATA->PCIE cable for the card to prevent shutdowns but the 5.25" bay supplemental PSU's don't seem to be in production anymore. Soldering doesn't scare me, but I don't want to mod the power supply just to find out that it won't provide enough juice.

This is the article that got me scared:
http://www.macvidcards.com/blog/the-pesky-power-issue-with-pascal-1080ti-and-titan

The SATA->PCIE power cable seems like more of a hack than a stable solution.

Any help is greatly appreciated!!

Pretty much depends on which 1080Ti you get. For the heavy factory OC model (e.g. MSI 1080Ti Lightning Z), sure you can't power them by just the mini 6pins. But if you get the reference card, then it may be OK.

https://forums.macrumors.com/threads/getting-my-5-1-ready-for-vive-vr.2046670/page-3#post-25429070

I can confirm that CUDA-Z heavy test can shut down the cMP (even with reference card). However, apart from CUDA-Z heavy test, I never experience any issue (as you can see from my link, including Furmark).