Mac Pro Pixlas Mod, is it really needed?

Mac Pro Pixlas Mod, is it really needed?

Reading Time: 11 minutes

A few things have bothered me about the reasons for recommending the Pixlas Mod.
1. The mini 6-pin connector can only provide up to 75W of power.
2. The traces in the Mac Pro’s logic board are too thin to have more power demanded than 75W.

I needed definitive proof (at least definitive for me) of these two claims so I decided to really dive into this. I’ve gathered all the relevant information and put the puzzle pieces together, along with anecdotal evidence from Mac Pro users in MacProUpgrade. Here are my findings…

Claim 1: The mini 6-pin connector can only provide up to 75W of power
Finding a spec sheet for the mini 6-pin (or ‘Molex Mini-Fit Jr. connector housings’) has proven not possible for me. But just like the mini-USB or mini-DVI, the mini 6-pin PCIe connector is just a smaller form factor of it’s full sized counterpart. This counterpart does have specs available online. Presenting the “Molex 45559-0002” connector. This is what you’ll find on the cable connecting your GPU to one of the mini 6-pin auxiliary power connectors on the board. The datasheet for that connector can be found on the linked page. This connector is rated at 9 Amps per contact (pin). The standard for this connector is that only two 12V pins are connected and a third is optional. In the Mac Pro 4,1/5,1, this third pin is connected and provides 12V. This leaves us with the following conclusion:
• 3 * 12V available
• 3 * 9A available
• 12V * 9A = 108W
• 108W * 3 pins = 324W total available power for a mini 6-pin connector.
This is simply going off of the datasheet. Mind you, this connector and others like it (Molex 39-01-2060) are rated for 600V at 9A so 12V at 9A should be very much possible without any issues for these connectors (using the appropriate gauge wiring of course). To my knowledge, no-one has ever squeezed more than 210W out of a mini 6-pin connector. Draw more than 210W and the Mac will shut down. This has been proven by a few people now that have documented their tests.

That’s quite a bit short of the 324W the connector should be able to do, 114W short to be exact. So what causes the Mac to shut down at 210W on that port? Not sure but it can definitely handle more than 75W.

Applications such as Hardware Monitor and iStat Menus will show a max power draw of 7.99A on a PCIe auxiliary port. Some say this is a display limit, others say this is the actual limit meaning Apple’s board design caps the current at a max of 8A. The max rating for the port is 9A so it doesn’t matter much but if we take the 7.99A then the total available power would be 288W.

Claim 2: The traces in the Mac Pro’s logic board are too thin to have more power demanded than 75W
I took a Mac Pro 4,1 apart and lifted out the board. On both sides of the board thick 15mm traces lead to and from the mini 6-pin connectors. I figured one was ground and one was 12V, this is not the case. The traces that are visible on both sides of the board are both ground planes. The PCIe auxiliary ports get their 12V from traces inside the board. We don’t know how big the traces for the 12V line to the PCIe auxiliary connectors are!

With this information available I’d say both claims, which is what the Pixlas Mod is largely based on, are debunked.

A lot factors in to how much power a port can provide. Power supply rating, 12V line rating, traces in the board and wire gauge just to name a few. The rabbit hole goes deep and there is still some room for speculation as Apple does not make certain information available to the public. Below are some of the bits of information that lead to my current recommendation on the Pixlas Mod (which I will get to as well). If you have no interest in reading my rambles then just stop here and skip to the bottom of the page 😉

Power Supply Rating

Starting at the top of the chain, the power supply. The Mac Pro 4,1/5,1 is powered by a Delta Electronics DPS-980BB, rated for 980W maximum continuous total output. Not bad! Power supplies receive their rating at a set temperature. We don’t know what temperature the Mac Pro PSU was rated at. It could be a very unrealistic “980W at 25C” or a very good “980W at 80C”. For a PSU to maintain it’s maximum rating at 80C or higher, you’re typically talking about véry expensive power supplies (think military use). These delta PSU’s are pretty good but not thát good.

Power supplies typically come with an operating temperature range, not the Delta PSU’s in Mac Pro though. They mention nothing about temperature. If we look at the Mac Pro tech specs page we see “Operating temperature: 50° to 95° F (10° to 35° C)”
Typically the operating temperature means that anything over the high range, in this case 35C you lose output very quickly.
35C = 980W
45C = 75% power or 735W
55C = 50% power or 490W
This is called the “de-rating curve” and de-rating typically starts around 50C so assuming (hoping) this is the case with the Delta PSU as well we’d have 980W at 50C, 735W at 60C (75% output), 490W at 70C (50% output) and at 80C the power supply is operating at 25% output (245W) capacity so you’ll probably trigger thermal shutdowns under even a light load.

Greg Thompson helped me test to see what Apple’s PSU operating temperature was at least. He put a maximum load on his Mac Pro with Prime95, Heaven and other utilities and pushed his power supply to 561W at 60C. This shows that at 60C the PSU is capable of providing at least 57% output. We’d need someone with dual X5690’s and a much beefier GPU to continue this test and see how much wattage can be drawn at which temperatures to narrow this scale down. At any rate, this shows the de-rating curve for the Delta PSU does not start at 35C, so why Apple lists 35C as their maximum operating temperature is anyone’s guess at this point. It’s a safe assumption that the Mac Pro PSU follows the same average de-rating curve as most other PSU’s out there. As long as the PSU is kept under 50C, it’s maximum output is probably 950-980W. Over 50C you’ll probably lose 25% output for every 10C.

950W? Ah, good eye. Yes, in this case we’re dealing with 12V and according to the sticker on the Delta PSU, the 12V output has a max of 79A. This translates to 948W for the 12V output at most (12V*79A=948W). A minor technicality but still a relevant one. This means dealing with GPU power requirements (and in general) one should consider a 948W max, not a 980W one, as the GPU uses 12V on the mini 6-pin port. The 5V and 3.3V power lines share components with the 12V lines so the max output of 79A is ONLY if the 5V and 3.3V lines are not in use. This means your RAM, hard drives, other PCI cards, optical drives, your processors, USB devices etc. etc. Of course there wouldn’t be any Mac Pro left if you take that all away and while the 5V output has it’s own 5A rating (25W), shared component use inside the PSU lower these 12V and 5V ratings. the 5V and 3.3V power rails also create lower power rails of 1.5V for example that power other components which again take away from the overall rating.

Taking my Mac Pro 4,1/5,1:
It uses between 185 and 335W between normal and load. And this varies depending on the type of load etc. The GPU is not included in this and I use stock CPU’s, not the power hungry 3.33 or 3.46GHz ones.
This brings you down from 948W to ~800W (idle), ~760W (normal load) or ~600W (heavy load, again this depends on your setup etc.). And now you can start thinking about your GPU, it’s power draw and how you want to power it. The above mentioned temperature factors in to this too. This applies to a power supply that is operating at 100% output but if your PSU is running at 70C you’ll only have 50% output to work with. Proper fan control is essential in a Mac Pro as Apple’s default fan curves can easily have your internal components running at 80+C before the fans wake up.

Wire Gauge
A common misconception is that sudden shut downs are related to improper wire/trace gauge (see above Claim 2). This is not the case though. Sudden shutdowns are a protection from the power supply when too much power is pulled (or some current sense circuitry on the board). If the traces or wire gauge was the problem your wires or traces would simply melt.
We know the traces in the board can handle at least 8A. The American Wire Gauge standard mentions that 8A requires at least a 13 gauge (AWG) wire (assuming properly rated insulation) and since these wires are not melting, we can assume the traces in the board are not the problem. In fact, most of us use 18AWG wire either for the Pixlas Mod cable or for the mini 6-pin to GPU connection and 18AWG wire is only rated for a max of 2.3A. With three 12V connections in the pin that’s a total of 7A our cables are technically rated for. Something we often exceed. Also assuming the AWG rating is a tad conservative, we can definitely rule this out as a cause.

Real world tests
By now I have seen and heard enough people that say they use mini 6-pin to 8-pin conversion cables, draw way more than 75W and never experienced any shutdowns. On the other hand there are those that use mini 6-pin to 6-pin alongside mini 6-pin to 8-pin conversion cables and they do sometimes mention sudden shutdowns. The issue appears to be power draw balance. I haven’t found anyone yet that experiences shutdowns while using a single mini 6-pin to 8-pin conversion cable but those that use both 6 and 8 pin cables mixed statistically experience them far more.
These connectors offer more than just 12V and ground wires, there ae sense cables as well that tell a GPU whether a 6-pin ot 8-pin cable is connected. If the GPU ‘sees’ an 8-pin cable connected it can demand up to 150W from that port, not knowing the other end of that cable is a mini 6-pin. In itself that’s not a problem because we now know the mini 6-pin can provide more than that. Most evidence points to the Mac Pro board circuitry being the problem. The Mac Pro knows there are two mini 6-pin connectors present. Identical connectors should always have more or less the same power draw (this is not an issue if two GPU’s are detected from the testing I’ve seen). If one of the connectors suddenly has a draw spike of 150W while the other connector is sitting cush at 80W (random example), something MUST be wrong even if total power draw is well within the technical limitations! And there’s your shutdown.
Power balancing is handled by the GPU and it balances based on the info it gets from those sense lines. A PCIe slot can provide up to 75W but realistically I’ve never seen it take more than 40-45W. The rest has to come from the 6 or 8 pin ports.

120W the magic number?
I’ve seen it mentioned several times now that drawing more than 120W on a mini 6-pin connector can cause a shutdown. As the connector can provide at least 280W this doesn’t make sense. Earlier you read me saying that drawing more than 210W causes shutdowns so which is it? I have not been able to get a clear answer or any solid documented testing to answer that question. Unless the GPU draws 120W or more on a single pin, then this becomes a problem as the single pin can only provide at most 108W (assuming 9A limit) or 96W (assuming 8A limit). If this is the case I would see that as a flaw in how the GPU balances power, not just across cables but across pins for a single cable.

As you can see, details are still a little vague on this part of the equation but what most do agree on; it’s all about the right balance.

Load balancing
By using a mini 6-pin to 8-pin cable, you tell the GPU that 150W is readily available. If that one cable is all you’re using, no problem. If your second cable is a mini 6-pin to 6-pin cable, your GPU may think there is only 75W available there (assuming GPU manufacturers maintain an indurty limit/understanding that 75W is the max for a 6-pin port). How do you balance that out? Well, there’s the Pixlas Mod of course but there’s also a little gadged called the EVGA powerlink. The powerlink basically does away with the sense lines and only concerns itself with the 12V and ground. You can configure the powerlink with 6-pin, 8-pin or a mix of both connectors to fit on a wide range of GPU’s. Once it’s fitted, you are left with two 8-pin ports that you can connect with mini 6-pin to 8-pin conversion cables. Without sense wiring and with both ports on the GPU side being connected to the same 12V and ground rails, any power draw will end up with the same two 8-pin connectors and is automatically balanced out. Even if the GPU requests more power from it’s 8-pin port, the powerlink funnels it all through the same rail.
At $20, it’s a few bucks cheaper than the Pixlas Mod and of course a lot easier to set up and install. Of course your GPU has to be compatible, there are a dizzying amount of GPU models, brands, rebrands, variations etc out there and the powerlink doesn’t fit all of them.

So, goodbye Pixlas Mod?
Eh not so fast, doc. There is the small matter of us not knowing what the actual gauge of the traces in the Mac Pro board are. They appear to be enough to handle whatever we throw at it but we really don’t know. If by chance they are not good enough, pulling hundreds of watts through them will cause a significant amount of heat and significantly speeds up electromigration. Read into electromigration a little if you haven’t, it’s really cool 🙂 But it’s not so cool when it happens in your Mac Pro to the point it punches a hole in your traces (on a microscopic level).
As mentioned, the Pixlas Mod will also handle power balancing as it basically is the exact same thing as the powerlink, just less pretty.

I will continue to use the Pixlas Mod to power my GPU’s and prepare my Mac Pro’s for whatever card(s) I may have in the future. But you now know that you may not need to use the Pixlas Mod, certainly not as often as it’s been recommended in the past (by me as well). As there are still a few unknowns, if you want to take the safe route, go Pixlas. If you like to tinker and have your Mac Pro ready for any GPU, go Pixlas. If your GPU requires both a 6-pin and an 8-pin, go Pixlas or powerlink. But if you have:
• Single or dual 8-pin
• Single or dual 6-pin
Everything points to that you should be fine to connect these straight to the board. Unless of course you have a GPU that tries to draw more than 120W on one of the connectors, or pins… or was it 210W? See, some questions remain and until they are definitively answered, my Mac Pro’s will keep using the Pixlas Mod.

This was a shortened version of what went on in my head, I hope you found it interesting! A big thanks goes out to the users of MacProUpgrade that helped me by answering some questions and explaining some things and if you have any comments or feedback, please drop them in the comments below. This article will probably get updated and revised over time so if you call on it’s information, always check to see if something has changed since your last visit. I do not claim to have all the answers and I may be wrong about some things or a lot of things. I am always willing to learn, even if that means being proven wrong or being corrected 🙂

Cards that really should be using the Pixlas Mod:
Sapphire Radeon RX Vega64 8G HBM2
Even though spec states this card has a power consumption of 295W, this card has been documented to pull 570W (thank you Johnny McClung Photography for testing and documenting).
System shutdowns were triggered when using:
• 2x mini 6-pin to 8-pin directly into the GPU
• 2x mini 6-pin to 8-pin to an EVGA powerlink
No shutdowns were triggered when using:
• 1x dual mini 6-pin to single 8-pin + dual SATA to single 8-pin
As the SATA connector is officially rated at a max of 54W, using one or multiple SATA ports to power a GPU is risky. We don’t know how adequate the traces in the board are and pulling 250W or more from these connectors can risk your connectors, board and/or SATA controller. Pixlas Mod for this card is definitely recommended.

Do you have a GPU that causes system shutdowns? Let us know so we can add it to this list if needed.

Did you like this content?
Take a second to support me by donating!

21 thoughts on “Mac Pro Pixlas Mod, is it really needed?

  1. Im following your lead on my Mac Pro 2010 rebuild and grateful for it. I use Pp Cs6, Ae & would like to have ray trace capability as well.
    I have a NVIDIA Titan X and am now a bit uncertain as to installing it. If the Titan X is not suitable which NVIDIA would you recommend ?

    1. I am experiencing shut downs on upgraded Macpro 5,1.
      *dial X5690 12 cores
      *32 gig RAM
      *Radeon AMD RX 580
      using dual mini 6 pin to single 8 pin connecting directly to RX 580. should I be using another config?

  2. Hey there,
    thanks for your great post! I was running a GTX980TI under OSX on a Macpro 5,1 with dual X5690 without any Problem. Also a running Prime95 + Luxmark torture test did not caused the system to shutdown.

    Interestingly when i installed windows (natively on an SSD) and started some games (Outer World for example) the system randomly shut down. — independently of the temperature. Somtimes it took only 10s after starting games, sometimes 20 minutes.
    After installing Pixlas mod these random shutdowns disappeared completely.

    So there might be some limiting in OSX? No clue, but maybe interesting for you…

  3. RX 580’s are typically not known for overdrawing current, however I found my MSI RX580 Armor X 8GB to shutdown under certain load in Windows. Specifically playing modded Skyrim. I’m powering it with a dual mini 6 pin to 8 pin adapter.
    As a temporary solution I powered it with an external PSU and I haven’t experienced shutdown since. I will be doing the Pixlas mod tomorrow.

    1. My 4,1 running Mojave 10.14.3 is experiencing shutdowns.
      My son extensively plays Roblox (a non-intensive game), and the Mac Pro is shutting off by itself. It has a RX580 Sapphire Nitro+ 8gb.

      What would be a solution here? He gets mad because the Mac shuts off while he plays… I tried have him play the game with all fans running at the maximum, which keep the computer very cool, but it still shut down by itself.

      I got two mini 6-pin to 6-pin and 6-pin to 8-pin
      Maybe I should link the 12V rails from both 6-pin connectors to provide a better load balancing? What would be the best course of action here?

      1. I’d keep a close eye on the temperature sensors, make sure it’s not overheating. With a mix of 6 and 8 pin connectors if you don’t want to go the Pixlas Mod route, at least get the EVGA powerlink to balance the load on the cables, that might be all you need.

        1. Hi,
          Thanks for your efforts: I was researching upgrading my MacPro, and was concerned about power, and thought of doing this.

          With respect to your analysis, my comments:
          (1) Rated Temperature:
          The 35 C is for the ENTIRE system, and not the PSU. Basically it’s saying if your room is hotter than that, the OEM cooling will likely be enough to run the individual components at their specced performance. So get an air con…

          (2) PSU Output:
          This is a (very) standard switched-mode power supply. The outputs for the various voltages are COMMON. ALL of the 12V is from the SAME source. Same for the grounds.

          And you can confirm that yourself many ways:
          (a) From the PSU circuit diagram
          (b) From loking at the wiring inside the PSU
          (c) By looking at the male end of the connector on the logic board. There are a total of four connector spades: 2 x 12V, 2 x ground. So you can see that each 12V lead is joined again on left and right side. And then you can see it is all connected in the logic board into one common 12V bus: test the resistance, and it shows that.

          The only reason it has multiple grounds and voltage leads is to split up the current to different wires, and especially to different connectors at each end of the leads attached to it.

          The weak link in a system ialwsays the connectors: as you commented, a Molex Jr. contact carries a max of 9A. And the Apple logic board has somewhat heavier duty contacts for the 4 x 12V leads (again, can see for yourself)

          If your PSU is actually going to supply up to its max 12V output, in this case 79A (which is the 12V rating: look on the PSU, it’s given there ), each contact of its 4 x connectors has to carry 20A.
          That’s quite a lot.

          For wires , I haven’t bared mine yet, but it looks like AWG 12. And for the short length in the Mac Pro, you can easily carry 20amps per wire (confirm on Google of you want).

          (3) MacPro 300W for 4 x PCIe slots limit:
          That’s going to be the circuit board traces that limits that. Based on the components that Apple sold at the time, the designer won’t have seen any reason to provide any more. Adding more greatly increases the cost of the circuit board.

          (4) MacPro PCIE AUX A and B limits:
          The GPU can get power from its PCIe connector (2 or 3 or 4 x 16). But that’s limited to 75W.
          Anything more, and it needs a separate power supply input.
          The PCIEe spec covers that, and the manufacturer can use either 6 or 8 pin cables, where the 6 pin suplies up to 75W (using either 2 or 3 lines at 12V) and the 8 pins have 150W supply.

          So the MacPro as designed, with 6-pin Aux supply be able to provide a GPU the following: 75 W from the PCIe x 16 and 2 x 75W if it were to use both 6 pin Aux, total of 225W.

          Note that the connectors in the PCIE Aux are rated at 8 Amps per contact max, but the PCIe spec doesn’t go to that extreme.

          This is a good reference:

          Anyways, I agree with your conclusion: this modification makes sense.

          In terms of a given GPU drawing more than its TBD, remember that the TBD is not actually the maximum peak power a GPU or other component) might draw. Depending on manufacturer, the actual peak might be LESS, or even more (as much as 50% more).

          It’s affected by marketing as much as engineering. As long as a given GPU can live within the PCIe supply spec (75W for 16pin + 75W per 6 pin, 15W for 8 pin Aux) then there’s obviously some pressure to not admit the max/worst case since the Customer would prefer a lower TDP.

          And they know that the PSU’s are usually overspecced, so there’s not a big risk there.

          The only time you’ll run into problems if the End User ends up trying to do too much with too little: like run too big a GPU in a system not designed for it. If yous system shuts down because of excessive power draw after you’ve changed from OEM, it’s “pilot error”…..

  4. Would it be possible to use a Radeon VII without the Pixlas Mod by using a dual mini 6pin to 8pin & a two SATA to 8 pin cable?

    If not does anyone have an external PSU to recommend?

    I want to do the Pixlas Mod as a last resort.

      1. SilverStone Nightjar Fanless Titanium NJ600 600W

        I own this one and I am very happy with it.
        It’s super expensive, though.

  5. Just another configuration: Nvidia GTX1080TI with EVGA powerlink to balance the load and dual SATA drive supplemental power sources, still shuts down under load. The load is Premier Pro exporting a time line to an h264 movie.

  6. I am going to try a VEGA 56 in my Mac Pro (an RX5700XT would have been better but the price of the VEGA was too good to pass). Any hint as to whether I should go for
    * 2x mini 6-pin to 8-pin; or
    * 1x dual mini 6-pin to single 8-pin + dual SATA to single 8-pin?

    1. I would not sacrifice SATA ports personally. You can try the mini 6 to 8 pin, at least power draw *should* be balanced that way. If the GPU manufacturer can not guarantee the card does power balancing over both ports, I’d highly recommend an EVGA PowerLink though.

  7. So you mention that its not recommended to go 2x 6 pin to 1x 8 pin
    2x sata to 1x 8pin
    because we don’t know the capacity of the sata traces. My question is, if i use the sata cables from the optical drives those cables come directly from the PSU not from the Logicboard so, theory says it should be OK right?

    1. SATA does not come directly from the power supply, follow the cable, it runs to the backplane, only way to draw power directly from the p/s is Pixlas mod, or solder to the power supply pcb itself, or to solider the backplane connection spades , for most who care not to solider , or not have a high current solider station, Pixlas mod is how to do this….
      I have done all 3, they all work fine.
      Drawing more current that the backplane intended on these old machines may work for awhile, but will eventually cause the backplane to have issues, no boot, no sleep….ect. And you have to replace the backplane.
      I have done that as well….
      Bottom line is do the Pixlas mod, and call it a day, it’s easy.

  8. I would love to know your thoughts on using a Radeon VII? Would you recommend using Pixlas or would I be okay to get by using the 2x 6 to 8 pin adapters? Thanks in advance!

    1. I’ve heard from people using the Radeon VII that pixlas mod is definitely preferred. You can downvolt your card and it *should* be fine using just the board connectors but you will end up with a less powerful card so this may not be worth it.

  9. Recently discovered the Mac Pro Cheese Graters… very nice and upgradable systems to make impressive performers under $3k… much better than new high cost options. With the power & associated shutdowns as a concern, was looking at options to optimize /balance power distribution & requirements. As CPU (3.46), GPU requirement go up, can loss of other drains be improved: remove physical drives, upgrade Fans,etc.

    I plan doing the Pixlas mod, ordered cable. It may be duplication, but for cable mgmt or cosmetic appearance, no harm using the EVGA Powerlink with Pixlas that I see, maybe just a little more balance in the 6/8 pin connectors too with rails?

    Is there need for external PS if build includes 12 core 3.46 CPUs, 128GB Ram, 1080 Ti, and all SSD drives: PCIe Nvme RAID and SSD’s off SATA on the drive bays. My thought was avoiding mechanical HD’s saved on Watts… leaving room for optical drive(s) too.

    Along the power discussion; any mod for chassis Fans. Newer fans perform well with less power too don’t they?

  10. My experiences so far:

    Mac Pro 4,1 -> 5,1
    2x X5675 @ 3.06 GHz (90W TDP)
    64 GB DDR3 ECC RAM (6 modules)
    1 PCIe nvme module
    1 spinning disk

    1x RX 560 without any extra power requirements:
    Of course it works. =) 35W tops pulling all from the PCIe slot. Under Windows, I managed to overclock it using AMD Auto Tune and it started pulling 48W.
    System idles at 130W according to HWMonitor.

    1x RX 580 (Saffire Pulse):
    2x mini-6 pin to single 8 pin, it pulls 7W from the slot and 10W from each port in idle according to the HWMonitor.
    Under Windows, playing Flight Simulator, it pulls 155W total. I’m not sure how these are distributed. System never shuts down.

    1x Vega 64, reference design:
    2x mini-6 pin to single 8 pin and 1 SATA to 1 8 pin. Constant shutdowns. Setting card’s maximum power to -25% gives me 150W on Flight Simulator and the system would still turn off sometimes.

    2x mini-6 pin to single 8 pin and 2 SATA to 1 8 pin. After flying for many many hours, I got once a shutdown during a load screen. I set the maximum power to -20%, the card runs at 180W and I haven’t had a shut down since.

    My plan is to use an external power supply to drive the Vega 64 and connect the 2 mini 6 pin to a thunderbolt card, use the RX 560 and connect the Vega only when I want to play. The system makes weird electronic noises with the Vega, even in idle, and it sounds fine with the RX 580 or the RX 560.

Leave a Reply

Your email address will not be published. Required fields are marked *