One card like this (combine both mini 6pin as a single power source).
2nd card by Pixlas mod.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 1,1-5,1 AMD Radeon VII on the Mac Pro 5,1 - Mojave 10.14.4 and 10.14.5 Beta driver results.
- Thread starter monoton
- Start date
- Sort by reaction score
One card like this (combine both mini 6pin as a single power source).
2nd card by Pixlas mod.
Thanks for the quick reply!
That makes more sense as it will balance the load over the 2x 6 pin socks.
I already have a 2x mini 6pin > 1x 8pin that was using on the RX580, ill just order a 1x8pin > 2x 8pin to split that.
Also I just checked the Custom PowerPlay table I'm using for the VII and its 981mv (240w max) so if I do run into any power draw issues, ill just run a second pixlas cable.
Thanks for all your help
Arghhhh, just made a bad mistake, in removing the aux cables from the board, Ive knocked a small chip off the motherboard, so annoying, I even remember somewhere someone saying to be careful of this!
I haven’t attempted to power the machine back on yet. Does anyone know what this small chip does? It looks a bit to small for me to attempt re-soldering...
I haven’t attempted to power the machine back on yet. Does anyone know what this small chip does? It looks a bit to small for me to attempt re-soldering...
Attachments
Just as a heads up to anyone removing the PCI power cables from the backplane, BE CAREFULL. Its so easy for the cable to brush the chip on the way out and you wouldn't believe how easily the chip brushed off the board.
Found this thread https://forums.macrumors.com/threads/scraped-an-op-amp-from-logic-board-accidentally.2180650/
It appears the chip I knocked off is part of the power monitoring circuit for the AUX A power socket.
As a precaution I removed all valuable items from the machine and powered it on.
Booted up and works fine, I can confirm from looking at iStat Menus that the voltage for "PCI Boost A" is 0.00v, I would have to plug something into it to confirm whether its able to report the "current draw" on that socket but I'm not sure I want to risk toasting a GPU to test it.
Due to the age of these machines, its not economical to repair and is easier to buy a 2nd hand backplane.
However as the circuitry involved appears to be solely for monitoring the draw over the PCI power cables, its easier for me to just not use them, even though im sure it would still be fine.
I've currently powered the VII from the Pixlas cable alone, leaving AUX A & AUX B unplugged and everything is running fine, fully stress tested and no side effects so far.
In theory, I should be fine to run a second PIXLAS cable and plug the 2nd VII into that without any issues.
Looks like I've been lucky and got away with it this time, but ill be sure to be much more cautious if I have to remove the cables on any of the other MacPros.
Found this thread https://forums.macrumors.com/threads/scraped-an-op-amp-from-logic-board-accidentally.2180650/
It appears the chip I knocked off is part of the power monitoring circuit for the AUX A power socket.
As a precaution I removed all valuable items from the machine and powered it on.
Booted up and works fine, I can confirm from looking at iStat Menus that the voltage for "PCI Boost A" is 0.00v, I would have to plug something into it to confirm whether its able to report the "current draw" on that socket but I'm not sure I want to risk toasting a GPU to test it.
Due to the age of these machines, its not economical to repair and is easier to buy a 2nd hand backplane.
However as the circuitry involved appears to be solely for monitoring the draw over the PCI power cables, its easier for me to just not use them, even though im sure it would still be fine.
I've currently powered the VII from the Pixlas cable alone, leaving AUX A & AUX B unplugged and everything is running fine, fully stress tested and no side effects so far.
In theory, I should be fine to run a second PIXLAS cable and plug the 2nd VII into that without any issues.
Looks like I've been lucky and got away with it this time, but ill be sure to be much more cautious if I have to remove the cables on any of the other MacPros.
So the Dual VII's are working!
Only tested Luxmark and Bruce X FCPX so far.
Bruce x has gone from 12 seconds with the single VII down to 8-9 seconds.
Luxmark has gone from 52861 with the single up to 105791 with the duals.
Only tested Luxmark and Bruce X FCPX so far.
Bruce x has gone from 12 seconds with the single VII down to 8-9 seconds.
Luxmark has gone from 52861 with the single up to 105791 with the duals.
I've done exactly the same thing with my main MacPro5,1. It makes no difference to the power delivery through the logic board, the system just can't monitor usage on that particular connection any more. Thank goodness it was nothing more serious, I was sure I'd killed the thing.
So, you've successfully powered a Vega VII through the combined two AUX power ports on the logic board? And two through Pixla's mod? Interesting...
Last edited:
I'm powering both cards from 2 new PIXLAS cables, I could of however, powered one of the cards from the 2x standard 6 pin plugs (via 2x6pin cables> 1x 8pin > split to 2x 8pin to balance power across both connectors) and the other card via PIXLAS.
I only went the route of a double PIXLAS cable to be safe after my accident of knocking that chip off.
Its also important to stress that I would ONLY run dual VII's if they are downvolted otherwise they could in theory draw 300 watts each, where as downvolted to 981mv, they will only draw a theoretical max of 240 watts each. In real world usage however, even running LUXMARK, I've only managed to get them to draw around 180 watts each, meaning the PSU was still barely at 50% capacity in total.
Also worth noting the Dual Vii's comparison with the Vega II Duo that Barefeats ran. https://barefeats.com/mac-pro-2019-four-gpus.html
7.1 MacPro, LuxMark scores: Vega II Duo (£5000) 89284, Dual VII's (£1000 total) 100972.
The Dual VII's undervolted in my 5.1 scored 105791
Just need apple to continue optimising there software to lean on the GPU's more in FCPX and there might be a bit more life in these 5.1 yet before I purchase the 7.1
With the 2nd Radeon VII in Slot 4, is there still room above the card to install SATA SSDs? Or are the SAT ports too blocked.
It is physically possible, the space is there, however the SSD's is BAYs 2,3 & 4 would completely block the fans of the VII so I would strongly advise against it.
I have moved 2x SSD's into the optical drive bays, a 3rd SSD in Bay 1 and a 4th SSD on a PCI card (running a 6gbps rather than the slower 3gbps on the standard SATA ports)
Sonnet just release details of a new card that would be a good option, this combines a USB C port out the back of the card and Dual SSD's running at 6gbps (550mb's per second read/write), this would sit fine in slot 3.
www.sonnettech.com
I have moved 2x SSD's into the optical drive bays, a 3rd SSD in Bay 1 and a 4th SSD on a PCI card (running a 6gbps rather than the slower 3gbps on the standard SATA ports)
Sonnet just release details of a new card that would be a good option, this combines a USB C port out the back of the card and Dual SSD's running at 6gbps (550mb's per second read/write), this would sit fine in slot 3.
Fusion Dual 2.5-inch SSD RAID - SONNETTECH
Dual 2.5-inch SSD PCIe 3.0 card with hardware RAID controller, plus 10Gbps USB-C port. Add your own choice of SSDs.
First of all, thank you so much, Ian, for this super interesting discovery. I am really thrilled. You opened up more possibilities for the future because you might as well put two AMD 5700 XT Red Devils in there or even two AMD 5700 Pro cards.
They need less power, so it would be feasible from a power perspective. On top of that, chances are that the 5700 drivers will be great because of the upcoming 7.1 version we all wait for. The drivers will decide if a Catalina cMP with DosDude firmware and two 5700's will be the top dog on a very low budget. This is almost too good to be true. If Barefeats reads this, THIS config is what you need to benchmark next..
They need less power, so it would be feasible from a power perspective. On top of that, chances are that the 5700 drivers will be great because of the upcoming 7.1 version we all wait for. The drivers will decide if a Catalina cMP with DosDude firmware and two 5700's will be the top dog on a very low budget. This is almost too good to be true. If Barefeats reads this, THIS config is what you need to benchmark next..
Found an XFX Radeon VII for a decent price today, so I'm going to give it a try tomorrow as a potential replacement for my GTX 1080 Ti. Living on High Sierra is getting a little tricky these days, but I've loved the power of the NVIDIA card, so this looks like my best option.
I'll be sad to lose the boot screens (it's an MVC modified 1080 Ti). What el cheapo cards do you folks keep around in case you need boot screen access?
My initial plan is to use the dual mini 6-pin -> single 8-pin -> dual 8-pin trick mentioned by @h9826790 above, then switch to Pixla's mod (or downvolt) if I encounter any issues. Might see if I can squeeze in the EVGA PowerLink I'm using at the moment for the 1080 Ti. Sound like I'm on the right path?
I'll be sad to lose the boot screens (it's an MVC modified 1080 Ti). What el cheapo cards do you folks keep around in case you need boot screen access?
My initial plan is to use the dual mini 6-pin -> single 8-pin -> dual 8-pin trick mentioned by @h9826790 above, then switch to Pixla's mod (or downvolt) if I encounter any issues. Might see if I can squeeze in the EVGA PowerLink I'm using at the moment for the 1080 Ti. Sound like I'm on the right path?
If you've got an EVGA PowerLink, use that with 2x 6pin cables going into it.
I would downvolt the VII anyway as not only does it use less power, you actually get an increase in performance due to the card not having to throttle down to control heat.
Thanks to @h9826790, it couldn't be easier:
How to - downvolt a Radeon VII
OK, the goal achieved. Now, we can downvolt the Radeon VII with a kext (which only contain an info.plist to inject the required parameters), no need to modify the system kext anymore. OS update won't revert the modded kext. PRAM reset won't break the GPU driver. In fact, we can simply reset PRAM...
forums.macrumors.com
download the zip file and follow his instructions A-K
regarding cards with boot screens, I've got a Saphire 7950 Mac addition in the draw but the cheapest option is the NVIDIA GeForce GT 120 Mac Edition
Thanks for your advice, spent some time over the weekend setting this up - the hardest part was finding the second 8-pin output plug for the PowerLink!
I was amazed by how easy it was to be honest. I use the internal SATA drives for boot, so put one in a caddy and used my MacBook to install Catalina onto it and shut down at the first setup screen. Checked I'd done the
sudo nvram boot-args="-no_compat_check" setting from High Sierra Recovery on the MacPro5,1, then took out all the expansion leaving only that drive and the VII, just to be on the safe side. It booted (very quickly) and hit the setup screen on the first try. As a bonus, I guess the logic board checks for Handoff have been removed - that and the Apple Watch unlock were enabled and worked without having to tinker with anything.Played with downvolting the card in Windows and it looks like the one I have hasn't faired so well in the silicon lottery, I couldn't get it under 999 mV without artefacts at stock peak of 1801 MHz in Heaven and Valley. As mentioned, it's powered from the two mini 6-pin logic board connectors via the PowerLink and under LuxMark stress testing it reached about 7.5A on those (ooh!) and 2.3A on the slot. No shutdown or stability issues.
Thanks, I have one of those GT 120s lying around, so I've got a way out in case of trouble. I was toying with leaving a low-end card installed, but there's nothing single-slot and fanless which supports boot screens and Catalina. I suppose I could try the GTX645 tweaked by MVC, though I can't imagine Kepler GPUs have a bright future in macOS.
Time to edit the sig!
Grabbed a second VII today and just finished testing in Windows. The first one was definitely a Friday afternoon job; this one downvolts to 975 mV without issues - 981 mV crashed the whole system with the first one. Memory to 1100 MHz also not a problem. Now to tweak the kext for macOS and test over there.
Okay, at Vcore of 975 mV, HBM clock of 1100 MHz and a power limit setting of 220W I get 54063 in LuxMark, which is pretty tidy.
Hit 7.75A on the aux power and 2.88A on the slot, so (2x(7.75x12)+(2.88x12))=220.56W. That's 93W per aux power connector with no shutdowns or weird smells, thankfully.
The card seems pretty conservative on the draw from the slot; I'm going to try dropping the max draw down to 200W and see what happens.
UPDATE: Didn't make a blind bit of difference. Exactly the same power draw.
Hit 7.75A on the aux power and 2.88A on the slot, so (2x(7.75x12)+(2.88x12))=220.56W. That's 93W per aux power connector with no shutdowns or weird smells, thankfully.
The card seems pretty conservative on the draw from the slot; I'm going to try dropping the max draw down to 200W and see what happens.
UPDATE: Didn't make a blind bit of difference. Exactly the same power draw.
Last edited:
Good result on that card!
Also bare in mind that Luxmark is worst case scenario for the cards power draw.
I currently use Redshift with Cinema 4D on the other 2x MacPros with Dual 1080Ti's, this taxes the card at near enough maximum, but somehow, Luxmark always pull more current from them than anything else.
I tested AMD ProRender in Cinema 4D on the machine with Dual Vii's and it was the same result, iStat showed 100% GPU processor usage on both cards and yet they were drawing significantly less power (around 180 watts each) than what Luxmark manages.
Ah, thanks, I didn't know that. Is FurMark similarly hungry? It seemed to be, but it was late when I was testing last night and to be honest, I was just happy everything was working.
I've seen that before where the card is pegged at 99-100% but the power draw is around 160-180W. I play a few games under Windows too and it's usually the case that the power consumption is lower while the usage is high.
There was a further price drop on the card yesterday so a third one has just arrived at my door. Though I think I've got lucky enough on the one I have, I'll check that one out as well.
I'm not above a shill link, so if anyone's in the UK or EU, here are the XFX Radeon VII cards I found on Amazon. I've not seen much availability of them anywhere else right now, certainly not at their current price of £512.
Last edited:
Furmark is the most stressful benchmarks. Of course, if you limit the card to 220W, it will still only draw up to 220W. But if you remove the limitation, you should see that Furmark can draw way more than Luxmark.
e.g. This is from my 1080Ti (N.B. 97W is the max display limit, actual draw is more than that)
Wow, over 8A from the aux supplies? I take it you've never had any issues?
I was starting to feel a little nervous at 7.75A, but I'm aware there's a fair bit of FUD out there about the current these aux ports can supply. It's clearly more than the 75W a piece which is usually mentioned.
Also worth noting that there is generally no need to worry about the speed of the PCIe slot for compute work. (more of an effect for gaming)
If you are using GPU's for 3D rendering, the lane width of the PCIe slot only really effect the pre-render stage where scene information and textures are uploaded to the GPU, once done the cards will render at the same speed no matter what slot they are in.
If you are rendering out an animation, the textures are only loaded into the GPU at the start before frame 1 is rendered, they are then held in memory for all subsequent frames meaning all GPUs will then be rendering at virtually the same rate regardless of slot speed.
Here are my Luxmark results for each card:
GPU 1, Radeon VII, Slot 1 (x16) with 3 displays plugged into it - 52710
GPU 2, Radeon VII, Slot 4 (x4 shared with slot 3) no displays plugged in - 52877
GPU 1 & 2 together - 105791
*Update*
To summarise, the benefit of spacing the cards apart to help dissipate heat far outweighs the tiny un-noticeable speed gain by having the cards in slots 1&2 with the increased potential of overheating (causing the card to throttle back anyway) and reducing the life of the card.
If you are using GPU's for 3D rendering, the lane width of the PCIe slot only really effect the pre-render stage where scene information and textures are uploaded to the GPU, once done the cards will render at the same speed no matter what slot they are in.
If you are rendering out an animation, the textures are only loaded into the GPU at the start before frame 1 is rendered, they are then held in memory for all subsequent frames meaning all GPUs will then be rendering at virtually the same rate regardless of slot speed.
Here are my Luxmark results for each card:
GPU 1, Radeon VII, Slot 1 (x16) with 3 displays plugged into it - 52710
GPU 2, Radeon VII, Slot 4 (x4 shared with slot 3) no displays plugged in - 52877
GPU 1 & 2 together - 105791
*Update*
To summarise, the benefit of spacing the cards apart to help dissipate heat far outweighs the tiny un-noticeable speed gain by having the cards in slots 1&2 with the increased potential of overheating (causing the card to throttle back anyway) and reducing the life of the card.
Last edited:
Each mini 6pin can deliver up to ~120W. I did that on my cMP for years.
Mac Pro 5.1 owner, make it your king of Mac with 2 D700s
All reference design cards should be fine. The bios attached were for reference card.
forums.macrumors.com
Nice, an interesting read, thanks - I'll stop panicking about my measly ~8A current draw in that case. We've certainly come a long way in performance-per-watt metrics for GPU in the last few years too!Each mini 6pin can deliver up to ~120W. I did that on my cMP for years.
Mac Pro 5.1 owner, make it your king of Mac with 2 D700s
All reference design cards should be fine. The bios attached were for reference card.