AMD stuff seems to love lots of watts, even threadripper needs more watts than any comparable intel cpu.
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.
AMD Radeon RX Vega Plug-and-Play in 10.13
- Thread starter theitsage
- Start date
- Sort by reaction score
AMD stuff seems to love lots of watts, even threadripper needs more watts than any comparable intel cpu.
Well, first you have it to turn on (an ATX PSU will not turn on without modification).
Use the paper clip trick, or buy a SWEX (ATX Power Supply Power-On Switch). With a SWEX you can turn it on and off.
Thanks. Yeah that I was aware of due to the lack of 24 pin connection. What I meant was once it's on. Can it be left on 24/7?
I don't know how to ask this without sounding rude so please don't take it that way.
Why would you turn it off? That's what I'm trying to understand. Is there any negatives besides using power from the wall when leaving it on? Will it hurt the video cards?
I do not recommend using the RX Vega 64 in the Mac Pro tower. We've seen power peaks exceeding 600W in an external eGPU setup (which has a 550W 80 Plus Gold single rail 12v PSU).
First, the PCIe AUX A power sensor in my MacPro4,1 seems to be defective. No current is reported at all in iStat. I've tried different cables and GPUs. Power still seems to be provided, but I can't be sure, so the following results need to be verified by someone else.
I've run some tests with a Sapphire Vega 56 in a MacPro4,1 (flashed to MP51.0084.B00).
Geekbench 4 (OpenCL):
GTX 680: 45372
RX 580: 135541
Vega 56: 157107
Geekbench 4 (Metal):
GTX 680: 35519
RX 580: 38629
Vega 56: 37341
Unigine Valley (3440x1440):
GTX 680: 628 (3.5fps - 33.3fps; 15.0fps avg.)
RX 580: 907 (12.6fps - 38.9fps; 21.7fps avg.)
Vega 56: 543 (6.8fps - 31fps; 13fps avg.)
In my opinion, the Vega 56 is currently not a suitable card. With improved drivers, it MIGHT work well at some point – as long as you don't stress it too much or provide additional power (SATA or additional PSU).
Sidenote: Using APFS volumes created with High Sierra under 10.12.6 may lead to data loss...
I've run some tests with a Sapphire Vega 56 in a MacPro4,1 (flashed to MP51.0084.B00).
- Doesn't work in 10.12.6
- Works in 10.13 Developer Beta 9
- Lots of artifacts in OpenGL
- Instant shutdown when running FurMark 0.7.0 @2560x1440 (on both dip switch settings)
Geekbench 4 (OpenCL):
GTX 680: 45372
RX 580: 135541
Vega 56: 157107
Geekbench 4 (Metal):
GTX 680: 35519
RX 580: 38629
Vega 56: 37341
Unigine Valley (3440x1440):
GTX 680: 628 (3.5fps - 33.3fps; 15.0fps avg.)
RX 580: 907 (12.6fps - 38.9fps; 21.7fps avg.)
Vega 56: 543 (6.8fps - 31fps; 13fps avg.)
In my opinion, the Vega 56 is currently not a suitable card. With improved drivers, it MIGHT work well at some point – as long as you don't stress it too much or provide additional power (SATA or additional PSU).
Sidenote: Using APFS volumes created with High Sierra under 10.12.6 may lead to data loss...
Last edited:
Well, no, not suitable currently because this is a High Sierra card and it's still in beta. The drivers should be very good in the future with the release of the iMac Pro.
The 7950 has 2x 6-pin connectors, limiting its draw from each port to 75w. Vega 56 has 2x 8-pin connectors, which can technically draw 150w each. Just because it doesn't make sense to you, doesn't make you right.
I got it now. Thank you for explaining it. I was going by a post from another thread where someone explained the 6-pins on a cMP can supply 120w each before it shuts down, but I agree that it's probably a bad idea to go over 75w on those. I would guess that the wiggle room is probably to allow for spikes in power consumption, so if the Vega 56 was consuming close to that limit sustained then a spike could damage the logic board.
Last edited:
I always thought the smart school of thought was 6 pin to 6 pin is fine, however 8 pin requires dual 6 pin connection.
Am I glad to hear that! You had us very worried about the RX 580!
Sorry to hear you have a power issue however, hopefully you can get it resolved somehow.
sorry if my last message was a bit blunt.
I find it quite odd that they included 2x 8-pin on the 56. The 5870 for example has a TDP of 188w, but this only has 2x 6-pin, giving it 37w of headroom. Vega 56 with 2x 6-pin or 1x 8-pin would only have headroom of 15w, probably not enough tolerance to compensate for power draw variance between GPU dies. But 1x 8-pin and 1x 6-pin should be more than enough.
Perhaps to cut costs, since the development of Vega cost so much, the PCBs for Vega (not including Vega Nano) are all the same, and the GPU dies are different. That would explain the rather bizarre reality of giving Vega 56 195w of headroom over its rated TDP.
All the new AMD stuff is high wattage even Tread ripper. It is a bit odd in comparison to Nvidia who seem to try to keep things power efficient. But it is also not surprising.
We can't really get mad at this from the perspective of it not being the best wattage situation for our older mac pros. I am sure our machines were not even in their thoughts when they built these cards. Also on the custom PC side they probably just had the logic that power supplies are available in any wattage you want these days people can just feed it whatever wattage it needs with no problem.
You had us very worried about the RX 580!
Sorry about that.
However, I still believe that the RX 580 can draw more than 75 watts from the slot. It would be helpful if someone else could run FurMark and check the values in iStat.
Wasn't so much trying to make the point that it doesn't fit simply into the Mac Pro pre 2013.. more that putting a 210w draw GPU on a board that can provide 375w doesn't make much sense.
Sorry about that.
However, I still believe that the RX 580 can draw more than 75 watts from the slot. It would be helpful if someone else could run FurMark and check the values in iStat.
I am quite sure it's programmed to draw that much from the slot, but not the cable's or cMP's issue.
If the cable is faulty, or Aux A can't deliver the required power, I believe the card will simply shows artifacts, but not so "smart" to re-distribute the power draw and get more from the slot.
I think the Vega 56 is a 64 that they crippled a little because it's chip didn't quite make the cut. However, it doesn't seem to be unlockable to a full 64 with a bios flash, which is fine with me since I wouldn't want it to draw that much power. I was disappointed that it isn't safe to run the 56 with the cMP's existing GPU power sources, but I don't have a problem with adding another connector to the PSU. You would think given the Mac Pro's intended use that Apple would have provided dedicated power from the PSU for high power GPUs.
Bear in mind that machine was designed 10 years ago when most PSUs were not modular and nobody foresaw such high power GPUs. We thought graphics processing power would increase while energy consumption would stay the same or better, mimicking CPU advances. Nvidia did follow this pattern after Kepler. AMD didn't.
There were high power GPUs available when the cMP 4,1 & 5,1 were designed such as the GTX 480 and HD 5970 that could not be safely powered using it's existing connections.
Really now? Can you tell me, why you did sell a GTX 1080? I am surprised, in my opinion the Vega 56 would be really not such a better deal, I really don't see such of a big difference or gain in GPU Power. - The 1080 is a great card that is not slow at all. Did you struggle witch the Nvidia Web driver?
In fact, by considering the cMP has 4 PCIe slot, and can install at least 2x double width GPU. I believe that Apple should provide at least 4x mini 6pin. Even though they only officially sell up to HD5870. They should allow us to choose 2x 5870 option (or 2x HD4870 on 4,1) to more utilise the PCIe slot (because technically it's impossible to install 3x5870 properly, therefore, I will say 4x mini 6pin is more reasonable than 6x).
If they have that. Then we can almost power any GPU we want now easily.
The 5,1 has good PSU, but no proper way to use the power. The 6,1 don't even have enough power at the beginning. IMO, none of the Mac Pro is really correct so far. Hopefully the 7,1 has a better "design".
If you read the post again I'm talking about the power consumption region Apple was happy to use. Nvidia filled that requirement after Kepler but they ****ed their relationship.
Yes, I was not satisfied with the Nvidia web driver. I was experiencing random slow-downs where the computer was not useable and occasional restarts due to a kernel panic. The panic reports specified the Nvidia web driver as the culprit. Those issues disappeared when I put my old HD 5870 back in. I also don't like having to wait for an Nvidia driver after each MacOS update. And besides that, I was excited to try the new Vega GPU. Since it's going into the new iMac Pro and will most likely be one of the officially supported eGPUs, it will have native drivers, so I thought that may work out better for me.