These are RX 580 2048SPs and not compatible with Apple AMD GPU drivers, since these are factory faked RX 570s flashed with a PCIe ID that is not present inside any of the Apple GPU drivers.
Works for Windows.
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 Video Card Recommendation for Mac Pro 5,1 in 2024
- Thread starter matthewpomar
- Start date
- Sort by reaction score
These are RX 580 2048SPs and not compatible with Apple AMD GPU drivers, since these are factory faked RX 570s flashed with a PCIe ID that is not present inside any of the Apple GPU drivers.
Works for Windows.
OK, good to know. Glad I ordered the Vega 64s. Gives me the two correct RX 580s that I need to pass down. Thank you for the heads-up on those newer cards.
There are several threads regarding the VEGA 56 which discuss this card in great detail.
In summary:
the VEGA56 has a low power bios mode something like 20% less power for 7% drop in performance. its a tiny micro switch on the side of the case.
The drivers are though thoughly optimised in Monterey.
It is prudent to use an EVGA power link ($20 on amazon) to distribute the power evenly from the mini 6 Pin power sockets on the motherboard.
I have been using a Vega 56 for a few years now for editing FCPX. I prefer the Sapphire branded product. Some reports of Powercolor cards not working properly.
Vega 56 is cheap on Ebay.
Last time I messed with a Mac Pro 5,1 a Vega56 was the best choice. Vega64 worked great as well, but it needs more power then the Mac Pro board can provide, either by modified PSU or because I was too lazy to do that, just a small 300w PSU to power the GPU.
I have old game rig PSUs just waiting to power the Vega64s. I like the Vega56/lower power bios idea though, sounds like a good solution if I ever need to upgrade the few remaining machines beyond 580s.
RX 460/560 is the best bang for the buck, while you can get cheaper cards like a GT 630, these are not supported after BigSur and will require patching.
So, spend a little more and get a RX 460 or a RX 560.
Also I do have a EVGA NVIDIA gtx 970, any idea if that would work. I bought dual-6to8 pin and it runs but black screen. I was reading maybe like the GT 630 (which I may just buy instead) you need to use the DVI port first then switch to DP , and it may need a flash to work ?
Forget NVIDIA cards.
Flashing a GPU is exclusively for pre-boot configuration support and solves absolutely NOTHING for driver support with macOS.
Support after High Sierra is only for Kepler cards compatible with Apple NVIDIA drivers, so, no support for a Maxwell one (your GTX 970 is a Maxwell based GPU) after 10.13.6. You can't even install any macOS release with a Maxwell GPU, you can only add the GPU after the NVIDIA WEB drivers are installed. You can make it work with a lot of intermediate steps and hacking, but then you have a Mac that is unmanageable and incapable of running METAL apps. Useless for a lot of workflows, IMHO.
Even for the Kepler compatible cards, that have METAL support from High Sierra to Big Sur, Apple completely terminated support for NVIDIA cards with Big Sur and then is an extremely ugly hack to make it work.
If you want to run Monterey and newer, you are extremely limited with AMD GPU choices, basically RX 460/560, 470/570, 480/580/590, Vega 56/64 and VII.
I would agree with others that the RX 560 is probably your best budget card. Reasonable performance if you aren't doing anything 3D. Some (all?) of them have garbage video issues at boot, but not so bad as to prevent you from being able to choose your startup drive etc. They work fine once booted.
I've got a couple Vega 64 cards now, but I'm seeing metal crashing on Ventura and Sonoma. These look to be reference-design cards (they appear to have the reference cooler design, anyway) (Powercolor brand, IIRC). I avoided XFX as the only brand listed on the OCLP pages to avoid.
Metal works fine in Monterey, but does not function in Ventura or Sonoma, throwing up:
Crashed Thread: 0 MainThrd Dispatch queue: com.Metal.DeviceDispatch
This is on a completely fresh build, with a GOP flashed boot-rom. (I built a fresh triple-boot system for testing Monterey, Ventura and Sonoma).
Any thoughts? Am I screwed on these? I have a couple random Vega 56 cards on the way also, along with EVGA Powerlinks, and am now expecting I may have problems there as well. (Trying to put together a few machines for a 3D Printing lab, and the slicing software on those can be GPU intensive.)
Ventura/Sonoma/Sequoia drivers are compiled with AVX2 instructions, which MacPro5,1 and 6,1 does not have.
So, it's an ugly hack installing Monterey drivers to overcome the missing AVX2 suport and make a GPU to work after 12.7.6 with the old Mac Pros.
Report your issues to OCLP devs, maybe they can address some of the issues.
Thanks, will do. Are there *any* known Vega cards currently working, or are Vega cards now considered non-functional from Ventura onward?
I expect I'll return all of these and simply invest in yet a few more RX 580s as I know they still function properly. Maybe I'll stick a Sapphire RX 590 in a couple of them even though the performance uplift is very low, since I already have the PSUs to power them setup (I'm thinking here, that the RX 590s draw just enough more power that RX 580s that it is inadvisable to run them from the dual-mini 6-pins?). I'll see if any of the Vega 56 cards work first though.
AFAIK, all GPUs that you can install to a MacPro5,1/6,1 running Ventura/Sonoma/Sequoia have some sort of issue with the drivers, even Polaris. Like you noticed, people seems to have less issues and better experience with Polaris. Not saying that is your case, but some workflows work, others doesn't. Btw, sometimes is not the GPU the real problem, but the app that you are running that requires AVX2, like several of the recent updated Adobe apps.
Check with OCLP developers for the best all around supported GPU model, seems to be Polaris, report your findings/issues with VEGA to them. Could be completely unrelated to your issue, but yesterday there was a commit for OCLP 1.6.1 about VEGA GPUs, if I remember correctly, a new METAL bin version.
macOS Sequoia Development by khronokernel · Pull Request #1137 · dortania/OpenCore-Legacy-Patcher
This PR is primarily for tracking development of patch sets and changes implemented in code base. Please note that these builds are highly unstable and will require frequent re-installs throughout ...
github.com
Even a standard non-overclocked RX 580 can trigger SMC emergency shutdown when running full throttle, since RX 590 draws more power…
Just curious, is the graphics performance of a Radeon RX 580 on a Mac Pro 5,1 running Sonoma or Ventura as good as in Monterey? And is it stable with OCLP 1.5? Is the AVX2 patch now complete?
Tricky question. The GPU drivers are from Monterey, OCLP removes the Ventura/Sonoma/Sequoia ones, so, "performance" is the same. Stability is not the same, tho.
If you don't use any workflow that triggers an issue with GPU drivers or any apps that require AVX/AVX2, Ventura/Sonoma is fairly stable.
Install and test your workflows.
Which AVX/AVX2 patch? You can't patch a missing family of CPU instructions that is now being widely used inside macOS. OCLP is trying to workaround it with drivers from Monterey (and earlier macOS drivers for Wi-Fi/BT) to replace the Ventura/Sonoma/Sequoia ones that require AVX/AVX2, but can't do anything about apps that are compiled to require AVX/AVX2.
This is a total waste of time. You can either run a 15 year old machine, or the latest versions of macOS. Not both.
If you need to run a macOS newer than Monterey, get yourself a newer Mac e.g. a 5K iMac. Even a 2012 MBP runs Sonoma fine.
I assume this Mac is only good in Sonoma for office work and some photo-editing. No video-editing, Blender etc.... Am I right?
There is no easy and simple answer, what works for one could not work for the other. You need to test your own workflow and see if it works acceptably or not. A lot of people is using Monterey for some workflows and Sonoma for others.
Get a spare disk and test yourself.
I tested Sonoma and it is disappointing on my Mac Pro 5,1. Wifi doesn't work on my legacy wifi card which was only supported in El Capitan and surprisingly in Monterey. There are glitches when I run a 3D test and it's not stable. Cinebench OSX and R15 don't run in Sonoma, they did in Monterey. The computer also boots slower.
I would agree with this statement. I'm in EDU, so it's important, for a variety of reasons, that we stay with a homogenous OS version. Would have stayed on Monterey otherwise for sure, as both Ventura (slightly), and Sonoma (more so), are less stable on our old MP 5,1s.
I do a fair amount of substantial work on 5,1s however, including 3D modeling, plenty of software that hits the GPU hard (Pixelmator's "ML Upscale, for example). Students also run Universe Simulator, which will saturate both the GPU and the CPUs fully. So far, knock on wood, they've held up very well in these scenarios. I expect this will be the last year for them however.
(A quick stat: Out of an original 24 (25?) machines that were donated to us we have thus far had a total of only one CPU Board failure and one fan failure (PSU.) That's it. Of course the original HDDs are nearly all long dead (a few survive), but that's some impressive reliability.)
A quick update on GPUs:
No luck with any of the Vega cards I tried; metal failures on anything later than Monterey. I recently found refurbished Sapphire RX580s available and snagged 4 of them. Only 2 of them worked, with the other two producing no boot screens (on machines with a working GOP flash) and eventually automatically dumping those systems into Apple's "Your operating system must be reinstalled to work on this computer" recovery process. My guess is that these two cards have been flashed with a non-standard firmware themselves, or are possibly a later revision of the Sapphire cards (looking briefly at the design, I feel there may be some very small differences, but I lack the time to investigate further and will simply be returning the cards that do not function.)
No luck with any of the Vega cards I tried; metal failures on anything later than Monterey. I recently found refurbished Sapphire RX580s available and snagged 4 of them. Only 2 of them worked, with the other two producing no boot screens (on machines with a working GOP flash) and eventually automatically dumping those systems into Apple's "Your operating system must be reinstalled to work on this computer" recovery process. My guess is that these two cards have been flashed with a non-standard firmware themselves, or are possibly a later revision of the Sapphire cards (looking briefly at the design, I feel there may be some very small differences, but I lack the time to investigate further and will simply be returning the cards that do not function.)
May I know how you check the RX580 can only output 5K 60Hz, but not 6K 60Hz?I thought I was finally able to get 6k @ 60Hz using this StarTech HDMI cable:
6ft 2m Certified HDMI 2.1 Cable - 8K/4K - HDMI® Cables & HDMI Adapters | StarTech.com
6ft (2m) HDMI 2.1 Cable 8K - Certified Ultra High Speed HDMI Cable 48Gbps - 8K 60Hz/4K 120Hz HDR10+ eARC - Ultra HD 8K HDMI Cable - Monitor/TV/Displaywww.startech.com
But I was only getting 5k (5120 x 2880) @ 60Hz and MacOS was just upscaling the resolution.
In general, macOS will only tell you the rendering resolution, and the UI resolution (with refresh rate). It won't tell you the actual resolution that was transmitted to the monitor.
e.g. If you are using 5K HiDPI 60Hz, then the screen is rendering at 10240x5760 60Hz, with everything in 5120x2880 2x scale. Then downscale back to 6144x3456 60Hz, and send this signal to the monitor. Therefore, you are actually using full 6K 60Hz, but the UI looks like 5K 60Hz.
macOS should not upscale 5k 60Hz signal to 6k 60Hz for you. This part usually done by the monitor. If your graphic card is really outputting 5k 60Hz, then it's 5k 60Hz. This is the final signal generated by the OS, no more upscale. And once the monitor get the signal, and it realise the signal won't fill up the entire screen, it may allow you to run in boxed mode. Or it will stretch the output to fill the 6k screen.
If macOS can upscale a 5k 60Hz signal to 6k 60Hz, then it means the card can output 6k 60Hz, then everything should be able to render at 6k 60Hz from the very beginning.
For example, the following is from my own cMP. As you can see, a RX580 can output 7680x2160 @144Hz (it's 87.5% more demanding than 6k 60Hz). There is no such monitor exist on the world when I made this screen capture. But it's because I enabled the HiDPI function, which makes the rendering resolution went beyond my monitor's resolution.
macOS can take this kind of order, and it is entirely normal. e.g. Many people use 2560x1440 HiDPI on a 3840x2160 monitor. So, macOS will actually render the screen at 5120x2880, then downscale back to 3840x2160, and send it to the monitor. But in macOS system report, you will only see rendering resolution (5120x2880), and UI resolution (2560x1440), and it won't show you the signal is actually transmitting at 3840x2160 (which means the monitor is fully utilised).
Also, different macOS version may have very different behaviour on those "non standard" resolutions. In some OS, everything shows up for user to select. On some other OS, you may need something like BetterDisplay to select what you want. In some extreme cases, you may need to build the entire profile in SwitchResX in order to use the exact resolution you want.