MP All Models AMD Polaris, Vega, & Navi GPU macOS Support

[PetteriKiller]

Oh..considering the Mac Pro 5.1.´s PCI-e slots the Sapphire RX5700/RX5700 XT has the nice height 36 mm.

Me like. It leaves a little extra space for ventilation compared to Sapphire Radeon RX580 Pulses 40 mm. height.

 

[bigtomato]

Finally my AMD 5700xt is recognized in Mac Catalina, thank you apple. !

 

[macguru9999]

Does the 5700/5700xt need the pixlas mod or can you power it from the motherboard ?

 

[PetteriKiller]

^ XT at least seems to have both 8-pin and 6-pin sockets side-by-side.

AMD Radeon RX 5700 XT Specs

AMD Navi 10, 1905 MHz, 2560 Cores, 160 TMUs, 64 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bit

www.techpowerup.com

..and the power needed 225 watts. Would it be enough to draw from the two Mac Pro's mini-pins one cable from each other plus the power from the PCI-e -slot?

 

[bigtomato]

Sorry, I am running a Mac mini 2018 with EGPU, its amazing !

 

[w1z]

Does the 5700/5700xt need the pixlas mod or can you power it from the motherboard ?

^ XT at least seems to have both 8-pin and 6-pin sockets side-by-side.

AMD Radeon RX 5700 XT Specs

AMD Navi 10, 1905 MHz, 2560 Cores, 160 TMUs, 64 ROPs, 8192 MB GDDR6, 1750 MHz, 256 bit

www.techpowerup.com

..and the power needed 225 watts. Would it be enough to draw from the two Mac Pro's mini-pins one cable from each other plus the power from the PCI-e -slot?

No, it does not need the Pixlas or any another power mod. You just need to connect both mini-pcie power ports to the card in a load-balanced setup. I would also recommend you go with the Sapphire Radeon Pulse RX 5700XT. The Nitro+ 5700XT series might be a problem both in terms of available space (length) and power requirements.

I am waiting for black friday/cyber monday deals as well as for 10.15.1 to be publicly released before I will order mine and would recommend you do the same unless you are in a rush.

Edit: Do keep in mind though that the Sapphire Pulse edition of the 5700XT is 6.5mm (2.3 Slot) wider than the Sapphire Pulse RX 580 (Dual or 2.0 Slot) so you 'might' have to use it in slot 2 of the cMP depending on what's currently occupying your PCIe slots.

 

[MIKX]

Looking at the RX 5700 XT pics from this link https://www.techpowerup.com/gpu-specs/radeon-rx-5700-xt.c3339
you could use an EVGa PowerLink either connected to the card if the power socket cable locks face upwards away from the GPU PCB ( like the HD 7xxxx & R9XXXX cards . . . . or . . . as a bridge PLUS extra 1 X 8 pin to 6 pin and 1 X 8 pin to 8 pin cable(s) FROM the PowerLink power out TO the GPU.

( NOTE : the PowerLink comes with a spare 6 pin installable socket. )

( NOTE : You can also add a TWO x SATA to 6pin or 8 pin TO the Powerlink power input. )

That way you'll have balanced power from . .

(a) 2 X Mini 6 pin to 8pin
+
(b) 2 X SATA to 6pin or 8 pin
+
(c) PCIe slot

Should be enough power and more.

=============================================

I use the above to power my MSI Armor RX 580 8gb. It works seamlessly.

My PowerLink sits ( securely ) on top of my PCI area Intake fan.
As my RX 580 has only one 8 pin power socket I needed one female 8 pin to male 8 pin from the PowerLink to the RX 580.
For any 8 pin & 6 pin GPUs two cables from the PowerLink will be required.

The pic below shows my messy cMP case interior ; I've been mucking around with the cable layout tonight.
It usually looks much neater, runs quietly, cool, fast.

Below is an older pic showing how a 'bridged' EVGa PowerLink can be installed.
That 'eraser' really holds the PowerLink very securely.

Currently I have added 2 X SATA cable to the PowerLink input.

 

[MisterAndrew]

The RX 5700 would be a good match for an unmodified cMP. The power levels are similar to RX 580. RX 5700 XT requires Pixlas mod or separate PS to be safe. It consumes more power than Vega 56 at full load (e.g. Furmark). It also runs hotter than the RX Vega 64.

The AMD Radeon RX 5700 XT & RX 5700 Review: Navi Renews Competition in the Midrange Market

www.anandtech.com

 

[MisterAndrew]

I don’t think we’ll see a huge improvement with better drivers. The Navi architecture wasn’t designed for compute workloads.
[automerge]1571449828[/automerge]


For gaming yes, but not for compute. Navi isn’t very suitable for workstation tasks.

Well, maybe that’s not so accurate what I said before. In the early benchmarks Navi compute performance was pretty bad, but perhaps that was due to immature drivers. It looks better in these benchmarks. It appears RX 5700 would be a solid upgrade over RX 580, but still not really an upgrade over Vega. I’m not going to recommend Navi for a cMP since it requires an unsupported version of macOS, but for those adventurous enough to enter the hackintosh realm, it may be a good option. (Note that the definition of hackintosh is running macOS on unsupported hardware. Running Catalina on a cMP does indeed fit that definition.)

RTX 2080 vs. Radeon VII vs. 5700 XT: Rendering and Compute Performance [Updated] - ExtremeTech

How do the AMD Radeon VII, Nvidia RTX 2080, and AMD's RDNA-based Radeon RX 5700 XT compare in professional applications? We take these solutions for a spin, with a particular examination of the new Blender version, 2.80.

www.extremetech.com

 

[w1z]

Well, maybe that’s not so accurate what I said before. In the early benchmarks Navi compute performance was pretty bad, but perhaps that was due to immature drivers. It looks better in these benchmarks. It appears RX 5700 would be a solid upgrade over RX 580, but still not really an upgrade over Vega. I’m not going to recommend Navi for a cMP since it requires an unsupported version of macOS, but for those adventurous enough to enter the hackintosh realm, it may be a good option. (Note that the definition of hackintosh is running macOS on unsupported hardware. Running Catalina on a cMP does indeed fit that definition.)

RTX 2080 vs. Radeon VII vs. 5700 XT: Rendering and Compute Performance [Updated] - ExtremeTech

How do the AMD Radeon VII, Nvidia RTX 2080, and AMD's RDNA-based Radeon RX 5700 XT compare in professional applications? We take these solutions for a spin, with a particular examination of the new Blender version, 2.80.

www.extremetech.com

AMD's reference and partner 5700XT cards have a cooling design flaw and run hot and noisy which is why I am recommending the Sapphire Pulse version. The pulse version runs cooler than the vega series but at higher performance levels than reference/partner cards. It is being recommended in all reviews I have seen of it in place of the reference/partner cards. The stock bios default mode will run fine under the cMP.

Update: Looking at @stupots card fit, photos, results and feedback on temp/noise levels, I'm going to try AMD's RX 5700XT card first before jumping on the Sapphire Pulse version of the 5700XT. The only difference between AMD's and Sapphire's versions of the 5700XT is the dual-bios option on the Sapphire which may be useful.

Update2: Digging deeper into the differences between AMD's reference PCB vs partner PCBs, it appears AMD has wasted no effort in populating their PCBs with quality components while partner PCBs are built with cost in mind. The only AMD partner that uses similar quality components is ASUS.

 

[stupots]

Comparisons for various Radeon cards in Heaven benchmark, all on Medium/2xAA/1920x1080 all on a vanilla Catalina install on the same machine.

RX 580 (Sapphire Nitro+)
View attachment 871049

RX Vega 56 (Reference blower model)
View attachment 871050

RX 5700 XT (Reference blower model)
View attachment 871053

Geekbench 5 results for the 5700 XT were:
Metal: 46182
OpenCL: 49839

the 5700 XT card draws around 220W during high load with instantaneous peeks up to around 260W. The reference board spec for the card is 225W so I’m quite happy running it on an un-modified 5,1.

I think what the benchmarks tell us is that the Navi drivers have a little way to go yet to see the cards full potential. I’m certainly already happy with the figures when taking into account the power draw of this card when compared to a Vega 64.

5700 XT Temps and noise was a lot better than when I tried this card in my PC. The blower models allow for a card directly above the GPU slot as well as getting all the hotter air directly out of the case and I prefer the look of it to the 5700 pulse models, but I can see why people would go for them.

And I think the blower models look neater

Stu

 

[h9826790]

Comparisons for various Radeon cards in Heaven benchmark, all on Medium/2xAA/1920x1080 all on a vanilla Catalina install on the same machine.

RX 580 (Sapphire Nitro+)
View attachment 871049

RX Vega 56 (Reference blower model)
View attachment 871050

RX 5700 XT (Reference blower model)
View attachment 871053

Geekbench 5 results for the 5700 XT were:
Metal: 46182
OpenCL: 49839

the 5700 XT card draws around 220W during high load with instantaneous peeks up to around 260W. The reference board spec for the card is 225W so I’m quite happy running it on an un-modified 5,1.

I think what the benchmarks tell us is that the Navi drivers have a little way to go yet to see the cards full potential. I’m certainly already happy with the figures when taking into account the power draw of this card when compared to a Vega 64.

5700 XT Temps and noise was a lot better than when I tried this card in my PC. The blower models allow for a card directly above the GPU slot as well as getting all the hotter air directly out of the case and I prefer the look of it to the 5700 pulse models, but I can see why people would go for them.

And I think the blower models look neater
View attachment 871056

Stu

Thanks for the info. But I suggest you may try Tessellation Extreme, 8AA, and highest possible resolution.

Otherwise, it’s easy to get into CPU limiting situation, and unable to compare the GPU performance difference.

 

[stupots]

@stupots You have the standard 5700 XT? I saw a test of a 5700 XT from MSI (overclocked with custom cooling and what not) and it could draw up to 390W on a chart there (Swedish website SweClockers):

https://cdn.sweclockers.com/artikel/diagram/18307?key=985a7a0bc0f8ab873e964220e1b8e4c1

@star-affinity Those look like total system figures to me i.e we know an RX 580 doesn't average 280w under load. Yes, I have the reference design RX 5700 XT.
This suggests around 222W Average under load.

@h9826790 Thanks, will do!

 

[stupots]

Thanks for the info. But I suggest you may try Tessellation Extreme, 8AA, and highest possible resolution.

Otherwise, it’s easy to get into CPU limiting situation, and unable to compare the GPU performance difference.

Here you go, Ultra Quality, Extreme Tessellation, 8xAA at 1920x1080.

I watched the benchmark all the way through and it looked stunning, sitting around 78 FPS for most of it. Card was above the system noise, but def not what I would call loud. Temperature of air coming out the back was warm but not hot. What tool can I run on a Mac to log temps during a benchmark?

 

[h9826790]

Here you go, Ultra Quality, Extreme Tessellation, 8xAA at 1920x1080.
View attachment 871123

I watched the benchmark all the way through and it looked stunning, sitting around 78 FPS for most of it. Card was above the system noise, but def not what I would call loud. Temperature of air coming out the back was warm but not hot. What tool can I run on a Mac to log temps during a benchmark?

Yeah, Tessellation really makes a huge difference on the visual. Especially when you look at the rocks, the dragon, the roof... in this benchmark.

Anyway, I just ran the same benchmark with the same settings in Mojave with my Radeon VII. As expected, at the same performance level as the 5700XT.

And I think if you run this benchmark (with this setting) again with the Vega56, more likely you can see the expected performance difference now.

Anyway, the very first thing to try in macOS to get GPU temperature is via this terminal command

ioreg -l |grep \"PerformanceStatistics\" | cut -d '{' -f 2 | tr '|' ',' | tr -d '}' | tr ',' '\n'|grep 'Temp\|Fan\|%\|\|Hz'

If this doesn't work, then I don't think there is anything can shows the GPU temperature in macOS.

 

[star-affinity]

Can't Unreal Heaven be run using Metal instead of OpenGL?

Like has been mentioned, maybe the CPUs in out MacPro5,1 are getting to weak to feed modern GPUs?

 

[h9826790]

Can't Unreal Heaven be run using Metal instead of OpenGL?

Like has been mentioned, maybe the CPUs in out MacPro5,1 are getting to weak to feed modern GPUs?

Unigine Heaven only support Direct X (Windows only) and OpenGL, no Metal support.

And I suggested to tune up the graphic options exactly because want to avoid unnecessary CPU bottleneck.

 

[Reindeer_Games]

Like has been mentioned, maybe the CPUs in out MacPro5,1 are getting to weak to feed modern GPUs?

At only 7.2% CPU usage at peak in scenes 2-4 (with RX 580), probably not (IMO).

RX 580 peaks out (with a slight dip, prob at scene change) at 100%.

Does the RX 5700 peak out at 100% usage as well, or cause a higher spike in CPU usage due to the higher frame rate?

 

[h9826790]

View attachment 871438

At only 7.2% CPU usage at peak in scenes 2-4 (with RX 580), probably not (IMO).

View attachment 871440

RX 580 peaks out (with a slight dip, prob at scene change) at 100%.

Does the RX 5700 peak out at 100% usage as well, or cause a higher spike in CPU usage due to the higher frame rate?

For a 6 cores 12 threads CPU, 8.3% CPU total usage may means 100% single thread loading.

If your cMP has 12 cores 24 threads, that will be further lowered to 4.2%. Even we put Turbo Boost into consideration, that will be also just about 4.5%.

If you turn off Tessellation, set the quality to low, AA to off, and resolution to 640x360... (make sure the benchmark is now CPU limiting). You should able to see that there is a process from Unigine Heaven cost about 110% in Activity Monitor (CPU single thread limiting).

And when the benchmark is GPU limiting, that process in Activity Monitor should stay below 100% (not CPU single thread limiting).

In no circumstance you can see 100% total CPU demand even you benchmark the GT120 or even worse GPU. Unigine Heaven isn't that CPU multi thread optimised.

 

[itdk92]

No, it does not need the Pixlas or any another power mod. You just need to connect both mini-pcie power ports to the card in a load-balanced setup. I would also recommend you go with the Sapphire Radeon Pulse RX 5700XT. The Nitro+ 5700XT series might be a problem both in terms of available space (length) and power requirements.

I am waiting for black friday/cyber monday deals as well as for 10.15.1 to be publicly released before I will order mine and would recommend you do the same unless you are in a rush.

Edit: Do keep in mind though that the Sapphire Pulse edition of the 5700XT is 6.5mm (2.3 Slot) wider than the Sapphire Pulse RX 580 (Dual or 2.0 Slot) so you 'might' have to use it in slot 2 of the cMP depending on what's currently occupying your PCIe slots.

In an ideal world, of course it needs Pixlas.

Any card drawing more than 200W, with two power sockets, of which one is an 8 pin requires Pixlas to be 100% safe, for power balancing purposes.

 

[w1z]

The nitro version of the card, yes.

 

[itdk92]

The nitro version of the card, yes.

Any version of this card, even the reference 225W model.

 

[w1z]

Any version of this card, even the reference 225W model.

That's incorrect. There are several forum members that are already using the 5700XT reference card in their cMPs without any power mod.

The Pulse version 'might' have issues when stress testing but I haven't tested that as I don't have the card yet.

 

[itdk92]

That's incorrect. There are several forum members that are already using the 5700XT reference card in their cMPs without any power mod.

The Pulse version 'might' have issues when stress testing but I haven't tested that as I don't have the card yet.

I beg to disagree, that's totally correct. Running a 5700 XT without Pixlas might be damaging to your Mac Pro logic board's traces unnecessarily.

Without restarting this years-long discussion - Just because some people use them and haven't experienced issues so far, and for what they know (because the traces on the logic board could already have taken some damage, but obviously users can't know before an actual failure) - that doesn't make it safe.

Of course we are talking ideal scenario here, I understand not everyone is up to perform a Pixlas mod, or pay somebody to do that.

 

[w1z]

I beg to disagree, that's totally correct. Running a 5700 XT without Pixlas might be damaging to your Mac Pro logic board's traces unnecessarily.

Without restarting this years-long discussion - Just because some people use them and haven't experienced issues so far, and for what they know (because the traces on the logic board could already have taken some damage, but obviously users can't know before an actual failure) - that doesn't make it safe.

Of course we are talking ideal scenario here, I understand not everyone is up to perform a Pixlas mod, or pay somebody to do that.

While I would never recommend powering an 8pin+8pin card without a power mod, my GTX 780 (8pin+6pin) SC has a tdp of 250W. I have been using it for over a year, with heavy gaming, and with 'ZERO' issues and no visible wear or damage to the traces on the board. The reviews state that this card peaks at 415W.

At 225W with short ms peaks up to 305/310W, the 5700XT (8pin+6pin) reference cards should be safe to use without a power mod unless it is proven otherwise. Though I would certainly recommend balancing the power draw which I have done with my GTX.