MP 7,1 AMD Radeon Pro VII

[jtgil]

As promised, attached you see my new Vega VII workstation makeover with blue accents this time. (My dual tray is in storage at the moment)

Since I had many different GPUs in this thing I am used to the Vega VII level of performance or higher. What really really surprised me is the low noise duct fan. The reference Vega 64 / 56 back in the day was really a very bad designed and noisy card. Because of this I was very sceptical. However, since the 6800XT is not supported in the Sonoma, I sold my trusted mutant card and fell back to my 580 with AIO water block. But this time AMD surprised me full heartedly. The Radeon Pro Vega VII is very calm and the duct fan is not very noisy. I think that is the most important and surprising information. Big kudos to AMD Pro Series, they really did their homework when it comes to noisy duct fan solutions. Another surprise was the weight of that card and the fact that it's shroud is metal. It is really next level when it comes to AMD build quality. The hefty 1899,00 Dollar price tag did go somewhere.
Just for kicks, you will find my past RX 580 AIO setup attached as well.


View attachment 2435864View attachment 2435865

Wow, I would have never thought to see the day someone would do what I did back in 2020 and putting a sweet Pro VII into a 5,1.

I think I only found 2 people by 2023 that had done this but they hadn’t shared a lot of info on setup or performance, for that matter I had not either.

I also did it mid gpu-inflation, but even if I managed to pay AMD retail price which was lowest among resellers and other shops it was still 2k😭

I will post pictures of my setup later when I get back home.

Kudos to you friend, Pro VII is a hell of a card on our ageing beasts even in 2024-2025.

Super quiet and performant in everything I need from CAD, FEA, CFD, gaming, graphics works and so on.👌🏻

Edit: if anyone needs info or specs on this setup feel free to ask.

 

[mikas]

Answering your call a whopping 4 years later, I am delighted to report my upcoming project. About A week ago, I was able to snatch a brand new sealed in box Radeon Pro VII workstation card for 300 bucks. I did sent it to my friends in Poland to get it flashed with boot screens and modified device ID for a perfect fit for my 2012 Mac Pro 5.1.
More to come with a short performance review towards the end of October.

Waiting eagerly for your benchmarks.

I've got one too now as of today. I just put it in my Lenovo P710 monster (2xE5-2696v4, 44C/88T), and it seems to work fine and is indeed quiet too.

It's just that Cinebench 2024 fails to render with this Pro VII GPU. Just says "render failed", GPU is seen by CB2024 allright, and it tries to use HIP, but it don't even start before fail. No crash though.

Just wondering if CB2024 and Radeon Pro VII runs on a Mac? And with what results? Maybe @jtgil

 

[mikas]

I had in my thoughts to get two Pro-VII's and an infinity fabric link, and to get 32GB of GPU render RAM. I guess that was an afterthought or a surreal belief, and as such, realistically something of a not-gonna-happen-ever-thing.

 

[AlexMaximus]

I had in my thoughts to get two Pro-VII's and an infinity fabric link, and to get 32GB of GPU render RAM. I guess that was an afterthought or a surreal belief, and as such, realistically something of a not-gonna-happen-ever-thing.

As promised, attached you will find the screenshots from Heaven Benchmark that I took from 2019, 2023 and today in 2024. I used to have the regular Vega VII with the Alu shroud, the Alienware two slot 6800XT and now the Vega Pro VII card, which is the slowest and less noisiest of the bunch.
I sold my Alienware 6800XT card because my PS cramped out after a year of usage. So I figured, for the long run a less hungry card would be great, so I did go for the Pro Version this time. A real nice treat and a good middle of the road solution compared to the other mutant cards. All cards have been flashed by my friends in Poland and had a mac rom, including my current Pro card.

 

[jtgil]

Waiting eagerly for your benchmarks.

I've got one too now as of today. I just put it in my Lenovo P710 monster (2xE5-2696v4, 44C/88T), and it seems to work fine and is indeed quiet too.

It's just that Cinebench 2024 fails to render with this Pro VII GPU. Just says "render failed", GPU is seen by CB2024 allright, and it tries to use HIP, but it don't even start before fail. No crash though.

Just wondering if CB2024 and Radeon Pro VII runs on a Mac? And with what results? Maybe @jtgil

Hi mikas and Alex,

Apologies on the lack of contact. I will upload pictures of setup and will try to answer l your questions tomorrow.

But to answer in a more general manner my Mac pro 5,1 with the VII pro works perfectly in both Mac and windows.
-192gb ram dual cpu on the tray
-I run OC for max OS and hardware compatibility and performance.
-VII is not flashed.

Cinebench last I remember it worked fine, but I have not tried the cb2024. Will test it tomorrow too.

To the best of my knowledge, the dual VII pro with fabric link is feasible on the Mac Pro but you will be tied to windows only. It was something I wanted to do too (but never materialized lol)

If you guys have any more questions, I’m happy to answer.

 

[shijan]

Will Pro VII work on Mojave 10.14.6 same as regular VII? I seems like in supported GPUs list but status "Unverified" https://dortania.github.io/GPU-Buyers-Guide/modern-gpus/amd-gpu.html#vega-20-series

 

[mikas]

Thanks AlexMaximus and jtgil for your replys.

I managed to get CB2024 to run, but it predicted dismal results first, and then it hang after just under a minute of the 10 min test.

LuxMark gave me what I excpected, a little bit under the Radeon VII, in the ballbarks of 43000 pts. VII is like 50000 pts.

I'll go and run furmark next for some stress testing.

It's just that my software (ArchiCAD) uses cinerender internally (which CB2024 is based on too), with Redshift nowadays if you want to use it. It's kind of dissappointing it won't run the CB2024. So maybe I'll just go and test ArchiCAD to see what's happening with the card. Need to install the software package first.

 

[shijan]

So i finally find working Radeon Pro VII the card and tested it on Mojave 10.14.6. It boots, system works, but Metal hardware acceleration is disabled. UI is slow and laggy. Any ideas how to make it work with HW acceleration on Mojave? Maybe spoof device-id somehow? Flash Bios from Radeon VII (i guess bad idea)?

On Catalina 10.15.7 Metal 2 is enabled.
On Sonoma 14.7.1 Metal 3 is enabled.

 

[AlexMaximus]

So i finally find working Radeon Pro VII the card and tested it on Mojave 10.14.6. It boots, system works, but Metal hardware acceleration is disabled. UI is slow and laggy. Any ideas how to make it work with HW acceleration on Mojave? Maybe spoof device-id somehow? Flash Bios from Radeon VII (i guess bad idea)?

On Catalina 10.15.7 Metal 2 is enabled.
On Sonoma 14.7.1 Metal 3 is enabled.

You need this:

Mac Flashing Service for AMD Radeon Pro VII 16 GB | eBay

Our service will turn your PC card into a Mac card with grey Apple boot screen (also on 4K@60 Hz displays via miniDisplayPort). This service does not include Mac Pro power cables. You need to move the mouse on boot picker to uncover icons.

www.ebay.de

Contact them for international shipping at their website.

 

[shijan]

I don't use real Mac. Also it is way easier to change pci-id at program level in OpenCore instead of bios flashing.
In macOS Catalina temp under LuxMark test was 82-85 C. Seems high to me, so i decide to disassemble and look inside.
It was plenty of dust inside as expected. VRM thermal pads where not flexible. Just in case if someone look for pads replacement, this card use 1mm and 0.5mm pads.
Copper plate had same permanent dark rectangle spot from graphite pad as it was in disassembled earlier Radeon VII. Suggested by local repair service, i will try to use industrial hybrid filler Honeywell HT10000 for VRMs and for GPU chip as well.
Blower model BFB1012SHA01. Blower fan axis mounted on two bearings, and fixed with washer latch hidden under the rear sticker (something similar to described in this video youtube.com/watch?v=nH7cC5C-vWg ). Do not attempt to pull off fan with force, similar to older simpler bowlers that don't have bearings.

 

[shijan]

One more disassembly note. Radeon Pro VII GPU chip seems use 0.2mm Hitachi TC-HM03 graphite thermal pad. It is "phase change" thermal pad that changes state from hard to soft within some temperature range. At normal room temperatures like 22 ℃ it is hard and almost permanently glued to the chip. You need to pre-heat board from the back side to 40~50℃ and only then detach heat sink.
Here i made better photo of the board and the thermal pads thickness:

 

[shijan]

Honeywell HT10000 arrived yesterday!

Test 1. Honeywell HT10000 on VRMs and GPU chip. After Luxmark test chip temperature quickly increased to 96℃ and keeps rising, so i stopped the test. (original Hitachi TC-HM03 graphite thermal pad was 85℃ max). So Honeywell HT10000 don't works on this chip.

Test 2. Honeywell HT10000 on VRMs, Arctic Cooling MX-4 (6 years old thermal paste) on GPU chip. 88-94℃. (better than Honeywell HT10000, but still far from perfect)

Test 3. I installed Windows, to check other temperatures. It appears that macOS show me HotSpot temp, instead of GPU proximity temp, this is why i got so high values. As you can see VRM temperatures are very nice, so Honeywell HT10000 works well there.
Seems CPU of this card require some special thermal pad. I'll try to experiment with phase change thermal pad later.
Here are test results idle and after 15 min of Lixmark beenchmark when temperetures are stabilized and don't increase anymore.

 

[flyproductions]

All cards have been flashed by my friends in Poland and had a mac rom, including my current Pro card.

Do they have "real" ( = UGA) Mac-ROMs or just GOP enabled?

 

[flyproductions]

Honeywell HT10000 arrived yesterday!
View attachment 2454239Test 2. Honeywell HT10000 on VRMs, Arctic Cooling MX-4 (6 years old thermal paste) on GPU chip. 88-94℃. (better than Honeywell HT10000, but still far from perfect)

If the pic shows your application, i'm afraid you are using way too much of the stuff. At least on the GPU-die.

 

[AlexMaximus]

Do they have "real" ( = UGA) Mac-ROMs or just GOP enabled?

Those ‘friends’ are these guys here:

MacVidCards Europe

MacVidCards store in Europe. The latest and fastest graphics cards upgrades for your Apple Mac Pro computer.

www.macvidcards.eu

They are not related to the company in the US with a similar name. In the last 5 years they have flashed 4 cards for me. All of them have been flawless. The best and most valueable service ihey provide, is the flash service where you send your own card to them. This GOP you mentioned seems new to me, I am sure they do the real deal only.

 

[flyproductions]

Those ‘friends’ are these guys here:

MacVidCards Europe

MacVidCards store in Europe. The latest and fastest graphics cards upgrades for your Apple Mac Pro computer.

www.macvidcards.eu

Yes, that’s what i was suspecting...

They are not related to the company in the US with a similar name. In the last 5 years they have flashed 4 cards for me. All of them have been flawless. The best and most valueable service ihey provide, is the flash service where you send your own card to them. This GOP you mentioned seems new to me, I am sure they do the real deal only.

...but my thinking was, that this "legendary" discovery of enabeling the GOP-driver, which nearly any more modern card has, would have dried ot their business model a bit...

...as it can be done by anybody with minor skills in using a HEX-editor themselfs. It's as easy as copying the "enable GOP"-image in front of the GOP-image to the BIOS, doing some minor fixing to the header, mainly fixing the device-ID, deleting the same size of zero-bytes behind the GOP an flashing the result to the card. Only problem in some rare cases with some AMD-cards is having not enough "empty space" in the BIOS after the GOP-driver to delete.

I have done this to nearly any of the cards i own: GeForce 1070 and 1080, Radeon 570, 590, 6600xt, 6800 and Vega 64 by now. The only difference in functionality i could notice with the 1080 (which is dual BIOS and has GOP- as well as UGA-bootscreen-BIOS on it) is the GOP-bootscreen needing a few 10ths of a second longer to pop up after the starting-bong. That's it! Anything else, i. e. getting the boot-picker by holding Option, is perfectly the same. So nearly any "5,1 relevant" GPU can be turned into a "Mac Version" with a few simple steps today. A situation which we have been waiting for for years!

 

[shijan]

If the pic shows your application, i'm afraid you are using way too much of the stuff. At least on the GPU-die.

It is not so much as it looks like. HT10000 is way more dense and not too sticky compare to normal thermal compound, so it is near impossible create thin uniform layer. Also the extra of HT10000 was squeezed out as you can see on the photo after disassembly.
I ordered few other industrial thermal solutions from same local supplier. DOW TC-5550 compound and Honeywell PTM7950 0.2mm phase changing pad. Will test them soon...

 

[shijan]

New tests are here! DOW TC-5550 appears very nice compound, but Honeywell PTM7950 is slightly better. So i put DOW TC-5550 to CPU and to my old RX 580, and Honeywell PTM7950 to Radeon Pro VII.
Also one more important note. I didn't do long 20 min tests for original graphite pad and didn't monitor rpm fan speed, so it is not too honest to say that original graphite pad had 85℃ hotspot.

By default Radeon Pro VII GPU seems prefer to preserve hotspot temperature around 90-95℃ and lower fan rpm speed. So first time with Honeywell PTM7950 i got same high temps but fan speed lowered from 2200rpm to 1600-1700. I personally think that it is very bad logic, because card blower fan noise is near the same loud, but due lower airflow, vrm's and other components temperatures increases:

If manually set fan speed to something close to 2200rpm, i got these nice temps:

By the way, in Windows pro driver Radeon Pro VII have only 2 basic sliders for fan control. No any pretty custom charts to set rpm according to temperatures. Very unprofessional:

 

[h9826790]

Honeywell HT10000 arrived yesterday!
View attachment 2454239Test 1. Honeywell HT10000 on VRMs and GPU chip. After Luxmark test chip temperature quickly increased to 96℃ and keeps rising, so i stopped the test. (original Hitachi TC-HM03 graphite thermal pad was 85℃ max). So Honeywell HT10000 don't works on this chip.

Test 2. Honeywell HT10000 on VRMs, Arctic Cooling MX-4 (6 years old thermal paste) on GPU chip. 88-94℃. (better than Honeywell HT10000, but still far from perfect)

Test 3. I installed Windows, to check other temperatures. It appears that macOS show me HotSpot temp, instead of GPU proximity temp, this is why i got so high values. As you can see VRM temperatures are very nice, so Honeywell HT10000 works well there.
Seems CPU of this card require some special thermal pad. I'll try to experiment with phase change thermal pad later.
Here are test results idle and after 15 min of Lixmark beenchmark when temperetures are stabilized and don't increase anymore.
View attachment 2454256

You may need liquid metal, otherwise, it will be very hard to go back to the original cooling performance.

I replaced the graphite pad with liquid metal, and only 3°C improvement. I was very surprised how well the original graphite pad worked.

[Just saw your update that has a new and working graphite pad now]

Also, if you can control the voltage, you may lower it to 920mV (if you want to play safe, you may start from 940mV) for 1691MHz, 1037mV is too high. Lower the voltage can reduce a lot of heat.

 

[shijan]

Well, my thermal compound tests just illustrate a paradox: with better compound and cooler chip, you may got higher memory and VRM temps due lower fan speeds generated by default fan profile in GPU firmware.

I didn't try to adjust or undervolt Pro VII yet because i still try to spoof device-id and make it work on Mojave. This seems not so simple because Pro VII don't have usual acpi path. Also undervolt may not work at all because Pro VII is locked and it is impossible to overclock in on Windows or modify bios. I find a old video where Windows users create registry patch to spoof device-id and use non pro drivers to overclock Pro VII. But card became not stable and creates alarm noise from special internal speaker:

IORegistryExplorer shows that Mojave don't use AMDRadeonX5000 driver. I really don't know why this happens. By the way, i find reports from some users that regular VII card don't works on Mojave as well. So it could be some bug in OpenCore WEG kexst, or we may need some special version of Mojave, or need to use some specific mac model.
I also may try to use Catalina AMDRadeonX5000 kexst and linked bundles in Mojave, but i need to setup separate test system for that.

 

[h9826790]

Well, my thermal compound tests just illustrate a paradox: with better compound and cooler chip, you may got higher memory and VRM temps due lower fan speeds generated by default fan profile in GPU firmware.

I didn't try to adjust or undervolt Pro VII yet because i still try to spoof device id and make it work on Mojave. This seems not so simple because Pro VII don't have unusual acpi path. Also undervolt may not work at all because Pro VII is locked and it is impossible to overclock in on Windows or modify bios. I find a old video where windows users create registry patch to spoof device-id and use non pro drivers to overclock Pro VII. But card became not stable and start to sound alarm noise from special internal speaker:

IORegistryExplorer shows that Mojave don't use AMDRadeonX5000 driver. I really don't know why this happens. By the way, i find repors from some users that regular VII card don't works on Mojave as well. So it could be some bug in OpenCore WG kext, or we may need some special verion of Mojave, or need to use some specific mac model.
I also may try to use Catalina AMDRadeonX5000 kext and linked bundles in Mojave, but i need to setup separate test system for that.

You need a control environment (same room temperature, same fans speed, same loading) to confirm if the new thermal compound works as expected. But anyway, from your previous posts, no matter is it better than the factory graphite pad, there is no danger of over heating.

Regular Radeon VII can work in vanilla 10.14.6, I tested that. But not all versions of Mojave have driver for Radeon VII.

 

[shijan]

I looked for acpi path in Windows as described in this tutorial https://dortania.github.io/Getting-Started-With-ACPI/Universal/spoof.html and it didn't help. Card itself just use default aspi path for first pci slot, and after that creates some additional hidden pci-e routing inside (this is probably how infinity fabric internal pci-e connection works in this card). So older systems probably just can't detect it in usual way.


I done one more interesting hardware test. Blower fan in my card creates not too loud but very uncomfortable tr tr tr tr tr sound even in idle low rpm. Not sure if it helps, but i will try to replace it with similar model.

Test 1.
External shell and fan where removed and i tested card in passive cooling mode. It works well in idle with passive cooling with temps about 50℃. During Luxmark render hotspot and core temps quickly increased, so i sopped the test.

Test 2.
I put single 120mm 1250rpm fan at the bottom and run Luxmark render test. It holds 2-3 minutes until hotspot reach 100℃. Core temp still was rather low (75℃), but hotspot increased quickly to 100℃.


Test 3.
I put two 120mm 1250rpm fans at the bottom and angle them a little bit to create more directed airflow. Overall result was near the same - all components where very cool, but hotspot reach 100℃ after 3-4 minutes, and so i stopped the test.

 

[shijan]

Seems i find an explanation about unusual Radeon VII aspi path:

Windows Location Path is useless here, because some data just hidden under 0000:
ACPI(_SB_)#ACPI(PCI0)#ACPI(PEG0)#ACPI(PEGP)#PCI(0000)#PCI(0000)

Hackintool.app is more useful and gives me path like this:
IOService:/AppleACPIPlatformExpert/PCI0@0/AppleACPIPCI/PEG0@1/IOPP/PEGP@0/IOPP/pci-bridge@0/IOPP/GFX0@0

As you can see macOS adds special parameter "pci-bridge" but path \_SB_.PCI0.PEG0.PEGP.pci-bridge.GFX0 can't be used by normal aspi naming specification.

So here is problem fix https://github.com/acidanthera/OpenCorePkg/blob/master/Docs/AcpiSamples/Source/SSDT-BRG0.dsl "pci-bridge" should be renamed to "BRG0", saved as SSDT-BRG0.aml and applied in config.plist.
Now i have normal visible aspi path \_SB_.PCI0.PEG0.PEGP.BRG0.GFX0 and this allow me to apply device-id spoothing in Catalina with SSDT-GPU-SPOOF.aml. Spoofed card was detected as Radeon VII.

The bad news - Metal acceleration still don't works in Mojave

 

[h9826790]

I looked for acpi path in Windows as described in this tutorial https://dortania.github.io/Getting-Started-With-ACPI/Universal/spoof.html and it didn't helps. Card itself just use default aspi path for first pci slot, and after that creates some additional hidden pci-e routing inside (this is probably how infinity fabric connection works). So older systems probably just can't detect it in usual way.
View attachment 2457996

I done one more interesting hardware test. Blower fan in my card creates not too loud but very uncomfortable tr tr tr tr tr sound even in idle low rpm. Not sure if it helps, but i will try to replace it with similar model.

Test 1.
External shell and fan where removed and i tested card in passive cooling mode. It works well in idle with passive cooling with temps about 50℃. During Luxmark render hotspot and core temps quickly increased, so i sopped the test.

Test 2.
I put single 120mm 1250rpm fan at the bottom and run Luxmark render test. It holds 2-3 minutes until hotspot reach 100℃. Core temp still was rather low (75℃), but hotspot increased quickly to 100℃.
View attachment 2457998

Test 3.
I put two 120mm 1250rpm fans at the bottom and angle them a little bit to create more directed airflow. Overall result was near the same - all components where very cool, but hotspot reach 100℃ after 3-4 minutes, and so i stopped the test.
View attachment 2457997

Hotspot touch 100°C is normal, no need to worry about that.

 

[shijan]

Replacement fan arrived! It looks like improved version. Same size 25mm height and 75mm diameter. Same weight. Externally fan looks more accurate, cable is flatter. It is quiet at low rpm. It allow very low rpm spinning and also 0 rpm mode when temps are about 45℃ (old model always spin at 650-700 rpm in idle).


By the way, it is possible replace fan without radiator disassembly:
1. Unscrew screws around blower, to decrease tension between PCB and aluminum plate.
2. Slightly bend plastic fixing pins of molex connector with very tiny screwdriver and remove connector enclosure.
3. Carefully insert cable.