MP 7,1 Sanity check on Mac Pro 2019 7,1 build in 2025

[Flint Ironstag]

When I build a new rig for VR, I'll be very sorely tempted by these (knuckles being unobtainable for all intents and purposes here):

FlipVR - The VR controller allows you to grab anything in reality | Shiftall

FlipVR, a new style of VR controller. It has the control panel (control buttons, thumbstick, grip, and trigger), which normally placed on the palm side of the hand, can be flipped (jumped up) to the back of the hand with a simple twist of the wrist. While flipped, you can grab a drink, control a...

en.shiftall.net

Interesting controllers - wonder if they will catch on? I have definitely needed to hold something or type and it breaks the immersion big time. I'm really looking for that killer app that will put me over the edge to get an Apple Vision Pro. I'll definitely wait till after the Sep 9th event.

Why are knuckles hard to get in Australia (right?)??

I want to do some architectural visualizations (walking around in buildings and able to adjust the lighting, etc), mechanical repair (exploding views of engines or other components), all kinds of stuff. Guess I need to shut up and schedule a demo.

Also raised my curiosity. All 'Bay searches seem to be for Apple's infinity fabric bridges or for the Mi100 (4-connections). They probably share design but Apple changed the orientation.

Since the Pro VII was based on Mi50, there may be similar parts, but likely incompatible.

I'm guessing from your image the 2-slot is "adjacent" and the 3-slot is "non-adjacent."

199-999762 AMD Infinity Fabric Link 2P Adjacent
199-999763 AMD Infinity Fabric Link 2P Non-Adjacent

199-999849 AMD Infinity Fabric Link 2P Adjacent (Pro VII)
199-999850 AMD Infinity Fabric Link 2P Non-Adjacent (Pro VII)

Yes, both ebay vendors (although I really suspect they are the same shop 😂) told me the Pro VIIs had their own 2 distinct IF bridges and were not compatible with any other GPU. I am looking for a pair of the 2P adjacent.

 

[mattspace]

Interesting controllers - wonder if they will catch on? I have definitely needed to hold something or type and it breaks the immersion big time. I'm really looking for that killer app that will put me over the edge to get an Apple Vision Pro. I'll definitely wait till after the Sep 9th event.

I'm not sure there's ever gong to be a killer app for AVP - most commercial use of VR requires the precision of lighthouses (Film & TV production) built around Vive MARS.

I think Apple is trying to find a way to use a headset to do traditional computing a different way, rather than to to the sort of innately 3D tasks that can't be done with a flat screen computer.

It's telling to me that there seems to be very little like Open Brush on AVP. It seems very much the content consumption device the iPad was initially criticised for being.

Why are knuckles hard to get in Australia (right?)??

Valve doesn't like having to provide a 1 year minimum warranty, from what I understand. EB Games (Gamestop owned) was set up as a distributor for Valve VR gear for a while, but seems to have given up on it.

I want to do some architectural visualizations (walking around in buildings and able to adjust the lighting, etc), mechanical repair (exploding views of engines or other components), all kinds of stuff. Guess I need to shut up and schedule a demo.

I'd be interested to see if any of that actually had AVP apps, beyond tech demos. It's a device very much caught between being the onboard processing of a cheap portable standalone headset, with the optics of an expensive tethered headset.

What Apple should have made, was a peripheral that talked to an iPhone, or Mac, rather than a dedicated platform device (which could VNC in to a Mac).

 

[Regulus67]

...I'm really looking for that killer app that will put me over the edge to get an Apple Vision Pro.
...
I want to do some architectural visualizations (walking around in buildings and able to adjust the lighting, etc), mechanical repair (exploding views of engines or other components), all kinds of stuff.

Shapr3D is an app that I have used on the iPad. Very easy to work with. But, and this is a big but.
The subscription model is expensive for anyone who aren't doing a lot of 3D work.

I got the 1-year sub upon starting out 4 years ago. But ended up using it only for the first 1.5 month after having paid $209. 😕
And starting all over would require me to relearn the app, as it has added a lot of new stuff.
That said, I would have been very happy to pay for it again, if I got a lifetime license, even with no free future upgrades.

Meet Shapr3D Apple Vision Pro | Edit and collaborate in CAD

Enterprise teams build better together with real-time 3D design and editing in the immersive environment of Spatial Computer-Aided Design in Shapr3D Vision Pro.

www.shapr3d.com

 

[mattspace]

Shapr3D is an app that I have used on the iPad. Very easy to work with. But, and this is a big but.
The subscription model is expensive for anyone who aren't doing a lot of 3D work.

I got the 1-year sub upon starting out 4 years ago. But ended up using it only for the first 1.5 month after having paid $209. 😕
And starting all over would require me to relearn the app, as it has added a lot of new stuff.
That said, I would have been very happy to pay for it again, if I got a lifetime license, even with no free future upgrades.

Meet Shapr3D Apple Vision Pro | Edit and collaborate in CAD

Enterprise teams build better together with real-time 3D design and editing in the immersive environment of Spatial Computer-Aided Design in Shapr3D Vision Pro.

www.shapr3d.com

Does that actually let you edit the model's geometry on the model, by interacting with it directly, or is it using the 3D as just a live preview for stuff you change in palettes / dialogues / other devices?

 

[Regulus67]

Does that actually let you edit the model's geometry on the model, by interacting with it directly, or is it using the 3D as just a live preview for stuff you change in palettes / dialogues / other devices?

I believe so. The website does not say there is any restriction on the Vision Pro, compared to the other platforms.

"With Shapr3D on Vision Pro, you can create fully editable digital prototypes that allow for quick adjustments without committing to physical builds."

When I used it on the iPad, it was just a matter of editing any value in degrees or length, with mm precision.
It can not import usdz files. Created in Apple's PhotoCatch (photogrammetry).
But you can export files in the usdz format.

The free version is very limited, but I would suggest to try that for several weeks, before considering subscription.

Free version:

2 draft projects. Delete them to create new.​

Low-res STL and 3MF only export​

* I also checked the subscription prices. Still high, and then VAT added (25% in Sweden) 😢
But you have all platform options included

 

[Regulus67]

An example when I used Shapr3D, compared to using photogrammetry.
I spent over 100 hours drawing and learning, and I was still not able to complete this complex model.
Sorry that I don´t have the rendered model to present.
But you can see the precise measurements.

100+ photos in PhotoCatch rendered the model in about half an hour on my iMac Pro with Vega 56 8GB.
I have not tested rendering speed on my Mac Pros yet.
No idea if any measurement is possible.

 

[Flint Ironstag]

An example when I used Shapr3D, compared to using photogrammetry.
I spent over 100 hours drawing and learning, and I was still not able to complete this complex model.
Sorry that I don´t have the rendered model to present.
But you can see the precise measurements.

Thank you for clear illustrations - cool stuff. Yes I would like to be able to scan objects in the field, then ingest that data on MP and possibly AVP.

I just upgraded to Sequoia from Big Sur, and Luxmark maxed all 8 GPUs immediately. Unfortunately neither my admittedly underpowered UPS nor the motel wiring were happy with it (shut down 15 seconds in), so no further benchmarking for the moment. Was grin-inducing while it lasted.

 

[mattspace]

I just upgraded to Sequoia from Big Sur, and Luxmark maxed all 8 GPUs immediately. Unfortunately neither my admittedly underpowered UPS nor the motel wiring were happy with it (shut down 15 seconds in), so no further benchmarking for the moment. Was grin-inducing while it lasted.

You're travelling with that rig??

 

[Flint Ironstag]

You're travelling with that rig??

Temporary setup. Super annoyed I can't run benchmarks. 😂

[edit] So I'm a fool and couldn't resist trying some more benchmarks.

Cinebench only recognizes a single GPU, and when I click RUN - it idles with zero CPU & GPU utilization for ~45 seconds then has to be force quit. My hardware is supported.

DiffusionBee only recognizes and uses a single GPU.

If some clever coder out there wants to write a power virus benchmark that can test ALL subsystems in a single app, the world wants it. Who else remembers when apps came with proper demos and benchmarks that only required clicking in the menu to run? Gee whiz.

 

[Regulus67]

I see iPower Resale has two used 4TB SSD kits for sale $999.
Not to bad a price, I would think

 

[Regulus67]

[edit] So I'm a fool and couldn't resist trying some more benchmarks.

Have you tested the Davinci Resolve benchmark yet?
Just a small reminder, if you forgot it

https://forums.macrumors.com/threads/davinci-resolve-studio-on-mac-pro-7-1.2458523/

 

[Flint Ironstag]

Davinci launches, does nothing for about 45 seconds then has to be force quit. I haven't been in the mood to read crash logs yet.

 

[Regulus67]

Davinci launches, does nothing for about 45 seconds then has to be force quit. I haven't been in the mood to read crash logs yet.

Oh I see. Not a great situation to be in, when the system crashes in both Cinebench and DaVinci

 

[Flint Ironstag]

Oh I see. Not a great situation to be in, when the system crashes in both Cinebench and DaVinci

I was wrong about Davinci - have tried a bunch of apps on little sleep. Davinci installs and launches fine, I just need to figure out how to run a benchmark this weekend. And I will be relocating somewhere with more robust electricity soon.

 

[shijan]

Try to test with IndigoBench https://indigorenderer.com/indigobench#download It can use multiple GPUs.

 

[Flint Ironstag]

Try to test with IndigoBench https://indigorenderer.com/indigobench#download It can use multiple GPUs.

This is exactly what a modern benchmark app should look and feel like. Unfortunately, This is as far as I get before the app crashes. 😂 It will do the CPU, a single W6800, and a single VII, then while rendering the next scene, whether it be supercar or multi GPU - it doesn't indicate which is up next - the app will crash.

I still need to relocate to better power, so I'll 100% try it again.

 

[Flint Ironstag]

Here it is with just the dual W6800X duos selected:

Indigo Benchmark v4.4.15, OS X 64-bit build.

Bedroom: 55.790 M samples/s (1600x900, 60.046 s, 1967.031 samples/pixel)

Supercar: 130.014 M samples/s (1600x900, 60.047 s, 4952.050 samples/pixel)

[edit to add]

Single 6800 core and single VII together (any more VII added and app will crash):

Indigo Benchmark v4.4.15, OS X 64-bit build.

AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon Vega Compute Engine (Apple) -

Bedroom: 20.908 M samples/s (1600x900, 60.047 s, 798.979 samples/pixel)

AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon Vega Compute Engine (Apple) -

Supercar: 49.208 M samples/s (1600x900, 60.047 s, 1902.139 samples/pixel)

[edit 2]

Both W6800X duos + single VII. Pretty linear scaling - I think we can safely extrapolate numbers once I get the power sorted out - at least for this benchmark.

Indigo Benchmark v4.4.15, OS X 64-bit build.
AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon Vega Compute Engine (Apple) -

Bedroom: 60.505 M samples/s (1600x900, 60.047 s, 2161.706 samples/pixel)


AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon PRO W6800X Duo Compute Engine (Apple), AMD Radeon Vega Compute Engine (Apple) -

Supercar: 143.186 M samples/s (1600x900, 60.046 s, 5341.401 samples/pixel)


Everything internal to the MP will run at full tilt just fine, but if I fire up the external GPUs there's definitely a power issue. This benchmark seems to scale in a linear fashion.

[edit] Sorry for the horrible formatting - once I have power sorted I will run benchmarks again and put them in some pretty graphs. Just wanted to capture and share some info with those interested. This is what makes the internet great.

 

[shijan]

Nice! Just to compare here is my test on Mojave with 6-core i7-8700K (overclocked to 4.6GHz) and Radeon Pro VII (flashed with Radeon VII bios, and using RadeonVII-PPT-Generator set to Core 1801MHz@1000mV and HBM 1100MHz@1075mV, and also with DeviceProperties Force_Load_FalconSMUFW and PP_WorkLoadPolicyMask 1)
Indigo Benchmark is ok, but not perfect because it always have rather large variation from test to test, especially in Supercar test. On my system it varies somewhere from 17.4 to 18.4 and also depends of monitor resolution. They also have benchmark results page https://indigorenderer.com/benchmark-results

 

[Flint Ironstag]

Noticeable boost with the BIOS flash. How's the fan noise vs. stock?

 

[shijan]

It is complicated to answer because currently my card hardware is far from stock. I changed thermal pads to Honeywell PTM7950 for chip and Honeywell HT10000 for VRM. New pads produce better cooling and hotspot is cool at default settings, but same time because cooler spins slower and internal airflow is low other components temperatures are increased (I tested this on Windows). Also in addition i replaced cooler which is quieter and have different default speed, so it require adjustment to spin when card is idle. Overall i set TedgeLimit (°C) to 75 in PPT and at max load hotspot remains within 89-91°C with any settings. Currently i set core 1801MHz@1037mV (whis was default voltage in Windows) and HBM 1100MHz@1168mV (i got some problems with undervolted memory in specific situations) Fan noise feels slightly louder than default, but internal components are cooler, even rear metal shell feels cooler due increased airflow. Anyway it is not a quiet card. (I somehow miss my legacy RX 580 NITRO+ which was 100% silent even during full load due giant heatsink)

In both macOS and Windows with default settings cooler behave very strange during longer 10-15 minutes LuxMark tests. First it spins well, but after some time it became slower and slower and temp hotspot temp increased to 100° (and maybe higher) and you start to feel some strange smell from melted plastic inside. And AMD says don't worry it is normal temperature. And then full internet of dead Vega and Radeon VII cards . This is because they set default TedgeLimit (°C) to 100. On Windows it is not a problem because driver allow to control fan speed manually with slider. But on macOS you can only rely to automatic settings, or adjust TedgeLimit (°C) in PPT.

I described about maintenance, repair and flashing here https://forums.macrumors.com/threads/amd-radeon-pro-vii.2235715/page-2?post=33576679#post-33576679

I personally use settings from RadeonBoost.kext plist injected using OpenCore plist. There is also different solution for newer macOS versions in last pages of those threads, but seems it works only on real Macs and don't works on Hackintoshes with WEG kext.
Threads about Power Play Tables and RadeonBoost settings:

https://forums.macrumors.com/threads/how-to-downvolt-a-radeon-vii.2194899/page-7?post=33901655#post-33901655

Tired of low Geekbench scores? Use RadeonBoost!

OUTDATED. PLEASE DON'T USE THIS KEXT ON ANYTHING LATER THAN CATALINA. THIS KEXT IS NOT BEING MAINTAINED ANYMORE BECAUSE I SWITCHED TO APPLE SILICON MACS Hello everyone! Together with the help of @DSM2.Hackintosh I built a Kext aiming at removing limitations put in place by Apple for PC...

forums.macrumors.com

https://www.hackintosh-forum.de/for...clocking-powerplaytable-lüfterkurve/?pageNo=1

[TOOL] Radeon VII PowerPlay Table Generator (OC, UV, Fan Curve)

Since apparently there is quite a big demand for it, I am now starting an official topic for my Radeon VII PowerPlay Table generator for macOS. It's a simple Excel spreadsheet where you can change the settings to your desired values. The result is a PowerPlay Table you can inject via Clover on yo...

www.insanelymac.com

Unfortunately i didn't test if it possible to apply Radeon VII PPT to original unflashed Pro VII. On Windows it is impossible to overclock/underclock Pro VII using driver UI. And if believe to video review to apply PPT you need to change Board ID in registry to fool Windows driver. Also, i guess, with flashed bios FP64 performance will be decreased and and all other specific Pro VII hardware options will not be available.

Here is also my collection of GPU performance tests. Don't expect exact precision because they are from different online reviews made on different machines and different drivers.

IndigoBench:            Bedroom Supercar
AMD Radeon RX 7900 XTX  29.08   74.11   FP32 61.39 TFLOPS, FP64 1.918 TFLOPS $1000 2022 https://www.techpowerup.com/gpu-specs/radeon-rx-7900-xtx.c3941 
AMD Radeon RX 7900 XT   22.47   61.58   FP32 51.48 TFLOPS, FP64 1.609 TFLOPS $900  2022 https://www.techpowerup.com/gpu-specs/radeon-rx-7900-xt.c3912 
AMD Radeon RX 9070 XT   22.12   57.41   FP32 48.66 TFLOPS, FP64 1.521 TFLOPS $600  2025 https://www.techpowerup.com/gpu-specs/radeon-rx-9070-xt.c4229
AMD Radeon RX 9070      21.85   57.52   FP32 36.13 TFLOPS, FP64 1.129 TFLOPS $550  2025 https://www.techpowerup.com/gpu-specs/radeon-rx-9070.c4250
AMD Radeon RX 6800 XT   19.50   48.22   FP32 20.74 TFLOPS, FP64 1.296 TFLOPS $650  2020 https://www.techpowerup.com/gpu-specs/radeon-rx-6800-xt.c3694 
AMD Radeon RX 6900 XT   17.24   38.68   FP32 23.04 TFLOPS, FP64 1.440 TFLOPS $1000 2020 https://www.techpowerup.com/gpu-specs/radeon-rx-6900-xt.c3481 
AMD Radeon PRO W6800    13.60   31.18   FP32 17.82 TFLOPS, FP64 1.113 TFLOPS $2250 2021 https://www.techpowerup.com/gpu-specs/radeon-pro-w6800.c3809 
AMD Radeon PRO W7700    13.46   33.50   FP32 31.95 TFLOPS, FP64 0.998 TFLOPS $1000 2023 https://www.techpowerup.com/gpu-specs/radeon-pro-w7700.c4184 
AMD Radeon RX 6800      12.83   28.11   FP32 16.17 TFLOPS, FP64 1,010 TFLOPS $580  2020 https://www.techpowerup.com/gpu-specs/radeon-rx-6800.c3713 
AMD Radeon RX 6700 XT   11.53   29.74   FP32 13.21 TFLOPS, FP64 0.825 TFLOPS $480  2021 https://www.techpowerup.com/gpu-specs/radeon-rx-6700-xt.c3695  
AMD Radeon RX 9060 XT   9.69    28.45   FP32 25.64 TFLOPS, FP64 0.801 TFLOPS $350  2025 https://www.techpowerup.com/gpu-specs/radeon-rx-9060-xt-16-gb.c4293
AMD Radeon RX 6700      9.11    24.25   FP32 11.29 TFLOPS, FP64 0.705 TFLOPS       2021 https://www.techpowerup.com/gpu-specs/radeon-rx-6700.c3716 
AMD Radeon Pro VII      8.28    18.36   FP32 13.06 TFLOPS, FP64 6.528 TFLOPS $1900 2020 https://www.techpowerup.com/gpu-specs/radeon-pro-vii.c3575 (my test on Mojave, flashed with Radeon VII vbios)
AMD Radeon VII          8.26    16.67   FP32 13.44 TFLOPS, FP64 3.360 TFLOPS $700  2019 https://www.techpowerup.com/gpu-specs/radeon-vii.c3358 
AMD Radeon PRO W7600    7.93    20.34   FP32 19.99 TFLOPS, FP64 0.624 TFLOPS $600  2023 https://www.techpowerup.com/gpu-specs/radeon-pro-w7600.c4169 
AMD Radeon Pro VII      7.53    17.49   FP32 13.06 TFLOPS, FP64 6.528 TFLOPS $1900 2020 https://www.techpowerup.com/gpu-specs/radeon-pro-vii.c3575 (my test on Catalina, original vbios)
AMD Radeon RX 5700 XT   7.09    15.58   FP32 9.754 TFLOPS, FP64 0.609 TFLOPS $400  2019 https://www.techpowerup.com/gpu-specs/radeon-rx-5700-xt.c3339 
AMD Radeon RX 6600 XT   6.95    20.10   FP32 10.60 TFLOPS, FP64 0.662 TFLOPS $380  2021 https://www.techpowerup.com/gpu-specs/radeon-rx-6600-xt.c3774  
AMD NITRO+ RX 580       3.80    9.7     FP32 6.502 TFLOPS, FP64 0.406 TFLOPS $280  2017 https://www.techpowerup.com/gpu-specs/sapphire-nitro-rx-580-8-gb.b4436  


LUXMARK:                Hotel   Mic     Ball
AMD Radeon RX 7900 XTX  11671   65254   88322
AMD Radeon RX 7900 XT   12941   53081   70395
AMD Radeon RX 7800 XT   11976   40273   53369
AMD Radeon RX 6950 XT   9602    48466   6130
AMD Radeon RX 7700 XT   9438    32374   41723
AMD Radeon RX 6900 XT   8761
AMD Radeon RX 6800 XT   8290    41850   56589
AMD Radeon VII          7728    20856   48890
AMD Radeon RX 6800      7211    36192   51553
AMD Radeon Pro VII      7303    25085   52613 (my test on Mojave, flashed with Radeon VII vbios, OC Core=1801MHz HBM=1100MHz)
AMD Radeon Pro VII      6884    23245   50688 (my test on Mojave, flashed with Radeon VII vbios, OC Core=1645MHz HBM=1100MHz)
AMD Radeon Pro VII      6413    22299   41463 (my test on Catalina, original vbios)
AMD Radeon RX Vega 64   5855
AMD Radeon RX Vega 56   5169
AMD Radeon RX 6700 XT   4779    24632   36689
AMD Radeon RX 590       3177    10881   15798
AMD Radeon RX 5700 XT   3052    10835   29975
AMD Radeon RX 580 NITRO+                15241 (my test on Mojave)

 

[Flint Ironstag]

@ Shijan tremendous work on your GPUs. Those benchmark numbers motivate me to upgrade my power situation.

I wonder exactly how much GPU can a 2019 rackmount MP accommodate? I have two ancient and idle Nvidia GPU PCIE expansion chassis (loud) that hold 4 passive cooling GPUs each. Even though Nvidia, I think it's agnostic and MP would recognize it just fine if populated with some cheap Vega 64s - same as the Cube3 on current box.

Even more interesting would be a cheap-ass RX580 mining rig (or several) 😂 - just to see. I would have to record that.

Y'all know in some apocalyptic future, MP 2019 with GPU expansion chassis is going to be one of the most highly sought after bartering items. With enough Radeons and the right LLM, you've got your own standalone genie in a bottle! 😂😐

 

[mattspace]

Y'all know in some apocalyptic future, MP 2019 with GPU expansion chassis is going to be one of the most highly sought after bartering items. With enough Radeons and the right LLM, you've got your own standalone genie in a bottle! 😂😐

Assuming we don't follow our current track, and end up in ~30 years with steam engines being the most advanced technology our species can produce.

 

[Regulus67]

I was considering upgrading to 1.5 TB RAM, for running LLM.
But the more I read about the models, and the failure to produce reliable, consistent content. I have decided, I will not touch it with a ten foot pole.

It does not make sense in a financially sense either. The billions that have been invested, and not one use case that proves that it will give any ROI (return on investment).
There is one clear use case. As part of a surveillance system grid, nothing comes close.
In this case, cost has no relevance at all. Because other priorities take precedence.

Search for "Larry Ellison predicts rise of the modern surveillance state" September 2024.

Norway is also jumping on the train, building an "Operation Stargate".

.... MP 2019 with GPU expansion chassis is going to be one of the most highly sought after bartering items. With enough Radeons and the right LLM, you've got your own standalone genie in a bottle! 😂😐

I would rather gamble on an iPad Pro for future usefulness, if it is the cellular version.
Electricity is already being routed to ever larger Data Centres. While the end user is burdened with increasing electrical bills.

 

[mikas]

A little off topic sorry, but to compare Flint and shijan Indigo results to my 3x Radeon VII in a 5,1 Monterey:

Downvolted with Open Core plist config and powered by dual Pixlas mod + dual SATA power.
The score seems roughly to match a single RX 6800XT or single RTX 4070 Ti.

This (Total 66,332) would rank as number 20 in Indigo user benchmark "highlighted" list, just between these two in the screenshot. Don't know what this "highlighted" list means though.

My single RTX4080 on HP Z4 scored about 86,500.