It looks like it really excels in FP64 performance so really great for those doing scientific compute workloads.
Last edited:
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 AMD Radeon VII on the Mac Pro 5,1 - Mojave 10.14.4 and 10.14.5 Beta driver results.
- Thread starter monoton
- Start date
- Sort by reaction score
That double-precision though… I know the Vega 64 would be capable of similar if not artificially limitted (which I actually didn't think it was to that degree before seeing that spreadsheet), but if you use DP, the Vega VII seems to be very good value... It's not like Nvidia competes that much on that front (in the GeForce market)
I am quite sure because we already hit a point that Unigine Heaven / Valley is CPU limiting on cMP. That's very clear on my 1080Ti (which has poorer driver efficiency than AMD card in macOS).
I simply put my 1080Ti into the Hackintosh with 8700k, then it can score something like 50% better then on cMP with W3690 (same OS, same benchmark, same setting).
does the GPU block the pci slot above it? can u still use a small pci-e card there like an usb3 pci-e card? I use every pci slot on my macpro so that could be a problem for me. thanks!
also... any chance amd included a mac bootscreen on the vii series? (i think nvdia did on the rtx)
i might actually finally replace my 1080ti with this radeon vii on my 4,1-->5,1 MacPro. I love my 1080Ti and I think is perfect for this machine but I kinda already lost hope for Nvidia Mojave drivers. But should I wait for Navi? Or go ahead with the Radeon VII? Only need to make sure it doesn't block any of the other pci-e slots.
I'm waiying for a Radeon VII Sapphire's.
Reason: I have a Hackintosh PC Gamer (Windows 10 Pro / Mac OS 10.14.4 Mojave / Linux Mint) & a Mac Pro 4.1 flashed 5.1 (Windows 10 Pro / MacOs 10.14.4 Mojave)
I sold my 1080ti because I was tired of waiting for Nvidia compatible Mojave webdrivers.
The first beta 10.14.5 brings the compatibility of the Radeon VII.
In addition I play less (mostly simulations) and I only have freesync screens.
In the meantime, I have a Vega 56 (Sapphire designe reference), normally planned for my Mac Pro, installed on the PC (with a Thermaltake Core P5 box & Tempered Glass Edition Kit).
I could compare with benchmark results that I kept (on all systems and both machines).
Of course, I will also be aware of the noise report on the sites that have tested the card.
Reason: I have a Hackintosh PC Gamer (Windows 10 Pro / Mac OS 10.14.4 Mojave / Linux Mint) & a Mac Pro 4.1 flashed 5.1 (Windows 10 Pro / MacOs 10.14.4 Mojave)
I sold my 1080ti because I was tired of waiting for Nvidia compatible Mojave webdrivers.
The first beta 10.14.5 brings the compatibility of the Radeon VII.
In addition I play less (mostly simulations) and I only have freesync screens.
In the meantime, I have a Vega 56 (Sapphire designe reference), normally planned for my Mac Pro, installed on the PC (with a Thermaltake Core P5 box & Tempered Glass Edition Kit).
I could compare with benchmark results that I kept (on all systems and both machines).
Of course, I will also be aware of the noise report on the sites that have tested the card.
Would you consider committing to a longer benchmark?
Blender foundation has a new CPU and GPU benchmark.
It's self sustained so you don't need a Blender install to run it and you don't need to know any settings—just click a button and check your score.
You can run a 'quick benchmark' on either CPU or GPU (CUDA or OpenCL depending on GPU), which renders two scenes and will result in a 'time to complete'. Faster is better.
The quick benchmark takes just over 20 minutes to run on a Mac Pro 5.1, 12 core 3.46GHz.
The full benchmark runs more scenes and takes 1h15min to 1h30min in average. Modern fast machines will do better, obviously, and high end GFX cards typically to better than CPU (in average).
When you're done you can upload your result on https://opendata.blender.org where you can compare your times against other configurations. I would recommend creating a profile before doing the benchmark, as it makes the upload of the result less error prone after committing over an hour of render time and testing.
In these times where we see a bit of thermal throttling in consumer machines, I do like that it takes a bit longer to run this, as I feel it's more representative of actual working conditions.
Unfortunately, Blender is dropping OpenCL for Mac, due to Apple dropping it. Maybe we'll see Metal support in an upcoming version.
Fans are great at idle in 10.14.5b1. Seems like the same experience I had with the Vega Frontier Edition in this beta. After testing I'll comment more on fan speed over load.
I'm running the quick test first then the full test in blender and will update with results as they come in.
The quick test took 17m16s to run, the full run took 1 hour 6 minutes to run. Uploading the results failed both times but I have the .txt's for both runs if the breakdown is important.
I'm running the quick test first then the full test in blender and will update with results as they come in.
The quick test took 17m16s to run, the full run took 1 hour 6 minutes to run. Uploading the results failed both times but I have the .txt's for both runs if the breakdown is important.
Last edited:
I tested Rise of the Tomb Raider at 2160p with all max settings except super scaling set to 1.0. I got the following results:
Mountain Peak: 58.82FPS (min:12.91, max: 100.84)
Syria: 43.41FPS (min 5.87, max: 77.53)
Geothermal Vally: 44.69FPS (min: 21.13, max: 80.33)
Overall score: 49.37FPS
I also ran Hitman (2016) maxed at 2160p and got about 58fps, but it outputted the results into a csv with frame times that aren't really useful for me to look into.
Mountain Peak: 58.82FPS (min:12.91, max: 100.84)
Syria: 43.41FPS (min 5.87, max: 77.53)
Geothermal Vally: 44.69FPS (min: 21.13, max: 80.33)
Overall score: 49.37FPS
I also ran Hitman (2016) maxed at 2160p and got about 58fps, but it outputted the results into a csv with frame times that aren't really useful for me to look into.
Did the fan behavior on the Frontier Edition improve with this beta?
Are you testing these games under macOS? The results look good for Rise of the Tomb Raider. Seems to be consistent with the results for Win 10 in the reviews.
Are you testing these games under macOS? The results look good for Rise of the Tomb Raider. Seems to be consistent with the results for Win 10 in the reviews.
Can behavior for FE was nice and quite in 10.14.5, and I tested these games in MacOS.
I do feel like running these benchmarks never got the graphics card very loud so I wonder if the fan curve tops off at a lower RPM than in Windows or if I wasn’t pushing it sustained enough.
I do feel like running these benchmarks never got the graphics card very loud so I wonder if the fan curve tops off at a lower RPM than in Windows or if I wasn’t pushing it sustained enough.
From the tests I've seen, the radeon VII yields 56-58fps at 4K on RoTR under Windows, considering that some graphical options are not available on the Mac (HBAO+, Tessellation). On the other hand, tested PCs have much faster CPUs than a 2009 Mac Pro.
So eksu's results look rather good and show that Metal drivers work well for this game.
[doublepost=1554744857][/doublepost]
I don't seen any "super scaling" setting for this game. I think it's in Hitman.
What type of anti-aliasing did you choose for RoTR?
Last edited:
On 10.14.5b2 my device name is "ATY,AMD,RadeonFramebuffer". I forgot to check before updating from 10.14.5b1, but have time machine if we need to go back and compare.
Thanks for catching me on that. I have everything maxed on the graphics tab, and under display tab I was just running with just "SMAA" (no 2x or 4x).
Last edited:
hey eksu. Does the Radeon VII block any other of the pci-e slots on the Mac Pro? Or are you able to use all pci slots on your mac pro? thanks man!
It's a reference cooler so it shouldn't block any slots, and it doesn't look like it to me. I haven't moved my SSD down into that slot to see how snug it is though.
I do worry with the open air design instead of a blower style GPU of blocking adequate airflow with something like the Highpoint in front of it though.
I have a question for those who upgraded to the Radeon VII. Compared to the Frontier Edition (or another card) has there been a difference in the temperatures of the case or other components? I ask this because I know some people want to know if the open-air design will make it warmer inside the case.
I don't have the Radeon VII, but from my experience between dual reference 7950 / R9 280, and open fan 1080Ti. I believe the PCIe compartment temperature will be at least 5C warmer with the Radeon VII.
Open the side panel will help a lot, but then the RAM will run warmer.
Remove all unnecessary little metal covers of the PCIe slot, leave the back end open, and increase the PCIe fan speed can help as well. But the PCIe compartment still warmer from my tests. And my 1080Ti is just 250W, no quite at 300W yet (Radeon VII).