I've been putting together a 2019 Mac Pro in order to run AVX2-required software that my beloved 5,1 can't run. In doing so, I've encountered what appear to me to be temperature and performance issues with the AMD RX 6900XT GPU I purchased to go in it. My system:
Mac Pro 2019 CPU: 16-Core Intel Xeon W
RAM: 192GB
Storage: Sonnet M.2 4x4 Silent PCI card with:
- 2T Samsung 990 PRO SSD NVMe)
- 6TB RAID0 (3 x Samsung 990 PRO 2T NVMe)
MacOS: Sequoia 15.5
GPU1: AMD Radeon Pro 580X (8GB) (Removed at the moment)
GPU2: AMD Radeon RX 6900XT (Reference Edition)
OS: Sequoia 15.5
I'm in the setup and testing phase; I want to optimize performance and iron out as many glitches as possible before installing it in my office/shop. I purchased the RX 6900 XT used, of course, and cleaned and installed it in the system.
In testing it, I found that, in the MP2019, it very readily moves into what would ordinarily be considered the hot- to very-hot range. In fact it moves into the yellow or red zone (85˚C./90˚C.) almost the instant any serious benchmark or load is applied. I tried using MacsFanControl to increase fan speeds to cool the card but even at full blast this has a negligible effect on GPU temp. I would have thought that this was an issue with the card, such as the need to repaste the GPU. But I have little faith that it would make any difference. Why? Because this is NOT its behavior when installed in my 6-year-old bench PC (3.7GHz Intel Core i5, Windows 10, 16GB RAM, SSD storage).
When installed in the PC and running five different benchmarks (Sacred Path, Cinebench 2024, GeekBench GPU, Unigine Valley and Unigine Heaven), I had to work hard to exceed 85˚C. In fact, I actually went to the length of running benchmark utilities while playing an MMO, on the PC, to try to provoke the card into cracking 85˚C. I couldn't exceed 90˚C. at any time with any test, even at max settings. And the PC scores were better; markedly better.
Here are the GPU's performance scores and operating temps, as measured by both MacsFanControl (with MFC in strictly monitoring mode; all fans set to sensor-driven) and by iStat Menus (also in sensor-only mode). I don't know if either of these utilities is measuring the hot-spot temp on the die or the average die temp, though I suspect that it is the former.
Each benchmark is cross-platform (can be run on both Mac Pro/MacOS/Intel). Only Basemark's Sacred Path explicitly claims that its score can be compared across platforms, but all of them test the GPU, rather than the system as a whole and so should be at least roughly comparable. No temps are listed for Sacred Path, as attempting to access any other app or utility to check the temp aborts the test. All tests were run at maximum settings and typically achieved 99%-100% GPU usage. Where the test duration was not preset, tests were run for at least 5 minutes:
As you can see, the card, when installed in the PC, outperformed the Mac Pro installation by anywhere from 26% to 83%. And its temps were lower by 11% to 18%. This temperature difference may seem small but it is at the upper end of the temp scale and so lowering the temp is quite desirable.
The purpose of benchmarks and stress tests, of course, is to tax the card's performance to its fullest extent. What I don't understand is the radical difference in both performance and operating temperature between the two platforms. My questions are:
1) Do these operating characteristics look "normal" for other users of this setup (Mac Pro 2019/AMD RX6900XT)?
2) If not, what is your experience? And what do you think might account for that difference?
3) Can anyone suggest ways in which I might lower the card's temperatures?
4) Is there any way I might improve the card's performance?
5) Do you know of any other cross-platform benchmarks that can be compared head-to-head?
Any other words of wisdom, perspective or just general hob-nobbery on this subject would be very welcome.
In any case, thank you for the time and energy you spent reading this rather lengthy post!
Mac Pro 2019 CPU: 16-Core Intel Xeon W
RAM: 192GB
Storage: Sonnet M.2 4x4 Silent PCI card with:
- 2T Samsung 990 PRO SSD NVMe)
- 6TB RAID0 (3 x Samsung 990 PRO 2T NVMe)
MacOS: Sequoia 15.5
GPU1: AMD Radeon Pro 580X (8GB) (Removed at the moment)
GPU2: AMD Radeon RX 6900XT (Reference Edition)
OS: Sequoia 15.5
I'm in the setup and testing phase; I want to optimize performance and iron out as many glitches as possible before installing it in my office/shop. I purchased the RX 6900 XT used, of course, and cleaned and installed it in the system.
In testing it, I found that, in the MP2019, it very readily moves into what would ordinarily be considered the hot- to very-hot range. In fact it moves into the yellow or red zone (85˚C./90˚C.) almost the instant any serious benchmark or load is applied. I tried using MacsFanControl to increase fan speeds to cool the card but even at full blast this has a negligible effect on GPU temp. I would have thought that this was an issue with the card, such as the need to repaste the GPU. But I have little faith that it would make any difference. Why? Because this is NOT its behavior when installed in my 6-year-old bench PC (3.7GHz Intel Core i5, Windows 10, 16GB RAM, SSD storage).
When installed in the PC and running five different benchmarks (Sacred Path, Cinebench 2024, GeekBench GPU, Unigine Valley and Unigine Heaven), I had to work hard to exceed 85˚C. In fact, I actually went to the length of running benchmark utilities while playing an MMO, on the PC, to try to provoke the card into cracking 85˚C. I couldn't exceed 90˚C. at any time with any test, even at max settings. And the PC scores were better; markedly better.
Here are the GPU's performance scores and operating temps, as measured by both MacsFanControl (with MFC in strictly monitoring mode; all fans set to sensor-driven) and by iStat Menus (also in sensor-only mode). I don't know if either of these utilities is measuring the hot-spot temp on the die or the average die temp, though I suspect that it is the former.
Each benchmark is cross-platform (can be run on both Mac Pro/MacOS/Intel). Only Basemark's Sacred Path explicitly claims that its score can be compared across platforms, but all of them test the GPU, rather than the system as a whole and so should be at least roughly comparable. No temps are listed for Sacred Path, as attempting to access any other app or utility to check the temp aborts the test. All tests were run at maximum settings and typically achieved 99%-100% GPU usage. Where the test duration was not preset, tests were run for at least 5 minutes:
As you can see, the card, when installed in the PC, outperformed the Mac Pro installation by anywhere from 26% to 83%. And its temps were lower by 11% to 18%. This temperature difference may seem small but it is at the upper end of the temp scale and so lowering the temp is quite desirable.
The purpose of benchmarks and stress tests, of course, is to tax the card's performance to its fullest extent. What I don't understand is the radical difference in both performance and operating temperature between the two platforms. My questions are:
1) Do these operating characteristics look "normal" for other users of this setup (Mac Pro 2019/AMD RX6900XT)?
2) If not, what is your experience? And what do you think might account for that difference?
3) Can anyone suggest ways in which I might lower the card's temperatures?
4) Is there any way I might improve the card's performance?
5) Do you know of any other cross-platform benchmarks that can be compared head-to-head?
Any other words of wisdom, perspective or just general hob-nobbery on this subject would be very welcome.
In any case, thank you for the time and energy you spent reading this rather lengthy post!
Last edited:
