Big Sur seems to be slower than Catalina in Geekbench Compute @ Metal
I just retested Geekbench with my Blackmagic eGPU Pro and granted, it's different host (Macbook Pro i9 vs Mac Mini i7) and it's slightly different revision of Geekbench, but the difference in performance is staggering:
Canny 80273 in Catalina vs 43073 in Big Sur, wtf?
Did you observe similar difference between Catalina and Big Sur with your setup?
@oo-viper-oo, somebody else in the forum has a thread about Big Sur performance and he said that Big Sur was better than Catalina.
A) 2020 MacBook Pro, i7-1068NG7, 32GB RAM, 1TB, EGPU Razer Core X, Nitro+ RX5700 xt 8Gb, Samsung 49 1440p UltraWide C49RG
Mac OS Catalina 10.15.7, Internal Bootcamp Windows 10 2004 pci.sys swap.
B) 2.7 GHz I7 4 Cores, 16Gb, 1TB MBP 13 2018 TB3 , EGPU Razer Core X, Nitro+ RX5700 xt 8Gb, LG 32UK550
Mac OS Catalina 10.15.2, Ext SSD Windows 10 1903 V1 .295
(Unless Geekbench fixed their bug on Mac OS) you will need to attach (2) or more displays to the Mini/eGPU to see a higher score.
Ok, so I re-tested it with Mac Mini and I see the same drop as on Macbook Pro. Now it's the Apples to Apples, more less, apart from minor version difference in Geekbench itself, and then, of course, the os version.
I will try the displays trick as pointed out by @mini-i5 above (thank you!). To be frank I do have 5K and 4K monitors plugged into my machine normally, but I'm not really sure whether I had them in back then, when Vega achieved 71102 and I had just one connected to eGPU now, when I re-tested it, as I thought display would cause unnecessary load on the eGPU and lowering the score.
If it's such stupid bug/incompatibility, then I'm fine. I was more worried that maybe Vega drivers in Big Sur were f\@\@ked up.
Ok, I have retested it with two monitors (5k and 4k) attached to eGPU, one via thunderbolt, the other via displayport. And the results under Big Sur with last Geekbench stay around 53-54k ( https://browser.geekbench.com/v5/compute/1965136). So I conclude that those 71102 generated previously were artificially inflated by Geekbench.