Low Bandwidth (Host to Device)
Hello, I have a question for you experts. I searched the forum, but perhaps because of my poor English, I have not found an answer to my curiosity.
I noticed that if I turn on my MacBook Pro 16 with the egpu (Razer Core X + Gigabyte Radeon 5700 XT) already connected, the bandwidth (Host to Device) that measures the program CL!ng is only 15-16 Gbit/s. While if I connect the egpu when the system is already started (for example when the login screen appears) then CL!ng measures 20-21 Gbit/s, which is absolutely correct.
However, trying to make quick benchmarks there is no substantial difference. On the contrary, paradoxically, the results are slightly higher with lower bandwidth.
Geekbench gave me these results:
20-21 Gbit/s bandwidth: Metal 64176, OpenCL 59282
15 Gbit/s bandwidth: Metal 65065, OpenCL 59609
Has anyone already noticed the same thing?
Maybe lower H2D means lower latency? Probably needs testing with other eGPUs with various firmwares that have different H2D.
Hello, thank you very much for your answer. Unfortunately I don't have any other EGPUs available, with other controllers and other firmware versions. Maybe someone else can help us out.
What I would be most interested to know is also if someone noticed a bandwidth difference between powering up with EGPU already connected and EGPU connected only with computer on.
And if someone else noticed it, do you know why this happens?
It is pure curiosity. Also because, as the benchmarks show, you don't actually notice differences.
Thank you very much
PS: I must perhaps add that my detections are done in MacOS Big Sur 11.0.1
I started Cl!ng for the first test [MTL GPU]. Then quit. Chose "Disconnect AMD Radeon RX 580" from the eGPU menu, unplugged the Thunderbolt cable from the OWC Mercury Helios, then reconnected it, then started CL!ng again.
H2D went from 16 Gbps to 20.89 Gbps.
D2H remains at around 21.84 Gbps.
Then I restarted.
H2D went back down to 16 Gbps.
D2H remains at around 21.5 Gbps.
D2D is between 700 and 1156 Gbps in either case.
Hi Joe, thanks a lot for playing the test with your Mac mini.
I notice that I'm not the only one who measures a lower bandwidth by connecting the EGPU before turning on the computer.
Who knows what this is due to? However, on the benchmark front, as mentioned, nothing changes. At least according to the measurements made with Geekbench.
In the next few days I will try other software. Advice?