[Solved] AMD Navi 7nm for eGPU - Radeon RX 5700 & RX 5700 XT  

Posted by: @itsage

@desmo_duo For non-gaming use the Radeon VII is much more powerful than the rest of Radeon cards. The best enclosure for it is NetStor HL23T-Plus. I've been using this exact same setup with my 2013 Mac Pro and it's working great [needs Purge-Wrangler script]. Apple will keep improving drivers for Vega II because they use it in the new Mac Pro.

 

@itsage  Thx mate! Netstore seems to be unavailable in Germany. What about Razer Core X? And XTX AMD Radeon VII?

Sorry guys for beeing off-topic

@desmo_duo Some members reported proper sleep behavior using the Core X with Radeon VII so you can certainly give it a shot.

@jefniro

I've run Geekbench 5 OpenCL test on Windows 10 bootcamp and I can get the same result as yours: around 60's. (see attachment)

Very odd.

Do anyone have an idea about what can be happening? Why my macOS Geekbench 5 OpenCL score is about 50% of what it should be?

@precision01 Something is definitely off with your macOS RX 5700 XT performance. I tried it on a reference RX 5700 just now and OpenCL was 49548 while Metal was 42447.

@itsage @jefniro

Very strange indeed.

The tests were done on a reference XFX Radeon RX 5700 XT (RX-57XT8MFD6).

I'm getting the same low Geekbench 5 score on MacBook Air 2019 (about 28000). So, I think it is not the MacBook itself. Very strange.

And Metal scores follow the same trend: ~50% lower Geekbench 5 metal score.

So, it must be Razer Core X interface or the GPU. I'll google about XFX 5700 XT, geekbench, Razer Core X and related topic to try to find out what is happening, since unfortunately I don't have another 5700 XT to make comparisons.

update:

It shouldn't be, but maybe the issue is related to thunderbolt 3 cable. it shouldn't because it is an Active 40gbps 100w 2 meters (~6.5 feet) thunderbolt 3 cable from CalDigit. But I will try with the 0.5 meter cable from Razer.

@precision01

Interesting discrepancy. I have my RMAed 5700 XT arriving today. Will run a bench to add a data point here. Mine is a PowerColor variant. Let’s see what we get.

Posted by: @precision01

@itsage @jefniro

Very strange indeed.

The tests were done on a reference XFX Radeon RX 5700 XT (RX-57XT8MFD6).

I'm getting the same low Geekbench 5 score on MacBook Air 2019 (about 28000). So, I think it is not the MacBook itself. Very strange.

And Metal scores follow the same trend: ~50% lower Geekbench 5 metal score.

So, it must be Razer Core X interface or the GPU. I'll google about XFX 5700 XT, geekbench, Razer Core X and related topic to try to find out what is happening, since unfortunately I don't have another 5700 XT to make comparisons.

update:

It shouldn't be, but maybe the issue is related to thunderbolt 3 cable. it shouldn't because it is an Active 40gbps 100w 2 meters (~6.5 feet) thunderbolt 3 cable from CalDigit. But I will try with the 0.5 meter cable from Razer.

 

I use Apple Thunderbolt 3 cable, the one 0,8 m. length.

What I need to say is that my Wavlink  TB3 external SSD disk performances dropped! 😕 

Before (10.15.1):

After (10.15.2):

@precision01 Allow me to add some more confusion to this mix. Same macOS build. Same GPU (different AIB/model). Yet another different set of scores. I tried multiple runs of course.  I have a feeling that Geekbench is being inconsistent. That or GPU BIOS makes a difference. I've had issues with Geekbench GPU benchmarking before - scores have varied widely (notwithstanding version changes that change normalization) on minor macOS updates even on Vega (same card). LuxMark is usually far more consistent. This is not even margin of error:

@mac_editor @precision01 here my new LuxMark benchmark. The same result that I had wit 10.15.1:

I have some gain with Valley benchmark (ca. 10%) from previous version of Catalina. But I think Geekbench has some bugs...

So the new 16" MacBook Pro is using AMD Radeon Pro 5000M series GPUs.  RDNA/Navi confirmed.