2013 13" MacBook Pro [4th,2C,U] + RX 5700 XT @16Gbps-TB2>TB3 (Razer Core X) + macOS 10.15.1 [vonthing]
- MacBook Pro (Retina, 13-inch, Mid 2014)
- 3 GHz Dual-Core Intel Core i7
- 16 GB 1600 MHz DDR3
- Sapphire Pulse Radeon RX 5700 XT 8 GB
- Crucial 1TB SSD NVMExpress @ x4 link width
- Pop the card into the enclosure
- (Only for Thunderbolt 3->Thunderbolt 2 setups) Disable System Integrity Protection and install purge-wrangler
Card recognized alright..
Then I score a 246 on the FurMark test.. I mean.. what.. the.. fuck?
I did a RX 590 build (thread here: https://egpu.io/forums/builds/2013-13-macbook-pro-4th2cu-xfx-rx-590-sapphire-bios-modded-16gbps-tb2tb3-razer-core-x-macos-10-15-1/ ) and that scored a 1170.
Again what the fuck? Two things are of suspect here:
- I'm running a PCIe riser this time around and it's causing some signal loss leading to bus faults dropped frames and all that.
- macOS 5700 XT (AMD Navi) drivers are straight up dog shit.
I'll redo the test with the riser out (I got the riser as a way to quickly switch between the Razer Core X and my PC laptop setup) but if the results turn out the same I see zero value in this upgrade.
Other than that; the Sapphire card is way more stable than the MSI card; no more kernel panics galore, can boot with the card plugged in, hotplug works, can even dirty hotplug and recover without a kernel panic. But with these kind of numbers I don't see any value in this upgrade what so ever and it seems like I'll return this card and go back to the RX 590 setup pretty soon.
Thanks for reading, and if you have any questions feel free to reach out!
Quick Update for the FurMark result fiasco
At first I was completely dumbfounded by the fact that a newer generation card priced 5 times higher, benchmarked 5 times worse compared to a card based on an architecture that's been around for three years (that's like 30 in gamer years!) so I redid the FurMark test with only the card changed which means:
- I am running the exact same operating system (macOS 10.15.1 Catalina)
- I am running the exact same PCIe riser cable (bought off Amazon)
- I am running the exact same binary for the benchmarking tool (seriously I hot-plugged the 5700 XT out, swapped the RX 590 back in -- didn't even
cdout of the directory I was in, on the shell I was logged in!)
- I ran the test with the exact same command line arguments
- Quite literally everything is the same, except one card is RX 5700 XT while the other card is RX 590.
And <drumroll here> here's the result I got:
The only words out of my mouth was holy fuck. A newer generation card priced 5 times as high, is beaten by an older generation card by almost an order of magnitude.
Well I mean, AMD has always been the superior hardware manufacturer, they beat Intel to 1GHz back in the day, they beat Intel to 64-bit again back in the day (Intel had its own, very short lived 64 bit architecture called Itanium but it had literally zero backwards compatibility so they ended up ditching it and just adopting AMD's x86-64 design) while Intel had the superior marketing, focused on flashy features (ray tracing) and it's kind of our generation's Edison vs Tesla analogy. AMD has always been lacking in the driver department but this kind of incompetency.. I could only say holy... fuck... and that's the end of this little story.
I don't know how you Boot Campers are doing out there but if you're planning to run macOS exclusively I suggest you stick to the RX 590, it is much cheaper and thanks to how bad the Navi platform (5700 XT) macOS drivers are, it's much faster as well. Apparently the drivers matter a lot -- these ones are such dog shit that an older, cheaper card beats a newer and much more expensive card by almost an entire order of magnitude.
Well that is all the testing on my plate for tonight. BRB printing the return label for this thing.
I don’t think that concluding GPU performance solely on one benchmark highlights the full picture (nonetheless result is alarming). Try comparing the XT vs the 590 on Unigine Valley and other benchmarks and see if you can reproduce this anomalous behavior (and if you search the forums a little there is plenty of evidence that the XT is indeed much faster vs. Polaris). Inversely, other 5700XT users on macOS should test with FurMark to verify if your problem is reproducible. Example of another benchmark that has been spitting out troubling results: Geekbench.
"Desultory reading is delightful, but to be beneficial, our reading must be carefully directed." — Seneca
Author: kryptonite ✧ purge-wrangler ✧ tbt-flash ✧ purge-nvda ✧ set-eGPU
Insights Into macOS Video Editing Performance
Launching Apps on Specific (e)GPUs on macOS
Definitive macOS Thunderbolt eGFX Compatibility Registry
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro
All the testing done by Barefeats, as well as the Basemark GPU Metal benchmark results, completely disagree with the original post.
The 5700XT is much faster.