I extensively benchmarked my 2018 Mac Mini+eGPU against my 2009 Mac Pro

[tu2thepoo]

Hi there! It turns out keeping a 2009 Mac Pro running can be a lot of busy work (as 25 pages of people who can't read instructions will attest)- especially since I have a Geforce 1080 and modern NVIDIA cards are no-go in Mojave until further notice. So I figured it'd be a good time to get the 2018 Mac Mini! There were a few builds on eGPU.io that suggested it would be simple to run my Geforce 1080 via eGPU in Win10 since I wouldn't be feeding video back through TB3 to the iGPU.

(I ran into electrical issues at my new apartment (AFCI circuits don't play well with my TV/Geforce/computer), so it ended up taking longer than planned to get everything running. But that's a story for another thread.)


I bought the store model and went with the base 256gb SSD since I thought I could easily run everything off SSD via external enclosure. My final hardware list was as follows:

Spoiler

• 2018 Mac Mini - 6 core CPU, 3ghz base speed (4ghz speed boost). 8 gb RAM, 256gb SSD
• Geforce 1080
• Sonnet TB3 Breakaway Box
• TCL 4k 43" TV connected via HDMI
• Terra Master D5-300C 5-bay USB3.1 enclosure
• Blu Ray drive via OWC USB3 5.25" enclosure
• Anker USB3 hub connected via USBA-USBC cable

My Mac Pro 2009 for comparison (note the video card and SATA SSDs were moved to my Mac Mini:

Spoiler

• 2009 cMP 4,1->5,1, firmware v140
• 6 core 3.46ghz Westmere Xeon
• 16gb RAM
• Geforce 1080
• 1tb SATA SSD boot drive

Here's how it looks:


How things feel: I kept a running list of notes comparing the user experience between the two machines. TLDR - almost everything felt just as fast if not significantly faster than my 2009 Mac Pro:

Spoiler

• As expected, the new machine is much less of a hassle to set up and keep running than the 10 year old MP. If you want things to Just Work, spend the money.
• The internal SSD is stupid fast, no SATA SSD compares (I haven’t used any PCIe SSDs before this). Everything from boot to file copies to opening programs felt faster, even after copying all of my files/programs over.
• The 4k TV works fine connected to HDMI, no fiddling needed to enable HiDPI mode. For what it’s worth, Win10 detects the setup as HDR-capable, MacOS shows it supporting a 30-bit framebuffer.
• Filevault enables instantly on the internal SSD, as expected with the T2 chip. External drives are still encrypted the old-fashioned way (progress meter/writing to drive).
• Google Maps and other WebGL-rich apps are definitely slower/laggier in 4k than the Geforce 1080 in High Sierra. I don't know why I was surprised, but I just figured integrated graphics would've been more than fast enough at this point.

Boot Camp notes: TLDR - boot camp works fine via internal SSD - if you need to boot via external drives or run games from USB it may get hairy. Get a larger internal SSD if you have the money.

Spoiler

• T2 defaults to "secure boot" mode, which allows Win10 boot camp only on the internal SSD. External booting/Linux/etc needs to be toggled in Recovery Mode.
• If installing on the internal SSD, Boot Camp Assistant partitions the drive and automatically copies the Win10 ISO to an empty portion of the SSD (I'm assuming). No USB drive needed.
• Don't use Bitlocker if you're using Boot Camp on the internal SSD - Win10 creates extra partitions and renders the Win10 portion unbootable.
• This may be just a quirk of my USB-C enclosure, but I had a lot of instability running my Steam games from my USB3.1 enclosure. I ended up wiping the internal SSD completely and doing a clean install of Win10 (booting MacOS Mojave from the USB enclosure). Your mileage may vary but I had no problems after doing things this way.

eGPU notes: TLDR - NVIDIA still doesn’t work (for the most part) in MacOS Mojave. Win10 works fine.

Spoiler

• The Sonnet Breakaway Box 350w is more than powerful enough for my Geforce 1080. You can upgrade the power supply if you want to, I ended up buying a Corsair SF600 power supply to troubleshoot the electrical issues I mentioned above.
• Once you install the NVIDIA drivers it is literally plug and play in Win10. I found the best method was 1) boot via iGPU 2) turn on eGPU 3) install drivers 4) shut down and disconnect monitor from iGPU 5) power back on.
• The iGPU disables itself if you boot with only eGPU connected to the monitor, but Intel Quicksync still works.
• No bootscreens via eGPU, at least on my Geforce 1080

Overall benchmark notes: I used Geekbench, GFXbench, and a few synthetic GPU benchmarks to see what I was gaining/losing. Overall, Mac Mini+eGPU was a significant win for video transcoding and gaming. Mac Pro won in some synthetic benchmarks but lost in most real-world situations (at least, in my use case).


I’ll be using these abbreviations:

• MP5,1MacOS = 2009 Mac Pro in Mac OS High Sierra 10.13.6
  
• MP5,1Win10 = 2009 Mac Pro in Win10 v1809, installed on dedicated SATA SSD
  
• Mini8,1Win10 = 2018 Mac Mini in Win10 v1809, installed on internal SSD and using Geforce 1080 in eGPU

(I skipped testing the Mini in MacOS since the CPU scores were similar on the Mac Pro, and I didn’t really care how fast/slow the iGPU is)


Geekbench (CPU) https://www.geekbench.com: The Mini is faster in single and multi-core (assuming you’re not running a 12 core Mac Pro).

Spoiler

• MP5,1MacOS = Single Core 3136 Multi Core 13432
  
• MP5,1Win10 = Single-Core 2936 Multi-Core 13462
  
• Mini8,1Win10 = Single-Core 5049 Multi-Core 19573

GFXbench (GPU) https://gfxbench.com: The Mac Pro + Geforce 1080 has a higher average/peak frame rate than the Mac Mini + Geforce 1080.

Spoiler

• MP5,1MacOS = Aztec Ruins 1080p: 154 fps 1440p: 280 fps
  
• MP5,1Win10 = Aztec Ruins 1080p: 186 fps 1440p: 413 fps
  
• Mini8,1Win10 = Aztec Ruins 1080p: 74 fps 1440p: 176 fps

Unigine Heaven (GPU) https://benchmark.unigine.com/heaven: Same as above, but with the note that the Mac Pro had a lot of stutter/jerkiness compared to the Mac Mini (see the minimum fps). This was reproducible across several test runs - pretty odd. Heaven is a pretty old benchmark so I’d take the results with a grain of salt.

Spoiler

• MP5,1MacOS = Score 2038 Avg FPS 80.9 Min FPS 9.2 Max FPS 168.5
  
• MP5,1Win10 = Score 3248 Avg FPS 128.9 Min FPS 9.3 Max FPS 256.0
  
• Mini8,1Win10 = Score 2713 Avg FPS 107.7 Min FPS 33.8 Max FPS 221.3

3DMark (GPU) https://www.3dmark.com: honestly, it’s not a super relevant benchmark (no games are based on it) but it’s useful for showing that GPU-heavy applications can narrow the gap between internal PCIe and eGPU considerably.

Spoiler

• MP5,1Win10 = Graphics Score 7044 Avg FPS test 1 45.5 Avg FPS test 2 40.7
  
• Mini8,1Win10 = Graphics Score 6562 Avg FPS test 1 42.8 Avg FPS test 2 37.6

Superposition (GPU) https://benchmark.unigine.com/superposition: another synthetic GPU benchmark, but pretty new (released in the past year), and very similar in graphics to most 2017-2018 games. It also has several detail settings that let me tease out the differences between PCIe and eGPU further:

Spoiler

MP5,1Win10

• 1080p “high” detail: 10989 (Avg FPS 82.2 Min FPS 64.9 Max FPS 109.9)
  
• 1080p “extreme” detail: 4030 (Avg FPS 30.1 Min FPS 24.7 Max FPS 35.1)
  
• 4k “optimized” detail: 6523 (Avg FPS 48.8 Min FPS 37.7 Max FPS 62.5)

Mini8,1Win10

• 1080p “high” detail: 10248 (Avg FPS 76.7 Min FPS 60.7 Max FPS 96.1)
  
• 1080p “extreme” detail: 3917 (Avg FPS 29.3 Min FPS 24.2 Max FPS 33.5)
  
• 4k “optimized” detail: 6242 (Avg FPS 46.7 Min FPS 37.2 Max FPS 58.5)

I tested the machines in my two favorite games right now, XCOM2 and Battletech. Neither have built-in benchmark options, but XCOM2 becomes very CPU-bound when you’re in the late game (especially if you have lots of Steam Workshop mods installed). Using the Steam built-in FPS counter, in XCOM2 the Mac mini was either the same as the Mac Pro or 15-20 fps faster, depending on the scene. FPS were basically the same in Battletech. I attached some screenshots and video capture so you can compare to whatever games you run:

Spoiler

• XCOM2 video capture: https://gfycat.com/HardShockingBeagle
• 
• 
• 

Lastly, I tested both machines in Handbrake. I rip my blu-rays to H.264/H.265, so I wanted to see what differences there were. The MP5,1 turned out to be faster in H.264, but the Mac mini was much faster in H.265 and - in windows - intel quicksync lets me play games while transcoding. The default quality isn’t as good, so I only use quicksync for extras/outtakes; still though, that frees up my computer when it would otherwise be chugging away on an encode.


The difference in x264 speed may also be due to testing in MacOS vs Win10. Whatever!

Spoiler

• Source file here: http://jell.yfish.us - jellyfish-40-mbps-hd-h264
  
• H.264 preset: Apple 1080p30 surround
  
• H.265 preset: H.265 MKV 1080p30
  
• Quicksync: Apple 1080p30 surround but using Quicksync instead of x264

MP5,1MacOS

• H.264 “work: average encoding speed for job is 40.475502 fps”
  
• H.265 “work: average encoding speed for job is 5.707329 fps”

Mini8,1Win10

• H.264 “work: average encoding speed for job is 30.362831 fps”
  
• H.265 “work: average encoding speed for job is 8.764733 fps”
  
• Quicksync H.264 “work: average encoding speed for job is 38.119728 fps”

Power/heat: the Mac Mini draws significantly less power (60-80w total under load) but gets significantly hotter than the Mac Pro (90-100C). This affects sustained boost speed - when using CPU transcoding I saw the CPU hit its max power limit and throttle back, resulting in a see-saw pattern between 3ghz to ~3.8ghz in the Intel Power Gadget:


Using Quicksync (or eGPU) resulted in sustained boost speeds of 3.9ghz/4ghz without any significant throttling:


Takeaway: Offload tasks to GPU if possible. If something is CPU-only, you should only expect the base 3.0ghz clock speed to be sustained. Mixed-use will result in see-sawing CPU/GPU frequencies as the Mini keeps itself within temperature/power limits:



Also, the exhaust gets stupid hot on the Mac mini, especially if you have lots of stuff plugged in. I ended up mounting a small fan above the Mini (you can see it in the picture above) - it doesn't really affect CPU/GPU temperatures but it helps keep the cables/plugs from getting hot.

 

[jinnyman]

Thank you for the great info! I'm really surprised by how much Mac mini can do.

 

[tonyunreal]

Great post! Could you add the resolution and preset you were using with Unigine Heaven to your post? I have Sapphire Vega 64 with the Razer Core X enclosure, and scored 2083 / 82.7fps (32.9 min ~ 168.5 max) at 1920x1080 8xAA / ultra / extreme.

 

[tu2thepoo]

Great post! Could you add the resolution and preset you were using with Unigine Heaven to your post? I have Sapphire Vega 64 with the Razer Core X enclosure, and scored 2083 / 82.7fps (32.9 min ~ 168.5 max) at 1920x1080 8xAA / ultra / extreme.

I think it was the default 1440p extreme. Maybe it’s 1600x900? I’ll need to double check.

[edit]
I left it at default, so it would be 1600x900 8xAA, I believe.

 

[tu2thepoo]

I re-tested Handbrake under MacOS with the Mac Mini for completeness sake (also because @pl1984 asked ) results as follows:

x264 "work: average encoding speed for job is 29.049265 fps"
x265 "work: average encoding speed for job is 8.390409 fps"

So about the same as Win10 results. Possible explanations?

1) x264 really likes having the 6 core/12 threads on the Westmere CPU vs just the 6 cores of the Coffee Lake i5
2) I'm not using the same versions to test (forgot to write down what version of HB the Mac Pro was running)
3) who knows?

 

[pl1984]

I re-tested Handbrake under MacOS with the Mac Mini for completeness sake (also because @pl1984 asked ) results as follows:

x264 "work: average encoding speed for job is 29.049265 fps"
x265 "work: average encoding speed for job is 8.390409 fps"

So about the same as Win10 results. Possible explanations?

1) x264 really likes having the 6 core/12 threads on the Westmere CPU vs just the 6 cores of the Coffee Lake i5
2) I'm not using the same versions to test (forgot to write down what version of HB the Mac Pro was running)
3) who knows?

The macOS version of Handbrake does not support Quicksync. Given the IPC improvements from 2009 up to now the results seem reasonable. Thanks for running the test.

 

[F-Train]

Hi @tu2thepoo,

If you are using a Mac mini plus an external GPU enclosure and an Nvidia card to run applications, could you explain how you are doing it and say what applications you are running this way?

Thanks

 

[tu2thepoo]

If you are using a Mac mini plus an external GPU enclosure and an Nvidia card to run applications,

I'm not using the Geforce eGPU in MacOS, just in Windows 10 - in MacOS I'm using the intel GPU.

 

[F-Train]

I'm not using the Geforce eGPU in MacOS, just in Windows 10 - in MacOS I'm using the intel GPU.

Thanks, I'm just trying to understand the connection between the Mac mini and this thread. Are you saying that these tests did not involve using a Mac mini with an Nvidia card at all?

 

[tu2thepoo]

Thanks, I'm just trying to understand the connection between the Mac mini and this thread. Are you saying that these tests did not involve using a Mac mini with an Nvidia card at all?

If you mean “did I use the GeForce in MacOS Mojave?” then no, and I stated that clearly in the post. The Geforce was used for all testing in windows 10 and the MacOS testing was all CPU benchmarking. I’m not sure if you skipped that part or are just disappointed that I didn’t write an NVIDIA driver for 10.14.

 

[F-Train]

If you mean “did I use the GeForce in MacOS Mojave?” then no, and I stated that clearly in the post. Otherwise I’m not sure if you’re being intentionally dense or just disappointed that I didn’t write an NVIDIA driver for 10.14.

Huh?

I asked whether you used a Mac mini with an eGPU enclosure and an Nvidia card. If so, I asked how you did it and which applications you ran. I didn't say anything about operating system.

The question is dead simple, and I've now asked it three times.

 

[tu2thepoo]

Huh?I asked whether you used a Mac mini with an eGPU enclosure and an Nvidia card. If so, I asked how you did it and which applications you ran. I didn't say anything about operating system.The question is dead simple.

Then you didn’t read my post closely enough. I clearly marked which test result came from each combination, whether it be MP5,1+GeForce 1080, Mac mini in MacOS, or Mac Mini+GeForce in Windows 10. Try clicking the “spoiler” buttons and you’ll see your questions were already answered

 

[F-Train]

Then you didn’t read my post closely enough. I clearly marked which test result came from each combination, whether it be MP5,1+GeForce 1080, Mac mini in MacOS, or Mac Mini+GeForce in Windows 10. Try clicking the “spoiler” buttons and you’ll see your questions were already answered

I actually don't understand your posts, which is why I asked the question in the first place, except that I know that part of my question isn't even touched on.

Could I now ask for the fourth time, and just get an answer. If you don't want to answer, that's cool, just say so, or just don't respond. This is now getting bizarre.

 

[tu2thepoo]

I actually don't understand your posts except that I know that part of my question isn't even touched on. Could I now ask for the fourth time, and just get an answer. If you don't want to answer, that's cool, just say so, or just don't respond. This is now getting bizarre.

I genuinely don't understand if you're skipping the part where I said what program was running on what GPU or what:



So Superposition was running on the Mac mini, using Windows 10, with the Geforce 1080 in the Sonnet eGPU box. Same goes for every other program I listed. Do the "spoiler" buttons not show up for you to click?

As for how I did it, here's another screenshot of the first post, if that helps:


I literally don't know what you're asking that those didn't answer.

 

[F-Train]

I genuinely don't understand if you're skipping the part where I said what program was running on what GPU or what:

So Superposition was running on the Mac mini, using the Geforce 1080 in the Sonnet eGPU box. Same goes for every other program I listed. Do the buttons not show up for you to click? You're literally the only one in the thread who doesn't seem to understand what I wrote in the post.

Why is this so hard? How were your running Superposition on the mini with an external enclosure and Nvidia card?

Have you run any actual applications this way, and if so, can you reveal which ones?

Let me tell you something about your posts. They are not exactly a model of clarity, I have no interest in your "buttons", and if I want a maze I can sure find more entertaining ones.

I just had a couple of questions. If you don't want to answer them, preferably in something approximating plain English, that is just fine. Having now asked the same questions five times, I'm sure not going to waste anymore time on this.

 

[tu2thepoo]

Why is this so hard? How were your running Superposition on the mini with an external enclosure and Nvidia card?

The Geforce is plugged into the eGPU enclosure. I installed the Windows 10 drivers. I powered off the computer. I plugged the TV into the Geforce. I rebooted. I double clicked the icon for the application Superposition. It did its thing. I wrote down the score. I wrote the post. I put all the detailed stuff in the "spoiler" buttons so it wouldn't be a giant wall of text. Other people figured out that the "spoiler" buttons had more info. They clicked those buttons and read the extra info if they wanted to. I didn't realize you had an objection to spoiler buttons. I really do apologize, I thought you were yanking my chain for some reason.

Have you run any actual applications this way, and if so, can you reveal which ones?

Are you asking if I've run like, Microsoft Word? Then yes. Everything you'd normally run, works perfectly fine using an eGPU in Windows 10. In MacOS, my Geforce isn't supported, so I have not run stuff on the eGPU.

 

[pl1984]

I actually don't understand your posts, which is why I asked the question in the first place, except that I know that part of my question isn't even touched on.

Could I now ask for the fourth time, and just get an answer. If you don't want to answer, that's cool, just say so, or just don't respond. This is now getting bizarre.

He answered your question in post #8. I didn't have any issues understanding the answer nor his subsequent ones. The original post explained everything in great detail and in a very clear manner.