2019 16" MacBook Pro (RP5600M) [9th,8C,H] + RTX 3090 @ 32Gbps-TB3 (Mantiz Saturn...
 
Notifications
Clear all

2019 16" MacBook Pro (RP5600M) [9th,8C,H] + RTX 3090 @ 32Gbps-TB3 (Mantiz Saturn Pro) + Win10 [bmagnien]  

  RSS

bmagnien
(@bmagnien)
Trusted Member
Joined: 2 months ago
 

System specs 

Apple MacBook Pro 16”

  • i9 9980HK 8 core 2.4 Ghz @ 3.5ghz 
  • 32 GB 2666MHz DDR4 memory
  • 1TB NVME SSD
  • AMD 5600m dGPU, 8GB HBM2 memory
  • LaCie 2TB Rugged SSD Pro TB3 External
  • Apple TB3 Pro Cable (2m)
  • Razer Blackwidow TE Chroma v2 TKL, Orange Switches
  • Razer Deathadder v2
  • Bootcamp Windows 10 Pro, 1903

Acer Predator X35
35” 3440x1440 180hz DisplayHDR1000 Curved Ultrawide

eGPU hardware

EVGA RTX 3090 FTW3 Ultra

24GB GDDR6X memory, OC vBios, Triple fan cooler, 3x8pin

Mantiz Saturn Pro II eGPU enclosure

Modifications: 

  • Top-mounted 120mm x15mm exhaust fan (Noctua NF-A12x15 PWM chromax.black.swap)
  • Side-mounted 120mm x 25mm intake fan (Noctua NF-F12 iPPC-2000)
  • Reversed PSU to intake off GPU backplate
  • 1TB Samsung EVO 860 2.5” SATA SSD

Hardware pictures

Top-mounted exhaust fan installed by having a local metal shop drill a circular cutout and 4 pass-through screw holes sized to match the steel mesh grill and case fan screws. Fan grills are used internally on the top exhaust and side intake to prevent wires from catching fan from the interior. PSU intake does not use an internal fan grill as it's nearly touching the backplate of the GPU with no risk of wire interference, and wanted maximum airflow there as it's the hottest part of the unit. Initial top-mounted exhaust fan sizing test was done with a standard Noctua 120x15 (pictured), which I replaced with a chromax black as you could see the white through the black steel grill. Also the side fan pictured here is the original Mantiz side fan, I've since replaced it with the Noctua NF-F12 listed above. Both Noctuas run off a single PWM-controlled header on the RTX 3090, and are individually controlled along with the 3 GPU shroud fans via EVGA Precision X1. GPU temps max around 66c at 100% fans, which is LOUD but I needed the extra cooling to push OCs for stability in these benchmarks. I run at auto fan curve while gaming with a more reasonable OC and the setup is quieter than the fans on my MBP

1
2
3
7
6
4
8
5
9

 

Installation steps

  1. Boot into Bootcamp, allow complete startup to desktop
  2. Plug TB3 cable into right side MBP (already connected to eGPU)
  3. Have all peripherals plugged into eGPU (mouse, keyboard, laptop cooler) and DP cable out of GPU directly connected to powered-on external monitor
  4. [Optional] LaCie TB3 external SSD plugged into left side MBP (confirmed full speed ~ 1500mbs write, 2500mbs read while eGPU running)
  5. SATA eGPU SSD disabled in Windows (only used as backup device when I use the eGPU as a glorified dock in OSX, split into two partitions for Time Machine and WinClone Bootcamp backup) so SATA doesn't take any bandwidth away from eGPU in Bootcamp
  6. Turn on eGPU
  7. All peripherals power on and become usable
  8. Manually download and install latest Nvidia drivers (not via Nvidia Experience) as I have all automatic Windows Updates/Driver downloads disabled
  9. No need to uninstall AMD drivers or disable dGPU
  10. Works like a charm
  11. To turn off: Select 'Shutdown' - after MBP shuts down, turn off eGPU.
  12. Boot back up and either log into OSX or repeat steps 1,2,3,6 above 

Benchmarks
There will be no data on internal monitor performance as I have no interest in utilizing this eGPU in that manner.

CUDA-Z: Host to Device: 2259.5MiB/s

CUDA Z 3090

Synthetic Benchmarks

3DMark:

Port Royal

(eGPU 3090) / (avg. Desktop 3090*)
Score: 10569 / 13265*
eGPU % Difference: -20%
Mem Offset: 1200
Core Offset: 85

PortRoyal


TimeSpy
 

(eGPU 3090) / (avg. Desktop 3090*)
Graphics Score: 17886 / 19592*
eGPU % Difference: -9%
Mem Offset: 800
Core Offset: 55

TS


TimeSpy Extreme

(eGPU 3090) / (avg. Desktop 3090*)
Graphics Score: 10202 / 10182*
eGPU % Difference: +<1%
Mem Offset: 1200
Core Offset: 85

TSE


Fire Strike Extreme

(eGPU 3090) / (avg. Desktop 3090*)
Graphics Score: 17209 / 24095*
eGPU % Difference: -29%
Total Score: 15850 / 21423*
eGPU % Difference: -26%
Mem Offset: 1200
Core Offset: 75

FSE


Fire Strike Ultra

(eGPU 3090) / (avg. Desktop 3090*)
Graphics Score: 10280 / 12492*
eGPU % Difference: -18%
Total Score: 10426 / 12489*
eGPU % Difference: -16%
Mem Offset: 1200
Core Offset: 75

FSU

 

3DMark Comparables: 3dmark.com/search

SuperPosition:

1080p Extreme

(eGPU 3090) / (avg. Desktop 3090**)
Score: 12468 / 12452**
eGPU % Difference: +<1%

SP 1080pExt


4k Optimized

(eGPU 3090) / (avg. Desktop 3090**)
Score: 16464 / 16597**
eGPU % Difference: -<1%

SP 4kOpt


8k Optimized
(eGPU 3090) / (avg. Desktop 3090**)
Score: 7645 / 7285**
eGPU % Difference: +5%

SP 8kOpt

**Superposition Comparables: https://www.overclock3d.net/reviews/gpu_displays/msi_rtx_3090_gaming_x_trio_review/19

Gaming Benchmarks:

For all games:

  • OC vBios on with additional +55 core offset &, +800 memory offset overclock (except for Metro High, Ultra, Extreme, and RTX presets which were only stable at +40/+600)
  • Vsync off
  • Gsync off
  • Full screen
  • 3440x1440
  • HDR1000, 10-bit color depth, full RGB
  • 144hz
  • Built-in benchmarks
  • Built-in named quality presets (except RDR2 as presets are broken, 2 custom combos below)
  • Average benchmark FPS listed, more details in screenshots

Shadow of the Tomb Raider

  • Highest Preset, DLSS On, RTX Off: 78 

     

     

     

     

     

     

    Shadow HighestPreset DLSS RTX OFF

     

  • Highest Preset, DLSS On, RTX Medium: 76

     

     

     

     

     

     

     

    Shadow HighestPreset DLSS RTX MED
  • Highest Preset, DLSS On, RTX High: 60 

     

     

     

     

     

     

    Shadow HighestPreset DLSS RTX HIGH
  • Highest Preset, DLSS On, RTX Ultra: 59 

     

    Shadow HighestPreset DLSS RTX Ultra

 

Metro Exodus

  • Low: 121

     

     

     

     

     

     

     

    Metro Low
  • Normal: 83

     

     

     

     

     

     

     

    Metro Normal

     

  • High: 67

     

     

     

     

     

     

     

    Metro High
  • Ultra: 53

     

     

     

     

     

     

     

    Metro Ultra
  • Extreme: 37

     

     

     

     

     

     

     

    Metro Extreme
  • RTX: 49

     

     

     

     

     

     

     

    Metro RTX

Forza Horizon 4

  • Very Low: 149

     

     

     

     

     

     

     

    ForzaVeryLow
  • Low: 127

     

     

     

     

     

     

     

    ForzaLow
  • Medium: 111

     

     

     

     

     

     

     

    ForzaMedium
  • High: 108

     

     

     

     

     

     

     

    ForzaHigh
  • Ultra: 93

     

     

     

     

     

     

     

    ForzaUltra
  • Every Setting Maxed: 76

     

     

     

     

     

     

     

    ForzaEverythingMax

Horizon Zero Dawn

  • Favor Performance: 79

     

     

     

     

     

     

     

    Horizon FavorPerformance
  • Favor Quality: 65

     

     

     

     

     

     

     

     

    Horizon FavorQuality
  • Ultimate Quality: 61

     

     

     

     

     

     

     

    Horizon UltimateQuality

Far Cry 5

  • Low: 117

     

     

     

     

     

     

     

    FC5 Low

     

     

  • Normal: 107

     

     

     

     

     

     

     

    FC5 Normal

     

  • High: 101

     

     

     

     

     

     

     

    FC5 High

     

  • Ultra: 94

     

     

     

     

     

     

     

    FC5 Ultra

     

Red Dead Redemption 2

Texture Quality Ultra, AF x16, MSAA Off, On/Off settings set to Off, Level of detail sliders set to midpoint

  • Quality settings set to High: 60

     

     

     

     

     

     

     

    RDR2 High

     

  • Quality settings set to Medium: 66

     

     

     

     

     

     

     

    RDR2 Medium

 

Additional Benchmark Comparisons (3080 vs 3090 vs Desktop vs eGPU):

Time Spy (1440p) Graphics Score

GPU

All Desktop (Average)

9980HK MBP w/eGPU

% Difference

GTX 980

4,365

4,154

-5%

GTX 1080

7,571

6,824

-9%

RTX 2080

11,116

9251

-17%

RTX 3080

17,551

15,234*

-13%*

RTX 3090**

19,710

17,886**

-9%

 

 *Intel Core i5-1035G7 Processor
**EVGA 3090 FTW3 Ultra

 

Time Spy Extreme (4K) Graphics Score

 

GPU

All Desktop (Average)

eGPU Sample (Best)

% Difference

RTX 3080

8,861

8,234

-7%

RTX 3090

10,185

10,202*

+<1%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra

 

Port Royal (Ray Tracing)

 

GPU

All Desktop (Average)

eGPU Sample (Best)

% Difference

RTX 3080

11,380

8,873

-22%

RTX 3090

13,248

10,569*

-20%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra

 

Firestrike Extreme (1440p) Graphics Score

 

GPU

All Desktop (Average)

eGPU Sample (Best)

% Difference

RTX 3080

20,966

15,213

-28%

RTX 3090

23,856

17,209*

-28%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra

 

Firestrike Ultra (4K) Graphics Score

 

GPU

All Desktop (Average)

eGPU Sample (Best)

% Difference

RTX 3080

10,775

9,496

-12%

RTX 3090

12,501

10,280*

-18%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra

 

Super Position (1080P Extreme)

 

GPU

Desktop Sample

eGPU Sample

% Difference

RTX 3080

10,722

10,389

-5%

RTX 3090

12,452**

12,468*

+<1%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra
**  https://www.overclock3d.net/reviews/gpu_displays/msi_rtx_3090_gaming_x_trio_review/19

 

Super Position (4k Optimized)

 

GPU

Desktop Sample

eGPU Sample

% Difference

RTX 3080

14,226

13, 391

-6%

RTX 3090

16,597**

16,464*

-<1%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra

 

Super Position (8k Optimized)

 

GPU

Desktop Sample

eGPU Sample

% Difference

RTX 3080

6453

--No-Data--

-N/A-

RTX 3090

7,285**

7,645*

+5%

 

*Intel Core i9-9980HK Processor, EVGA 3090 FTW3 Ultra

**  https://www.overclock3d.net/reviews/gpu_displays/msi_rtx_3090_gaming_x_trio_review/19

 

 

 

 

 

This topic was modified 3 weeks ago
2019 16" MacBook Pro (RP5600M) [9th,8C,H] + RTX 3090 @ 32Gbps-TB3 (Mantiz Saturn Pro) + Win10 [build link]  

itsage and nando4 liked
ReplyQuote
itsage
(@itsage)
Illustrious Member Admin
Joined: 4 years ago
 

@bmagnien, Thank you for taking the time to build and share results of your Mantiz Saturn Pro + EVGA RTX 3090 FTW3 Ultra eGPU setup. This should clear up doubts on Thunderbolt 3 bottlenecking high power GPUs. I wonder if the RTX 3090 FE cooling system is overwhelming the Desktop cooling in some of the comparable results. Your modifications to the Mantiz Saturn Pro definitely improved cooling for the graphics card too.

 

external graphics card builds
best laptops for external GPU
eGPU enclosure buyer's guide

 
2018 Mac Mini [8th,4C,B] + RX Vega 56 Nano @ 32Gbps-TB3 (Gigabyte Gaming Box) + macOS 11 & Win10 2004 [build link]  


bmagnien liked
ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 4 years ago
 

@bmagnien, thanks for sharing this build! Quick note about memory overclocks though - 3080/3090 have ECC so increasing clocks too much might actually cause performance loss - on non-ECC memory we would see the typical glitchy artifacting the display but with ECC it tries to correct and there is an FPS penalty.

 

purge-wranglertbt-flashpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance

Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro

 
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]  


ReplyQuote
bmagnien
(@bmagnien)
Trusted Member
Joined: 2 months ago
 

@mac_editor, all of the clocks listed were arrived at by starting at stock and ramping up until performance stopped increasing or the application became unstable. They were calculated per application. Would you recommend I handle it differently or test another way?

 

2019 16" MacBook Pro (RP5600M) [9th,8C,H] + RTX 3090 @ 32Gbps-TB3 (Mantiz Saturn Pro) + Win10 [build link]  

ReplyQuote
mac_editor
(@mac_editor)
Famed Member Moderator
Joined: 4 years ago
 

@bmagnien, sounds like you did due diligence while overclocking so everything is probably ok, but some of your offsets seemed unusually high (+1200 mem) to me. Clearly it isn't a stable overclock (which you do mention). Generally, baseline configuration performance numbers interest me more as that's how a typical user would be using their eGPU (but then I suppose typical users aren't purchasing the 3090 :p). Nonetheless, really appreciate your work in highlighting 3090 performance over eGPU! 4k60 gaming is a really good usage scenario for eGPUs in most cases and the 30 series seems to fit the bill well.

 

purge-wranglertbt-flashpurge-nvdaset-eGPU
Insights Into macOS Video Editing Performance

Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro

 
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]  


bmagnien liked
ReplyQuote
bmagnien
(@bmagnien)
Trusted Member
Joined: 2 months ago
 

Update 1 (9/28/2020, 7:21pm EST): Added the following to the end of the original post for context:

Additional Benchmark Comparisons (3080 vs 3090 vs Desktop vs eGPU)

2019 16" MacBook Pro (RP5600M) [9th,8C,H] + RTX 3090 @ 32Gbps-TB3 (Mantiz Saturn Pro) + Win10 [build link]  

ReplyQuote