2016 15-inch MacBook Pro: Quadruple GPU Power with External NVIDIA GTX 1080 Ti

2016 15-inch MacBook Pro

2016 15-inch MacBook Pro

Powerusers who demand powerful graphics performance from their laptops have lamented the weak GPUs Apple have been incorporating into its MacBook Pro notebooks. The lastest 2016 15-inch MacBook Pro did not allay that concern. Fortunately the high-end MacBook Pro comes with full-throttle Thunderbolt 3 connections.

Put in a NVIDIA GeForce GTX 1080 Ti, one of the most powerful desktop GPU NVIDIA has to offer (the top honor goes to the TITAN Xp, at the moment), into an AkiTio Node or a Mantiz Venus enclosure, connect the eGPU to a 2016 15-inch MacBook Pro and quadruple the GPU power. This is exactly what eGPU.io did.

AkiTio Node Thunderbolt 3 eGPU Enclosure

AkiTio Node

The AkiTio Node features a new Texas Instrument TI83 controller that macOS recognizes. A full-sized dual-width GPU — like the GTX 1080 Ti — fits inside. The Node is spacious: 17-inches long (all the way back to the rear handle), 5.75-inches wide, and almost 9-inches tall (including the feet). One x16 PCIe slot is available, with two PCIe 6+2-pin connectors, and is powered by a 400W SFX power supply unit (PSU). In the back is a single Thunderbolt 3 connection.

Make sure NVIDIA’s web drivers for Pascal driver support on macOS are the latest (https://images.nvidia.com/mac/pkg/378/WebDriver-378.05.05.05f01.pkg). Update the Node’s firmware too (https://www.akitio.com/firmware/node-firmware). eGPU.io is reporting some visual glitches using Goalque’s automate-eGPU script that enables eGPU support on macOS for NVIDIA GTX 10 series GPUs. Though compatibility is improving plug-and-play is not how eGPU and MacBook Pro’s work at the moment.

A well-architected Thunderbolt 3 host is a must. And the late 2016 15-inch MacBook Pro is currently the best Thunderbolt 3 host for an eGPU implementation. The reason? The PCIe lanes are directly connected to the quad-core i7 CPU, without having to go through the PCH (Platform Controller Hub). Maximum bandwidth per eGPU in a direct Thunderbolt 3-to-CPU configuration is 22Gbps with a single eGPU attached. Most other Thunderbolt 3-equipped laptops route the PCIe lanes through the PCH. Because the PCH shares bandwidth with other internal components such as network cards, USB ports, etc. the eGPU-connected Thunderbolt 3 connection can get congested, limiting the performance of the eGPU. The direct-to-CPU design of the Thunderbolt 3 ports maybe the key to future eGPU implementations not only for MacBooks but for iMacs and the new modular Mac Pro Apple is working on.

Unigine Valley Benchmark Screenshot

Unigine Valley Benchmark Screenshot

eGPU.io conducted benchmarks comparing the integrated GPUs — AMD Radeon Pro 460 + Intel HD 530 — and the NVIDIA GeForce GTX 1080 Ti TB3 eGPU. The results are a dramatic improvement in GPU capability for the 2016 15-inch MacBook Pro:

  • LuxMark 3.1: 6,056 vs 23,172 (+283%)
  • Valley 1.0: 895 vs 2,353 (+163%)
  • Heaven 4.0: 495 vs 1,422 (+187%)

eGPU benchmarks increase even more using an external display:

  • Valley 1.0: 3,031 (+239%)
  • Heaven 4.0: 2,640 (+433%)

Bear in mind the top LuxMark score for a single device (CPU+GPU) using OpenCL 1.2 and CUDA 8.0.0 running the LuxBall HDR simple scene is 28,456 with a TITAN Xp GPU and a 3.5GHz i7-5930K CPU. This eGPU solution isn’t too far behind the top performer, considering it started from a measly 6,056. According to the top Valley benchmark scores (Valley 1.0 Scoreboard Google Sheets) the highest score is 7,375 powered by a 2GHz Titan X and a Core i7-7700K.

The late 2016 15-inch MacBook with GPU scores like these are incredible, but this eGPU solution does not come cheap. A Mantiz MZ-02 (Venus) eGPU enclosure costs US$389 while the AkiTio Node will set you back $300. The GTX 1080 Ti goes for around $700. Combined is an extra $1000 or so. Of course not everyone needs a GTX 1080 Ti, but even the most affordable VR-ready AMD Radeon RX 480 will set you back about $200, for a total of around $400-$500.

A $400-$1000 investment to significantly improve GPU performance that enables a professional to accomplish the job on time and with less frustration might be more than worth it.

via The Verge