I’ve been looking for the best Thunderbolt 3 eGPU host. The requirements are quad-core CPU, iGPU only, and 4x PCIe 3.0 lanes. The true unicorn would also have the PCIe connection over Thunderbolt 3 directly attached to the CPU. There’s none out there atm unfortunately.
The 15″ Lenovo Yoga 720 is a bargain in the US market for its specifications. Best Buy currently has July 4th sale which reduces its price to $899.99. You can stack a $125 off education coupon. $774.99 gets you a nice 2-in-1 with an i7-7700HQ CPU, 8GB of RAM, and 256GB PCIe SSD (RAM and SSD are both upgradable). This particular configuration comes with iGPU only, Intel HD Graphics 630. I thought this would get me very close to the perfect eGPU host.
Sadly it’s allocated with only 2x PCIe 3.0 lanes for Thunderbolt 3. This is very disappointing. Intel Thunderbolt Software even shows this laptop is supporting eGPU. The Lenovo booth was spotted recently displaying a Razer Core rebadged with Lenovo logo so I’m running this Yoga 720 with a GTX 980 Ti Razer Core eGPU.
eGPU setup was plug-and-play. I’ve noticed the Razer Core has a built-in notification system prompting you to install Razer Synapse software. Other eGPU enclosures do not have this feature. Im wondering if it’s the communication between the Razer Core firmware and Intel Thunderbolt Software. It’s a very handy way to provide setup instructions for people doing eGPU for the first time.
| Lenovo 720 + Razer Core | 980 Ti eGPU Internal | 980 Ti eGPU External |
|---|---|---|
| Unigine Valley | 62.9 FPS | 75.7 FPS |
| Unigine Heaven | 61.6 FPS | 75.2 FPS |
| Unigine Superposition | 72.9 FPS | 84.4 FPS |
| 3DMark Time Spy | 29.7 FPS | 31.5 FPS |
| 3DMark Fire Strike | 58.4 FPS | 68.7 FPS |
| Rise of the Tomb Raider | 46.3 FPS | 50.0 FPS |
| Tom Clancy’s GhostRecon | 39.9 FPS | 52.5 FPS |
thanks for these tests! 🙂
these are benchmarks for a gtx980 ti with a desktop PC as a comparison
https://www.notebookcheck.net/NVIDIA-GeForce-GTX-980-Ti-benchmarks-and-specs.169446.0.html
http://forum.notebookreview.com/attachments/980ti-png.146663/
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts
I have a question for you 🙂
there is a 36% between nvidia gtx980ti and amd rx580
http://gpu.userbenchmark.com/Compare/Nvidia-GTX-980-Ti-vs-AMD-RX-580/3439vs3923
I'm watching at your new review
https://egpu.io/thunderbolt-3-host-dell-precision-7520-review/
| Radeon RX 580 eGPU | Precision 7520 Internal Display | Precision 7520 External Display | |
|---|---|---|---|
| Unigine Valley | 47.0 FPS | 49.2 FPS | |
| Unigine Heaven | 47.1 FPS | 48.7 FPS | |
| Unigine Superposition | 54.2 FPS | 57.6 FPS | |
| 3DMark Time Spy | 26.7 FPS | 27.6 FPS | |
| 3DMark Fire Strike | 56.4 FPS | 57.1 FPS | |
| Rise of the Tomb Raider | 55.7 FPS | 57.2 FPS | |
| Tom Clancy's Ghost Recon | 39.3 FPS | 43.1 FPS | |
| Shadow of Mordor | 69.7 FPS | 87.0 FPS |
and performance differences are about (gtx980ti vs rx580)
25% using internal display
30%-35% using external display
these differences (25% is better than 36%) are caused by tb3 x2 vs tb3 x4? there is also an influence of ti82 chip vs ti83 or different firmawares?
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts
@sicily428 when you compare the external display mode numbers, the difference is in line with that 36% overall estimate. The performance gap becomes smaller when you look at internal display mode numbers. This is due to both AMD Xconnect’s better optimization for internal display mode and the x2 PCIe lanes over Thunderbolt 3 on the Lenovo Yoga 720.
In Windows, the USB-C controllers (TI82 vs TI83) make no difference in performance. Both the Razer Core and Sonnet Breakaway Box have firmware to use ~22 Gb/s.
@sicily428 when you compare the external display mode numbers, the difference is in line with that 36% overall estimate. The performance gap becomes smaller when you look at internal display mode numbers. This is due to both AMD Xconnect's better optimization for internal display mode and the x2 PCIe lanes over Thunderbolt 3 on the Lenovo Yoga 720.
In Windows, the USB-C controllers (TI82 vs TI83) make no difference in performance. Both the Razer Core and Sonnet Breakaway Box have firmware to use ~22 Gb/s.
Thanks for your answer! does AMD Xconnect work only with AMD gpus?
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts
AMD XConnect is the graphics switching software for AMD eGPUs. Nvidia cards use Nvidia Optimus.
AMD XConnect is the graphics switching software for AMD eGPUs. Nvidia cards use Nvidia Optimus.
I didn't know. Thanks 🙂
my knowledge about amd gpus is really small
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts
I ran this Lenovo Yoga 720 with an RX 580 eGPU. The other three results were with a Dell Precision 7520 (x4 TB3 eGPU) and Z170 Test Bench (x4 TB3 eGPU & x16 PCIe dGPU).
 |
 |
 |
 |
If I understood correctly these tests are made with an external monitor. could you make these tests also with internal monitor? so we can also measure performance differences in that case
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts
Thank you a lot itsage for these tests: great comparison!
How do you explain these incoherent results:
- pci-ex16x vs tb 4x bench
Of course the 16x has all better scores, something between 14-48% BUT 3dsmax and snx has 278% and 160%... too much difference!
- tb 4x bench vs tb 4x dell
This one is interesting because tb is the same 4x but one cpu is 50% less power (i5 dell): scores are equal because bench goes 2-12% but also dell goes 1-14%. So, cpu power seems not important.
The strange thing is that 14% dell on catia04: that's the same also for lenovo tb 2x!! How is it possible tb 4x bench so low on that?
- tb 4x dell vs tb 2x lenovo
There is not a clear winner... 4x has better scores only in 3 tests 5-14%, other 2 tests are quite equal, one is better in the 2x with a 23%!
Overall not bad at all for 2x...
So, at the moment 2x seems not too different vs current implementation tb 4x (sadly limited to 22Gb...).
pic with percentages:
i.imgur.com/bQA04fu.jpg
Pending: Add my system information and expected eGPU configuration to my signature to give context to my posts