eGPU Macbook - Introduction
Always wanted to play the newest games on your Macbook? It's possible, and it got even more affordable. The eGPU is not something new, it's available for quite sometime, but it wasn't really a performance booster back then. But with the release of Thunderbolt 2 on the Macbooks it got interesting. So I decided to do some investigation about the technical aspects of this concept.
The main questions is, are we able to utilize full eGPU performance? Is there any loss? What about SLI or CrossFire? (Because we are connected through Thunderbolt 2).
For this example I have taken my own Macbook as example.
Macbook Pro Retina Late 2013 with nVidia GT750M.
This Macbook comes with an Intel i7-4850HQ, based off Intels product page we can conclude that this CPU supports upto 16 PCIe lanes.
Respectively; 1x16, 2x8, 1x8/2x4. (PCIe 3.0).
Reference: here.
So we have a total of x16 support for this Macbook. Let's see the PCIe parts of the Macbook.
512GB PCIe SSD uses x2 (PCIe 2.0) lanes.
If you have an upgraded SSD, your 1TB PCIe SSD uses x4 lanes.
Reference: here.
USB 3.0 Ports (two ports) uses x2 lanes (probably PCIe 2.0 too), each USB port x1 lane. Older Macbooks uses x1 lane for both USB ports.
Reference: here.
nVidia GeForce GT 750M 2 GDDR5 uses x8 lanes (PCIe 2.0).
Reference: mac sys info.
So how many PCIe lanes are left? What PCIe configuration did Apple use?
Let's proceed with the Thunderbolt part. The Macbook we are talking about supports Thunderbolt 2 (shortened TB2) and uses the DSL5520 from Intel. (Reference: here.) This is the Thunderbolt controller which supports 4 TB channels. Each TB port does have 2 channels, every channel caps 10Gb/s data transmission. With the new TB2 technology we can burst up to 20Gb/s. (Reference: here.) I personally think that there is no benefit for eGPU's using TB1 or TB2 technology. TB2 is mainly released for 4K video. (Reference: here.)
Conclusion: Thunderbolt connection supports up to x4 PCIe lanes.
Some more information about PCIe speeds, which comes in handy.
What's with this gigatransfers nonsense? Gigatransfers per second are the same thing (in this case) as gigabits per second, but they include the bits that are lost as a result of interface overhead. All PCI Express versions lose some of their theoretical maximum throughput to the physical overhead associated with electronic transmissions.
Per lane (each direction):
v1.x: 250 MB/s (2.5 GT/s)
v2.x: 500 MB/s (5 GT/s)
v3.0: 985 MB/s (8 GT/s)
v4.0: 1969 MB/s (16 GT/s)
So, a 16-lane slot (each direction):
v1.x: 4 GB/s (40 GT/s)
v2.x: 8 GB/s (80 GT/s)
v3.0: 15.75 GB/s (128 GT/s)
The modern day graphics card comes with a x16 slot, this means it can use x16 lanes maximum. Using Thunderbolt we are only able to feed x4 lanes of the 16. Wouldn't this mean 4/16*100 = 25% performance of the GPU will be utilized?
The dual GPU theory. When using the second Thunderbolt connection for a second GPU, would make it possible to run it CrossFire or SLI.
Theoretically, this will provide 50% performance of TWO GPU's.
IF both TB ports offer x4 lanes. But I doubt that, because I am still wondering what PCIe configuration Apple uses for this Macbook.
I am open for comments, let's discuss this! Please comment below, so everyone can benefit from this.
discoholicmusic
baratheonpilled
ellievsbear
izzy806
2087
areudirty