Thunderbolt and PCIE

[Deleted member 4644]

Not sure if this is the right forum..

http://developer.apple.com/library/m...011138-CH2-SW5

can someone confirm the speed of a current Thunderbolt cable (single direction). I think it is 10 Gb/sec or roughly 2.56 lanes of PCIE 2.0.

Is that right?

10 gigabit = 1280 megabit.

Pcie 2.0 lane is 500 MB/sec ( http://en.wikipedia.org/wiki/PCIE )

So 1280/500= 2.56.

Does that sound right?

So the next gen Thunderbolt (20gbs) should be around 5.12 lanes of PCIE 2.0?




Question 2:

Would 5.12 lanes of PCIE 2.0 be enough for a modern mid-high gaming card?

 

[Jaydip]

Isn't that should be 1280/(500*8) ?

 

[ViRGE]

Just as a minor correction: Thunderbolt can have multiple channels. Typically from a PC you'd have 2 channels, not just 1. But those channels cannot be bonded, so the most any device can use is 10Gb/sec.

Anyhow, your math is correct. As for whether 5 PCIe 2.0 lanes would be enough, the short answer is no. You wouldn't be badly crippled, but you do give up performance. And this doesn't take into account the additional latency of using Thunderbolt.

 

[Black Octagon]

As for whether 5 PCIe 2.0 lanes would be enough, the short answer is no. You wouldn't be badly crippled, but you do give up performance. And this doesn't take into account the additional latency of using Thunderbolt.

Hey ViRGE, what is this additional latency you speak of (and are you still talking about it in terms of Thunderbolt as a display interface between GFX card and monitor)? Additional as compared to what?

 

[Arkadrel]

USB 3.0 supercharged can do 10gbps as well.
(and is backwards compatible with older usb devices)

PCIe 3:

....PCIe 3.0's 8 GT/s bit rate effectively delivers 985 MB/s per lane, double PCIe 2.0 bandwidth.

- PCI express wiki

PCIe 4:

On November 29, 2011, PCI-SIG announced PCI Express 4.0 featuring 16 GT/s, still based on copper technology. Additionally, active and idle power optimizations are to be investigated. Final specifications are expected to be released in 2014/2015.

- PCI express wiki

PCIe 4.0 is probably gonna double the PCIe 3.0 performance.

Would 5.12 lanes of PCIE 2.0 be enough for a modern mid-high gaming card?

People and their dreams of external GPU drives for laptops.
Its gonna happend at some point, I guess.

I think 10gbps should be fast enough, so Im not sure why we dont see alot of USB3/Thunderbolt external GPUs.
Im guessing there is some reason for it though, because it would probably be popular next to a docking station (where you plant your laptop).



watch this video (lucid external GPU demo):
http://www.tomshardware.com/news/lucid-gpu-graphics-thunderbolt-external,17520.html

(same thing could work on USB3)

 

[ViRGE]

Hey ViRGE, what is this additional latency you speak of (and are you still talking about it in terms of Thunderbolt as a display interface between GFX card and monitor)? Additional as compared to what?

There are 3 points of additional latency when using Thunderbolt.

1) The TB controller in the PC

2) The TB controller in the sink device (even more if you daisy chain them)

3) The latency from the signal propagating down the TB cable.

The first two are similar in concept to the latency introduced by a PCIe bridge. Unfortunately a TB controller is a good deal more complex, and I don't believe anyone has a way of measuring it. The final one is just a matter of physics: a TB cable goes up to 3M (~10ft), and that takes time to propagate down the cable. Compared to PCIe traces which are measured in inches, that's quite a large propagation delay for a high speed bus.

Latency isn't an insurmountable problem, but there will be some kind of performance impact. More importantly though, as GPUs aren't currently done via TB, it's unclear whether AMD and NVIDIA's current hardware and drivers would play nicely with TB (i.e. how quickly do they expect a response?). It works in proof of concept demos, but those are under controlled circumstances.

 

[Black Octagon]

Thanks a lot. Very useful