Hi guys,
I'm here to tell you what we're discovering about this "Light Peak" that Sony used in Z21.
As you know, Sony used the former Light Peak Technology that is made of optical fiber instead of active copper cable. In addition it seems that the technology used by Sony differs from the former Light Peak and it's more like the actual Silicon Photonics Link by Intel. This technology use basically hybrid lasers, modulators,a muxer, an optical fiber, a demuxer and photo detectors. Each "subchannel" can go at 12,5 Gb/s and the former base project was made of 4 sub channel to obtain a 50 Gb/s.
http://techresearch.intel.com/spaw2/uploads/files/Intel_SiliconPhotonics50gLink_FINAL.pdf
We think that in reality that's the same technology that Sony used:
http://weekly.ascii.jp/elem/000/000/048/48360/110706gian_vaio003_1000x.jpg
With this technology, the demuxing of data would be made by photodetectors connected electrically to their own pins set. If the flux data is from a single peripheric and type, there's no need for additional demuxing and so no additional TB controller (eagle ridge or light ridge) avoiding Apple licenses.
We guess there's no TB controller since... there's no one on laptop and power media dock (maybe we didn't see it, who knows).
Stated this, if they used the former technology, which goes @ 50 Gb/s per channel/fiber (and we have 2 fiber/channel)(using 2 lasers in one direction and 2 in the opposite, not 4 in one direction) we think that one fiber is for direct connection of GPU to PCI-E bus from CPU and the other one is for DP connection to laptop screen. We're waiting for someone to run AIDA 64 and see if the screen accept DP input. Otherwise, it would need LDVS muxer/demuxer. As for the GPU, we think that the PCI-E bus being directely connected, and not having any other peripheric connected, it's set @ 16X 2.0 (granting a maximum bandwidth of 128 Gb/s). Since fiber channel goes at 50/ Gb/s we believe that GPU goes at least @ 4X 2.0 (32 Gb/s) or maybe like a 6X (48 Gb/s since the maximum bandwidth possible would be limited by fiber channel).
A configuration @ 4X 2.0 would show a decrease of performance nearly 20% against a 16X 2.0. And that's what we see from benchmarks between Sony SA (6630M) and Sony Z21 (6650M). 6650M is 10% faster then 6630M; in some test seen in a japanese review, PMD goes 24.86 and if it's a real 4X bandwidth, we'd have a loss of 20% (avarage through some review in scaling performance)..so, if it would be @ 16X we would have 31.075. SA/6630M goes 28.32 and since 6650M it's 10% faster -> 6650M would have to get 31.152. Results are quite similar, so it may be a real 4X.
I remember you that the loss of performance, changing real PCI-E bus can decrease performance in respect to 16X 2.0 by:
1x 1.0 20%-80% loss
1x 2.0 15%-60% loss
2x 2.0 10%-45% loss
4x 2.0 5%-25% loss
8x 2.0 0%-15% loss
[ better results in synth benchmark /worse result in games]
Even very low-hand video cards show the same or little reduced but always remarkable performance loss. A 7300 GT showed loss till 75% ( remember that GDDR memory bandwidth it's different from PCI-E bandwidth)
As for the additional ports on PMD, VGA and HDMI are controlled directely by eGPU and USB 2.0 3.0 ethernet and ODD are likely to go to the metal circuit of Light Peak port (Light Peak port is both optical fiber/USB 3.0 compliant). If those flux data still go with optical fiber, there must be an additional demuxer as TB controller to separate data and direct it in the proper pins out.
It's all speculation, for sure. But it's very likely to be so. We're trying to develop an eGPU home made in collaboration with some companies. For both TB eGPU and Sony's LP eGPU. For thunderbolt eGPU the problem would be to find light ridge/eagle ridge controllers and replicate TB cable which is the only cable PCI-E 2.0 compliant at the moment. The 2 chips on it are sold only to Apple till the end of their exclusive (so it ends on the starting of 2012).
For Sony's LP we have 2 ways: copy in some way the cable ordering trasmitter and receiver and guess pins out set or modifying power media dock substituting BGA gpu and re-rooting channels.
As for TB technology, there are two controllers: Eagle Ridge and Light Ridge. Eagle Ridge is capable of 4 x 10 Gb/s bidirectional flux, able to achieve a total of 80 Gb/s (more then enough to achieve a 8X 2.0 PCI-E when connected to such a bus) while Eagle Ridge which is on MacBook Air is capable of 2 x 10 Gb/s bidirectional flux for an overall of 40 Gb/s, more then enough to reach a 4X 2.0 PCI-E bus flux. But in reality, there won't be any difference. Actual Thunderbolt cable are able to go @ 20 Gb/s overall thanks to active chips on it (that's why they cost a lot). They cannot go beyond this limit at the moment. And I can assure you that without those chips it's very DIFFICULT to achieve a PCI-E 2.0 compliant cable (we're trying to do that). So This generation of thunderbolt, will be like a 2x 2.0 PCI-E bus. (16 Gb/s). Expect to have a loss from 10% till 50% in worst cases.
Here some links for what we're trying to do:
http://forum.notebookreview.com/gaming-software-graphics-cards/418851-diy-egpu-experiences.html [DIY ViDock thread - here there is a parallel project ti ViDock from Village Tronic to use home made boxes with eGpu. We're trying to develop things also for Thunderbolt and Sony's Light Peak ->
http://forum.notebookreview.com/7575216-post4031.html
http://forum.notebookreview.com/sony/600942-vidock-4g-light-peak-z21-possible.html [here is a thread meant to push Village Tronic to develop something also for Sony Vaio Z21, they've already wrote an Open letter to Sony Community like the previous one for Mac community. If there's around someone who's interested in Z21.. ->
https://www.facebook.com/topic.php?uid=89321949134&topic=17792#topic_top]