Thunderbolt powered GPU (TB-ExpressCard-ViDock)

Hi guy... I just stumbled over those two things:

Thunderbolt to Expresscard Adapter and ExpressCard to PCI Express


I have to admit I am a bit lost since I am spending no more time on "Hardware/Gaming" things... but I am still interested

So my question is, can this work (e.g. with the new MacBook Pro)?

I hope this question isn't too silly...
Thanks for any help/information! (does sb need +rep? )

 

Theoretically yes, it should work. We won't really know for sure until someone tries it out and I am afraid that spending $350 (plus the cost of a PCI-Express graphics card) is just too much for me to even test the waters let alone take a jump. I am also not sure if OS X has been programed to send and receive graphical instructions via the thunderbolt port. I don't see why not (as I thought that was part of the specs) but you never know as Apple could have done something differently or left something out of OS X.

 

The question I had is isn't OSX only designed to run on the hardware the computer shipped with? You can upgrade ram/hard drive but its designed for the processor and/or graphics card your computer came with.

So if you put in a graphics card not seen on any OSX machine, what happens? Does the OS support it? I don't know if it would unless Apple/the company programmed some stuff for it.

 

When it comes to gaming I think there is no better solution than to go with Windows (on the Mac) ... but are there any drivers for thunderbolt? Does TB need drivers on windows? So you guys tell me there is a chance? Great!

THX, +rep has been delivered

 

Yes there are TB drivers on windows.

Yes the egpu solution has a bunch of followers and people that are expecting it. Me included, it wont come cheap at first, since you will need the cable (50), the gpu, the dock, and possibly a monitor (not all egpu can be run on the notebook monitor)

Im sure as hell going to be one of the first to try it.

To keep you up to speed, there are 4 companies currently developing a solution for it (all the major players in egpu solutions and sonnet). As far as how far they are in the development process we dont have a clue.

And sure as hell that an egpu solution is mostly going to be available for windows and used in windows.

 

Thank you Mr MM

But reading your post... it sounds a bit to me like you're still waiting for a "real" TB egpu... Do you think the "solution" with the adapter I mentioned in the first post wouldn't work or has downsides?

Greetings
Julian

 

Yes.. this $150 TB-to-expresscard adapter would give you an expresscard interface so could attach a DIY eGPU using the provided expresscard slot. This Sonnect product must have the Thunderbolt chips on it to convert the TB signal to pci-e 2.0.

Since this is scheduled for release in October then Sonnet must have the specs and tech ready to release their Echo Express eGPU enclosure at the same time allowing full bandwidth eGPU connectivty. That is the higher performance and better integrated solution to go for.

 

Ah great! Thank you

Just one more question to that side...

... here


It shows TB "20 Gbit/s"... because of two channels... I thought it would "only" be 10 Gbit/s in each direction... am I getting this wrong or is it possible to "use" 20Gbit/s in one direction?

and as always THX 4 more infos
(sry I'm empty... spend too much +rep in the last 24h... is it limited to two times each 24h?)

PS: why is your name black?

 

didnt follow the links, and as nando4 pointed out its better to wait for the sonnet express, or the others who are working with this kind of egpu, currently nando4 is one of the few here who undertand the complexity of the task and is currently asking people on how the development is going

 

btw, if you want any other GPU to work in a mac be sure to use the flashing tool "zeus"

 

Considering that the old (pre-2010) PCI Express spec provided 64 Gb/s and PCI Express 3.0 provides 128 Gb/s, I also wonder whether TB can provide adequate bandwidth for modern GPUs even if both channels can be used in the same direction.

 

The TB cable (as far as the information that is available) can do 2 channels out of the four 10gbs channels that the light ridge controller have.

Each channel is bi directional, with the speed of each being 10gbs, so in theory you have 40gbs of bandwidth right there, which is similar to pcie x4 2.0.

Now regarding the gpus now using pcie x16 2.0, they use it because they can, not because they need it. a pcie x8 wont hamper any card out there, regarding pcie x4 yes there will be some titles as shown in several reviews that there will be some form of performance degradation, those include COD series and some others, so the performance can do 0-50% bandwidth degradation, being the most extreme 50% on the COD series

 

So with a better cable, but the same hardware (e.g. MacBook Pro), it could be possible to get 80Gbit/s out of it?

THX

 

I should have worded it better, there is a bottleneck along the way from the controller card to the destination device, I speculate that its the cable, from what I have seen in the marketing pdf from intel. I dont know if its the conector in the port, or if its the cable itself.

So yeah once you deal with that you can be sure as hell get the full bandwidth of the light ridge controller. It will be a good news for the mbp owners, not so much for the air owners, since they use the eagle ridge controller that only gives them 2 TB channels on the controller, differently from the light ridge wich gives 4 TB channels on it.

 

Why dont we get someone with some circuit building knowledge to source the conversion chips involved and get a new device manufactured by one of those PCB companies?

 

are you offering yourself?

 

A Thunderbolt port multiplexes a x4 2.0 pci-e and a Displayport signal. Need a Thunderbolt chip to split those signals back to it's components. Problem is that Intel and Apple are not releasing the Thunderbolt chips nor any specifications unless you are a partner and sign a NDA. Sonnet is a partner. Not sure if Apple/Intel will make Villagetronic one.

So then I believe the Sonnet's Echo Express should be the first to market but I'm wondering how many Apple folks will buy it given the the rumoured US$500 RRP cost for the basic enclosure? An early adopter premium? Can buy a whole desktop system for similar costs.

I'm of the opinion that a Lenovo X220 + DIY eGPU (soon to be x1 2.0 Optimus with a new cable so that approximates x2 2.0) is significantly better bang-per-buck for this round. Sure, it's lower performance - about half the bandwidth and some of that is used if running the internal LCD mode. Now if Ivy Bridge brings Thunderbolt support for a larger PC notebooks market then those costs will surely become significantly less.

 

I fear that most will use a light peak version (like Sony does) and not the Thunderbolt (Displayport) version... that would keep the prices for thunderbolt stuff high and (may be) the amount of devices low

 

WOW thats quite a premium, I knew that it wasnt going to be cheap, but now, I think Im going to wait for more implementations to get some competition going.

 

Except that, as I found out, Sony implemented something that wasn't standard. Anyone that applies now for lightpeak/thunderbolt would be implementing the same standards as Apple. Essentially Sony was able to get their request in before a standards committee was formed (or something along those lines). So you shouldn't see backwards USB ports being used for lightpeak/thunderbolt in the future as the implementation used by Apple is the standard.

 

Oh that is good news As is saw the Sony Notebook I thought: OMG they use the cooler/better fiber LightPeak version... TB gone die..

I read something about "USB 3.0 in Windows 8 but no TB?" ... as mentioned before there are already drivers for TB on W7 so what dies that mean?

And are we going to see mainbords with a display/TB port soon?

Greetings
Julian

 

There are 3 partners in the development of light peak/thunderbolt.

Intel, Apple and Sony.

Those OEMs, were given the rights to use the new form of I/O in their system with a one year advantage against the other OEMs.

The first implementations/demos of the that tech were demonstrated by intel using a USB connector. The controller is connector independent.

So how does that leave us?

Intel is part of the USB implementers forum. And it supports USB3, albeit it wont look like that way, due to the lack of integration of the chipset.

What people seen to forget (as I did) is that USB3 and light peak/thunderbolt arent competing, they are geared for different users, with different needs.

But I digress. The point here is that the specification, thus the standardization of that new tech is thunderbolt. The connector is mDP, there are 2 controllers, and for now there wont be a fiber connection.

 

But the sony notebook (Vaio Z) uses fiber doesn't it? (but they won't continue that?)

But you agree that the future of "the new I/O" is Thunderbolt -> great

 

I think what they did is to get into production some of the prototypes that they had and do it.

Indeed the future is the one dictated by the standards and the one that make those and the controllers is intel.

 

Tata: Is this it?

ExpressBox 3T

Greetings
Julian

PS: still not too sure about this is the right device^^

 

it can be...

I will add a correction here:

The TB controller uses a PCIe x4 2.0 connection. So in all due things it gives us that.

However here is where I got it all wrong again (I have corrected myself before, but my memory is giving me the kicks again)

A single pcie 2.0 lane is 5gb/s, thus giving us a theoretical connection of 40gb/s (as I said before)

The deal here is that the thunderbolt controller gets those lanes and converge/split those lanes in 2 channels, each lane is capable of 10gb/s bidirectional, and this is where I have been wrong here, its 5gb/s in/out and 5gb/s in/out. If I were right in my last statement we would be using the pcie x4 calculations using 16 lanes

So there isnt a limitation in the cable. The problem is in the controller, light ridge can provide 4 channels, which is good for the imac owners, since they can use the 2 TB ports that they have, but not good enough to use the full pcie x8 bandwidth that it could theoretically provide, since the pcie x16 is divided as follows:

x8 - gpu
x4 - usb hubs (keyboard, bluetooth, infrared receivers, usb)
x4 - thunderbolt

Thus the problem remains that in this instance 2 pcie x16 would be needed to provide the bandwidth necessary for the controller or whatever its blocking us of accessing higher bandwidth.

So yeah, we might be also limited by the cable, since it can theoretically provide only 2 channels, thus voiding the idea of putting another pcie controller in the mobos of the mbp/mba line, maybe even the imac line (dunno how much pcie that they have)

Thus a move to fiber might void the cable limitation thus expanding the power of the thunderbolt controller. Looking things at this perspective give me, at least, the picture of why the thunderbolt controller is limited to only those speeds, since it can go much higher.

We also have to remember that the fiber move is on prototypes, that actually just work at the speed of the copper cable that it we have now, one example is the vaio z21, and the prototype that was shown at IDF in 2009 and the one at may 2010. The problematic part is that the tech might be too expansive for full deployment, since it needs to be cheap enough to maintain the price of the higher end range of notebooks (800- to whatever dollars) contained.

One thing that caught my attention is that probably the problem regarding passing the info to the ones trying to come up with techs using thunderbolt aint in intels end, but might be on apple and sony. The exclusivity agreement that both share, might be the problem.

Giving companies that might produce solutions that werent sanctioned by both, might be a problem by diluting the quality of what is the new tech. Although it may sound stupid, since more availability of products is indeed a good thing, to maintain the quality of the products and the shared design ideas is another thing that is quite worthwhile, it provides strength to the brand, at the cost of leaving the widespread product development to next year and possibly the future

 

Edit: I guess you meant "x1 PCIe 2.0 0,5 GB/s"

So as a MacBook Pro user I could get a max of "theoretical" 2GB/s out of that one TB port related to the x4 PCIe 2.0 lanes and the fact that we can only access two channels? (I guess more channels wouldn't increase that?)

What I don't get is why the MacBook Air users cry because they get less channels than the Pro users... but the Pro can't benefit from more because it only has x4 and one TB port (as well)...
Or am I mixing this up? (I'm trying hard to understand it but I guess, as my mother language is not english, this makes it even harder for me...)

thank you very much!

 

Yes x1 = 1 lane, however its 5GB/s

No a single TB channel can have pcie x2 2.0 lanes, thus 10GB/s of bandwidth. The problem as I said in the afore mentioned post is that apple only uses pcie x4 2.0, thus limiting it to 40GB/s maximum theoretical. So yes for the second question, and we enter the other problem that is we dont know if the port or the cable is limited to deal with 2 TB channels, since clearly the controller isnt the problem

Indeed there is nothing to complain, and if I got this right both will (and are) providing the same speed (actually the eagle ridge controller in the mba while being cheaper is a bit faster)

It aint also my mother language. its difficult sometimes.

 

I looked it up twice now and it says:

Capacity
Per lane (each direction):
v1.x: 250 MB/s (2.5 GT/s)
v2.x: 500 MB/s (5 GT/s)
v3.0: 1 GB/s (8 GT/s)
16 lane slot (each direction):
v1.x: 4 GB/s (40 GT/s)
v2.x: 8 GB/s (80 GT/s)
v3.0: 16 GB/s (128 GT/s

( wikipedia)

But I get what you mean about the "bottleneck" now

Thank you very much one more time

 

I didn't want to make a new thread so I'll just post it here

Magma if offering the first Thunderbolt equipped PCI Express Box! ExpressBox 3T

It offers:
"Three PCIe 2.0 expansion slots (Two x8 and One x4)"
"220W power supply with (2) 4-pin auxiliary connectors"

but only?!?
"Up to 10Gbps Thunderbolt connection" (what about the "second channel"?)

... and a price of 979$ which is way higher than I excepted it to be...


ATI offers x8 and x16 with it's XGP so may be this is an other bottleneck... or do you guys think the 220W or the Thunderbolt (10Gbps) are even more bottlenecking the GPUs?

I am looking forward to other solutions because 979$ is way out of my league...

 

not to ruin your parade I knew that the price of the magma was going to be steep. its geared towards professional users, its a 3 pcie slot case

 

So you expect the GPU only versions to be way more affordable?

 

or at least I hope so