11" MacBook Air Owner Connects High-End Graphics Card With Complex Thunderbolt Setup

[ihuman:D]

The above information is incorrect by a significant margin.

No it isn't.
Expresscard 1.0 = 1.6GBbits/S
Thunderbolt = 10GBits/S

Therefore you're only capable of using 10% of the available bandwidth.

I wasn't on about the performance gain, I was on about the performance loss of using Expresscard as the "middle-man" instead of using a more direct Thunderbolt to PCIe adapter.

 

[nando4]

No it isn't.
Expresscard 1.0 = 1.6GBbits/S
Thunderbolt = 10GBits/S

Therefore you're only capable of using 10% of the available bandwidth.

I wasn't on about the performance gain, I was on about the performance loss of using Expresscard as the "middle-man" instead of using a more direct Thunderbolt to PCIe adapter.

It's not my style to be so forthright but I feel such misinformation can deter users from considering such a solution. Please do some homework to do before posting more. How to do that?

Look up ANY Intel datasheet for a Series-6 (2nd gen i-core) or newer chipset. It will show the southbridge has up to 8 x1 pci-e 2.0 ports. This is where a expresscard or mPCIe (wifi) slot attaches.

http://en.wikipedia.org/wiki/PCI_Express#PCI_Express_2.0 tells us:

The PCIe 2.0 standard doubles the transfer rate compared with PCIe 1.0 to 5 GT/s and the per-lane throughput rises from 250 MB/s to 500 MB/s.

That is, x1 2.0 is 5Gbps. Furthermore, I've linked you actual performance testing over a pci-e 2.0 expresscard and Thunderbolt link finding the performance difference between them averages 6.3% for external LCD and 14.6% when driving the internal LCD. We can then say with confidence that there is, on average, less than 15% real-world difference b/w 10Gbps Thunderbolt and 5Gbps Expresscard.

 

[jjahshik32]

My question is, does this only work with macbook/air/pros with integrated graphics or can you use this on any macbook pro equipped with thunderbolt?? If so, I would love to use this on my 2011 early macbook pro. Just turn that sucker into a gaming rig and move on.

 

[ihuman:D]

It's not my style to be so forthright but I feel such misinformation can deter users from considering such a solution. Please do some homework to do before posting more. How to do that?

Look up ANY Intel datasheet for a Series-6 (2nd gen i-core) or newer chipset. It will show the southbridge has up to 8 x1 pci-e 2.0 ports. This is where a expresscard or mPCIe (wifi) slot attaches.

http://en.wikipedia.org/wiki/PCI_Express#PCI_Express_2.0 tells us:

That is, x1 2.0 is 5Gbps. Furthermore, I've linked you actual performance testing over a pci-e 2.0 expresscard and Thunderbolt link finding the performance difference between them averages 6.3% for external LCD and 14.6% when driving the internal LCD. We can then say with confidence that there is, on average, less than 15% real-world difference b/w 10Gbps Thunderbolt and 5Gbps Expresscard.

Your on about Expresscard 2.0, my bad, you're right.

--------------------------------------------------------------------------------------------------

My question is, does this only work with macbook/air/pros with integrated graphics or can you use this on any macbook pro equipped with thunderbolt?? If so, I would love to use this on my 2011 early macbook pro. Just turn that sucker into a gaming rig and move on.

It works with any Thunderbolt-equipped Mac.

 

[mytdave]

why the frankenair?

Or... You could chose from one of the following (or many other) Mac compatible GPU cards:

NVidia GeForce 8800GT
NVidia GTX 285
NVidia GTX 680
NVidia Quadro FX4500
NVidia Quadro FX5600
ATI Radeon HD 2600XT
ATI Radeon HD 4870
ATI Radeon HD 5770
ATI Radeon HD 5870

...and simply put the card in one of these nicely finished chassis (depending on card size), plug it into the thunderbolt port (without having to use MS Windows) and call it a day:

http://www.sonnettech.com/product/echoexpresschassis.html

 

[HurtinMinorKey]

We're about to see a flood of these solutions...
http://www.anandtech.com/show/7040/computex-2013-thunderbolt-graphics-from-silverstone

----------

Or... You could chose from one of the following (or many other) Mac compatible GPU cards:

NVidia GeForce 8800GT
NVidia GTX 285
NVidia GTX 680
NVidia Quadro FX4500
NVidia Quadro FX5600
ATI Radeon HD 2600XT
ATI Radeon HD 4870
ATI Radeon HD 5770
ATI Radeon HD 5870

...and simply put the card in one of these nicely finished chassis (depending on card size), plug it into the thunderbolt port (without having to use MS Windows) and call it a day:

http://www.sonnettech.com/product/echoexpresschassis.html

This is more expensive, but a much cleaner solution.

 

[nando4]

We're about to see a flood of these solutions...
http://www.anandtech.com/show/7040/computex-2013-thunderbolt-graphics-from-silverstone

----------



This is more expensive, but a much cleaner solution.

The $250 450W Silverstone Thunderbolt enclosure would be the most cost effective high performance eGPU enclosure to date. However there is some scepticism that it will ever be released. kloper makes a point that such solutions have been promised before, eg: MSI GUS II, but nothing came of them. He then highlights that Intel's "not supported" stance on eGPUs being a likely reason.

The Sonnet Echo Express II is a dual-width, full length TB enclosure. At $799 RRP it's not cheap. Consider that kloper links the $170 PE4H 3.2 as an option to use rather than the $70 PE4L 2.1b if wanting a tidier solution:

Above: $170 PE4H 3.2 can be used to get a tidier solution for $100 more than kloper's PE4L 2.1b

 

[Dudical]

It's important to note that the Thunderbolt standard is not powerful enough to run a PCIe graphics card at anywhere near full speed. Even Thunderbolt 2 doesn't have what it takes.

Thunderbolt: 10Gb/s
Thunderbolt 2: 20Gb/s
PCIe 3.0 16x: 128Gb/s

Another reason why Thunderbolt is just an expensive port for external storage.

 

[nando4]

It's important to note that the Thunderbolt standard is not powerful enough to run a PCIe graphics card at anywhere near full speed. Even Thunderbolt 2 doesn't have what it takes.

Ivy Bridge PCI-Express Scaling with HD 7970 and GTX 680 (techpowerup) found:

x4 2.0 (16Gbps) = 94% of x16 2.0 (AMD)
x4 2.0 (16Gbps) = 88% of x16 2.0 (NVidia)

That is, x4 2.0 (16Gbps) averages to be 91% of x16 2.0 desktop performance. 'Thunderbolt 2' is 20Gbps so will be slightly better still.

 

[Psychj0e]

If he were to trade that 11in for cash, I bet that he could get a very nice gaming PC. Haha. Then.. no need for so much pain trying to run all that jazz, cables and win 7 on a MBA. Lol.. He definitely got the wrong type of computer. XD

I dont think you understand. it's what he's done that's so great.

Hows iPad matlab going btw?

 

[Michael CM1]

I think the other poster meant that you can upgrade the external Thunderbolt GPU, not the MBA's.

The iMac is upgradable, you can upgrade nearly everything but the GPU and you can use Thunderbolt for that anyway.

It's not as upgradable as the mid-towers I built for a few years, but yes, it is upgradable. I wish Apple would start developing iMacs with SSDs as easy to replace as RAM in this model. Most of the built-in SSDs look like DIMMs to me, so it would be nice to have an easy access option for two years down the road when twice as much storage costs half as much.

The new iMacs still have easy RAM access, so I don't see why it would be a leap to make SSD access like that.

By the way, didn't know you could upgrade a GPU using Thunderbolt. It probably has a lot to do with getting the last iMac model WITHOUT Thunderbolt. Boo. That makes my guesswork on the upgradability of those upcoming Mac Pros correct. I wonder if someone will develop an external processing box. I could use one of those for video encoding.

 

[AidenShaw]

umm, no

...

That is, x4 2.0 (16Gbps) averages to be 91% of x16 2.0 desktop performance. 'Thunderbolt 2' is 20Gbps so will be slightly better still.

T-Bolt 2 is 16 Gbps PCIe x4 2.0, same as any other PCIe x4 2.0.

 

[nando4]

T-Bolt 2 is 16 Gbps PCIe x4 2.0, same as any other PCIe x4 2.0.

U sure? If it is then it's not a doubling of TB 1.0, rather it's a +60% improvement. They'd need to up-spec the electrical pci-e link to x8 2.0 to be able to encompass a full 20Gbps downstream TB link.

 

[AidenShaw]

U sure? If it is then it's not a doubling of TB 1.0, rather it's a +60% improvement. They'd need to up-spec the electrical pci-e link to x8 2.0 to be able to encompass a full 20Gbps downstream TB link.

T-Bolt 1 was 8 Gbps PCIe - so 16 Gbps is a doubling.

PCIe 2.0 uses an 8b/10b bit encoding - 8 bits of payload have 2 additional bits of error detection/correction.

That's 20% overhead - so 10 Gbps on the T-Bolt wire is 8 Gbps of payload for PCIe. 20 Gbps becomes 16 Gbps of payload.

See http://en.wikipedia.org/wiki/8b/10b_encoding and various PCIe pages.

In any event, it should be obvious that PCIe 2.0 x4 over T-Bolt can't be faster than PCIe 2.0 x4 native.

 

[rodolfo7171]

I'm not sure if it was mentioned yet or not but what about an enclosure like this?

http://www.magma.com/expressbox-3t

 

[nando4]

T-Bolt 1 was 8 Gbps PCIe - so 16 Gbps is a doubling.

PCIe 2.0 uses an 8b/10b bit encoding - 8 bits of payload have 2 additional bits of error detection/correction.

That's 20% overhead - so 10 Gbps on the T-Bolt wire is 8 Gbps of payload for PCIe. 20 Gbps becomes 16 Gbps of payload.

See http://en.wikipedia.org/wiki/8b/10b_encoding and various PCIe pages.

In any event, it should be obvious that PCIe 2.0 x4 over T-Bolt can't be faster than PCIe 2.0 x4 native.

Not true. TB 1.0 is x4 2.0 (16Gbps) pci-e link but the downstream link across the TB channel is limited to 10Gbps. TB 1.0 and x2 2.0 (8Gbps) were bandwidth tested and compared finding the former about 12.5% faster confirming it is 10Gbps.

This means too Intel will need to upspec the pci-e link to x8 2.0 (32Gbps) to be able to get a full 20Gbps across it's TB 2.0 link. If they keep it x4 2.0 then it will only be 16Gbps and "20Gbps" will be false advertising.

REF: http://forum.techinferno.com/diy-e-...11-2012-gtx-660ti-@-2-2-no-opt.html#post33344

I'm not sure if it was mentioned yet or not but what about an enclosure like this?

http://www.magma.com/expressbox-3t

Starting at US$950 and you still need to buy the $36 Thunderbolt cable. Magma have dreams of big profits.

 

[rodolfo7171]

Not true. TB 1.0 is x4 2.0 (16Gbps) pci-e link but the downstream link across the TB channel is limited to 10Gbps. TB 1.0 and x2 2.0 (8Gbps) were bandwidth tested and compared finding the former about 12.5% faster. Meaning Intel will need to upspec the pci-e link to x8 2.0 (32Gbps) to be able to get a full 20Gbps across it's TB link. If they keep it x4 2.0 then it will only be 16Gbps.

REF: http://forum.techinferno.com/diy-e-...11-2012-gtx-660ti-@-2-2-no-opt.html#post33344



Starting at US$950 and you still need to buy the $36 Thunderbolt cable. Magma have dreams of big profits.

lol yea the price is ridiculous. you would think they would throw in the thunderbolt cable with what they charge. but that model has 3 PCIe ports. they have smaller cheaper model that houses 1 PCIe port, which would be good for this scenario since you are using one graphics card. That model is also almost 500 Bucks
http://www.magma.com/expressbox-1t

 

[AidenShaw]

REF: http://forum.techinferno.com/diy-e-...11-2012-gtx-660ti-@-2-2-no-opt.html#post33344

That link is to random chatter on a fan channel, with lots of misinformation and confusion.

If you think that I am wrong, please cite Anandtech, Tom's Hardware, Intel, or some other reliable source.

Not true. TB 1.0 is x4 2.0 (16Gbps) pci-e link but the downstream link across the TB channel is limited to 10Gbps. TB 1.0 and x2 2.0 (8Gbps) were bandwidth tested and compared finding the former about 12.5% faster confirming it is 10Gbps.

Link?

Does it not seem odd that a link specified for a certain number of Gt/s can be outperformed by the same link on an interface saddled with additional overhead and latency?

 

[minchanaooqq068]

first time i heard of the tb->pcie box i was amazed and couldnt wait to buy one, but nothing came and i forgot about it. then silverstone presented their box, i was still amazed but thought *its about time*... and damn nothing came.

 

[nando4]

That link is to random chatter on a fan channel, with lots of misinformation and confusion.

If you think that I am wrong, please cite Anandtech, Tom's Hardware, Intel, or some other reliable source.

Why Anandtech, TH, Intel or anyone else for that matter? For the attention challlenged, there are two CUDA-Z outputs reports showing real bandwidth at:

http://forum.techinferno.com/diy-e-...11-2012-gtx-660ti-@-2-2-no-opt.html#post33344

Top one is 10Gbps TB and bottom one in the spoiler is x2 2.0. 10Gbps TB works out practically to be ~12.5% faster than x2 2.0. There it's actually a x4 2.0 pci-e link to the device then 10Gbps across the TB channel. As oripash summarizes at http://forum.techinferno.com/diy-e-...-2012-gtx-660ti-@-2-2-no-opt-2.html#post33450

The TH05 system has a thunderbolt pipe (10Gbit) and a PCIe 2.0 x2 controller (8Gbit) hanging off it. It bottlenecks on the PCIe (at 8Gbit/s).

The Sonnet system has a thunderbolt pipe (10Gbit) and a PCIe 2.0 x4 controller (16Gbit) hanging off it. It bottlenecks on the thunderbolt (at 10Gbit/s).

Seems like the easiest way to explain the 10% speed difference

This was in response to your claim that TB 1 is 8Gbps, which is incorrect. Your assertion that TB 2.0 is 16Gbps is yet to be proven as we don't have it yet to test. If indeed Intel use a x4 2.0 pci-e link then it will be 16Gbps. If they uprate it to x8 2.0 then we can conclude they're giving a 20Gbps TB channel.

As much as I'd like 20Gbps TB 2.0 I've got the feeling that they'll actually give us 16Gbps. Why? Vendors like Magma, Sonnet would need to rework their existing TB boards to add an extra 4 lanes to get a x8 2.0 pci-e link, costing them time and money to do it. More likely they'll be giving new TB 2.0 chips that are pin-compatible with their TB 1.0 ones and just solder them on instead.

Intel have not exactly been generous in giving users pluggable bandwidth for their notebooks.

 

[moxin]

pokes Apple, "Watch and learn"

and then

Take my money.

 

[AidenShaw]

As oripash summarizes at http://forum.techinferno.com/diy-e-...-2012-gtx-660ti-@-2-2-no-opt-2.html#post33450.

Thank you for the interesting link, it does make it look like I was wrong.

For that performance improvement, it seems that T-Bolt must be compressing the PCIe data stream, then expanding it. (In other words, removing the 8b/10b encoding from the PCIe stream and then restoring it.)

 

[nando4]

Thank you for the interesting link, it does make it look like I was wrong.

For that performance improvement, it seems that T-Bolt must be compressing the PCIe data stream, then expanding it. (In other words, removing the 8b/10b encoding from the PCIe stream and then restoring it.)

Thunderbolt 1.0 is receiving a 16Gbps x4 2.0 pci-e stream using 4 TX/RX pairs (lanes), applying some signal processing and re-transmitting using 2 TX/RX pairs along the Thunderbolt channel. We can conclude this based on the TB pinout http://en.wikipedia.org/wiki/Thunderbolt_(interface)

Thunderbolt 1.0 has 2x10Gbps channels, 10Gbps dedicated for pcie and 10Gbps for displayport. No way of aggregating them.

Thunderbolt 2 achieves it's 20Gbps with aggregation. Same link tells us "At the logical level, Thunderbolt 2 enables channel aggregation, whereby the two previously separate 10 Gbit/s channels can be combined into a single logical 20 Gbit/s channel."

I'm guessing The TB 2.0 controller will uprate the pci-e electrical link to x4 3.0 to be able to transmit the full 20Gbps. Then manufacturers just drop in the TB 2.0 chip in their existing designs and can claim 20Gbps with pci-e 3.0 cards or 16Gbps with pci-e 2.0 cards. A cheap solution to doubling bandwidth and getting upgrade $$.

 

[phaedarus]

What primarily interests me is the possibility of connecting my 2011 Macbook Pro which uses the AMD 6770M and hooking up a modern Nvidia Geforce card to take advantage of CUDA technology natively in OSX.

What's the likelihood that this will be made a reality future?

 

[AidenShaw]

Thunderbolt 1.0 is receiving a 16Gbps x4 2.0 pci-e stream using 4 TX/RX pairs (lanes), applying some signal processing and re-transmitting using 2 TX/RX pairs along the Thunderbolt channel. We can conclude this based on the TB pinout http://en.wikipedia.org/wiki/Thunderbolt_(interface)

Thunderbolt 1.0 has 2x10Gbps channels, 10Gbps dedicated for pcie and 10Gbps for displayport. No way of aggregating them.

Thunderbolt 2 achieves it's 20Gbps with aggregation. Same link tells us "At the logical level, Thunderbolt 2 enables channel aggregation, whereby the two previously separate 10 Gbit/s channels can be combined into a single logical 20 Gbit/s channel."

I'm guessing The TB 2.0 controller will uprate the pci-e electrical link to x4 3.0 to be able to transmit the full 20Gbps. Then manufacturers just drop in the TB 2.0 chip in their existing designs and can claim 20Gbps with pci-e 3.0 cards or 16Gbps with pci-e 2.0 cards. A cheap solution to doubling bandwidth and getting upgrade $$.

Too bad that T-Bolt is a closed, secret, proprietary interconnect so that people are forced to conjecture about how it works.

This Tom's Hardware piece draws a pretty bleak picture of how T-Bolt fits into Intel's plans. http://www.tomshardware.com/reviews/z77x-up5-th-z77a-gd80-z77-oc-formula,3305-2.html

It will be interesting to see how Apple manages to feed six T-Bolt 2 ports and two workstation class graphics cards from the PCIe lanes on a single processor.