External GPUs -- Feasible with LightPeak / ThunderBolt?

Tower

Gawd
Joined
Oct 11, 2001
Messages
840
I'm debating purchasing a MacBook Air. Assuming the rumored specs are correct, I expect it to have a Sandy Bridge CPU coupled with 4-6 GB of RAM and a (pathetic) Intel GMA HD 3000 GPU.

I love the aesthetics, and the hardware specs are sufficient, but the GPU is anemic. I'm curious about the potential of attaching an external GPU via LightPeak/ThunderBolt, using something like this: http://www.sonnettech.com/product/thunderbolt/index.html

I know ThunderBolt's performance isn't as fast as a direct-to-motherboard PCI Express interface, but I'm wondering if it'll come close--and perhaps the only downside would be loading/buffering (adding textures), the rest of that data would be primary handled on the card's internal memory?

Anyhow, I'm looking for opinions, as otherwise my aging desktop with a Radeon HD 4850 still (sadly) blows most mobile GPUs out of the water, including the GeForce 5xx series. :rolleyes:
 
Not in the short term. Maybe a year or 2 down the line but probably not with a Mac.
 
I think their would be significant driver hurdles. Not to mention you would be forced to lug around that bulky enclosure if you ever wanted to game elsewhere (if not, then that would defeat the purpose of a laptop that can game in the first place).
 
Not with a mac currently, but the new Sony Vaio Z has a external GPU soon to be released.
 
Not sure about Thunderbolt, but Expresscard can do it now: http://hardforum.com/showthread.php?t=1522180&highlight=5850+expresscard

You beat me to it. This is a great device and many people on here and elsewhere have had great success with it.

I'm not sure if it works with OS X or not, but you should be golden with Windows if you dual boot or VM.

EDIT:
Well, I'm not really sure about the upcoming Macbook Airs, but the existing ones have 1.4-1.6GHz C2D's, which would bottleneck most modern graphics cards outside of the low-end.

OP, were you wanting to play games with the GPU and if so, which games?

It's an interesting concept, but even if it did exist, I don't think the Air would up for the task. For $1000, you could get a very nice laptop or even a standard Macbook which would be quite a bit faster.
 
Last edited:
Why would you do this with an Air? It would be like buying a dodge then putting 30 grand into it trying to make it something it's not supposed to be.
 
Why would you do this with an Air? It would be like buying a dodge then putting 30 grand into it trying to make it something it's not supposed to be.


heres a better one.. its like buying a toyota prius and putting 30 grand into it to make it into a corvette.. :p

ain't going to happen and you sure as hell know it won't happen with a mac to begin with. i mean come on apple is still using core 2 duo's in their crap. even if some one released a external gpu for lightpeak/thunderbolt it won't even see use on a apple product for another 10 years.
 
When i had my laptop i looked into some of the options available and they were all pcie-x 1 or 2x bandwidth and as it seems they still are. What's the point of hooking any current or even last gen card to a laptop when it will run at an eighth of the performance of what it should do.

Maybe you can use an old 8800gt and it would be just as fast as the the newest card today - 6990/gtx590, wow sounds great for gaming... ohh wait not really.
 
To answer some of the comments, the upcoming MBA (due to be released in a week or so) will use a Sandy Bridge ULV CPU, which should be a huge improvement over the previous C2D. However, the GPU will still be poor.

I would very much like to have an extremely mobile machine that I can easily use at home, work, and everywhere in-between--but I don't want to sacrifice performance in gaming. To answer the question as to what games I play, it's primarily Warcraft 3 (dota) and Street Fighter 4, which isn't much--but I want to be able to play Dota 2 and Diablo 3 with flying colors later this year.

I will investigate the Sony Vaio Z. Unfortunately, it really seems like mobile GPUs are incredibly lacking. For example, the GeForce 540M is the best variant I can order in a Dell XPS, and that's a significant downgrade from my aging desktop Radeon HD 4850.

I know mobile introduces a lot of differences, including power tolerances, heat tolerances, etc. but I'm surprised that the performance isn't generally better (or closer) to desktop models.
 
They need to just develop E-PCI-E 3.0. Just use a small high bandwidth connector similar to Raid Controllers with their SFF8088 connector which can move literally GIGs/second of data.
 
When i had my laptop i looked into some of the options available and they were all pcie-x 1 or 2x bandwidth and as it seems they still are. What's the point of hooking any current or even last gen card to a laptop when it will run at an eighth of the performance of what it should do.

Maybe you can use an old 8800gt and it would be just as fast as the the newest card today - 6990/gtx590, wow sounds great for gaming... ohh wait not really.

1/8 of the bandwidth does not equate directly to 1/8 of the performance potential.
 
To answer some of the comments, the upcoming MBA (due to be released in a week or so) will use a Sandy Bridge ULV CPU, which should be a huge improvement over the previous C2D. However, the GPU will still be poor.

The CPU will not get any magical performance boost either. Right now Apple uses a 1.6 GHz 10w CPU in their top-end Air, and they can replace that with a 1.8 GHz 17w Sandy Bridge (combines both CPU, memory controller and GPU so 17w is about right).

That should perform around a 2.2 GHz Core 2 Duo: still pretty slow for any serious gaming (would have been fast 2 years ago).

And yes, there is a TURBO mode to 2.9 GHz, but in a tiny form-factor like the Air, it's only sustainable momentarily. Also, any game that uses more than one core gets no turbo.

I know mobile introduces a lot of differences, including power tolerances, heat tolerances, etc. but I'm surprised that the performance isn't generally better (or closer) to desktop models.

It is ALL ABOUT POWER. Power is required to create fantastic 3D games. Then power must be removed to keep the system from overheating and dying.

You're fighting power on two fronts with laptops: first, the battery can only provide you limited wattage. Most ultra-portable laptops make-due on under 10w average power consumption (about 20w peak) due to small batteries (and the push for more battery life). Once they have consumed that power, they have to get rid of it, and that's extremely hard for the following reasons:

1. Laptops are very small, with limited ventilation and surface area. These are the two major keys to heat removal, and laptops are short on both.
2. Laptop ventilation is even more limited by noise. Small fans used in laptops move very little air at a given speed, and can get loud very quickly. Loud laptops are unacceptable to most users, because they're a lot closer to your ears than a loud desktop.

Also, it used to be a bit easier to get near-desktop performance from a mobile part, but then the average power consumption of high-end GPUs jumped from around 40w to 200w on the desktop:

9800 Pro, x800 Pro used about 40w average
6970, GTX 480/580 use 200w or more average).

How do you expect card makers to distill 200w of gaming power down to just 20-40w? They don't, and that's why you can't buy a mobile chip much faster than your 4850.
 
Last edited:
That's a good point, defaultluser.

This thread basically answers my question--there is really no good mobile gaming solution.

Even if there was, it'd probably have to be externally connected through an extremely high-speed bus, and from what I can tell, ThunderBolt isn't going to suffice.

Is there any evidence to suggest otherwise--that perhaps only loading/buffering would suffer if an external GPU was connected?

That seems to be perhaps the last hope for mobile machines right now.
 
This thread basically answers my question--there is really no good mobile gaming solution.

Depends on what you define as good. A GTX580m is essentially a slightly downclocked GTX560 ti. That particular card is no slouch. Now the price premium for the 580m is kinda crazy but it does exist.

as a comparison the 580m is about twice as fast on paper as an ATi 4850.
 
When i had my laptop i looked into some of the options available and they were all pcie-x 1 or 2x bandwidth and as it seems they still are. What's the point of hooking any current or even last gen card to a laptop when it will run at an eighth of the performance of what it should do.
.

Go look up the numbers instead of making them up. You'll be very surprised at how little PCI-E bandwidth matters for GPUs. Even at 1x, you still retain ~70% of performance.
 
Can you elaborate (or link to your references) Matrices? If that's true, an externally connected GPU (via ThunderBolt) may be a reality.

To reiterate, http://www.sonnettech.com/product/thunderbolt/index.html this enclosure provides power and a PCIE x16 slot for a single card, which could hypothetically drive a stand-alone GPU.

The question of how it'd work with drivers remains, but if what you're saying about bandwidth is true, then expansion for gaming with ThunderBolt might be a very real possibility.

EDIT: This seems to back up Matrices's comments; these results seem to show very similar performance of a GPU on PCI-E x16, x8, x4 and x1.

http://www.tomshardware.com/reviews/pci-express-2.0,1915-9.html

EDIT 2: Further review of PCI Express reminds me that it's bi-directional, but so is Thunderbolt. Thunderbolt boasts 10 Gbps (1.25 GB/s) bi-directional bandwidth. Here's the breakdown in bandwidth (for comparison with the above benchmark results):

(First value represents single channel speed, bracketed value represents overall bus speed bi-directionally):

PCI Express 1x 250 [500]* MB/s
PCI Express 2x 500 [1000]* MB/s
PCI Express 4x 1000 [2000]* MB/s
Thunderbolt 1250 [2500]* MB/s
PCI Express 8x 2000 [4000]* MB/s
PCI Express 16x 4000 [8000]* MB/s

If we can assume these values to be accurate, and that PCI x4 shows nearly identical performance for GPU-related tasks and gaming (right now), it may indeed be possible to attach a GPU to a Thunderbolt connection and utilize nearly all of the potential of said GPU.

I'm doing a lot of speculative analysis here--does this seem to add up correctly? My hopes on a MBA are being renewed.
 
Last edited:
well sony vaio z does it, im sure there are others to follow

besides 4x PCIe has enough bandwidth for single GPU high end gaming cards with a max of 2-5% performance hit
 
You're right, it probably equals less than an eighth for current cards.

No, it doesn't. NVIDIA cards are more reliant on the PCI-E bus speed than AMD cards are.

But still, there are a lot of charts out there showing the decreases in performance.

Many modern AMD cards can run at 4x PCI-E 1.0 with minimal hits to performance, not a 7/8 hit.
 
You're right, it probably equals less than an eighth for current cards.

Quit jumping to unfounded conclusions. A PCIe x1 retains 75% of the capability of a PCIe x16 when using an HD 5850. If you insist I will find the link and post it here (it has been posted in this forum at least once)

During the testing a PCIe x1 slot would vary between 73% and 80% of a full PCIe x16 slot depending on screen resolution.
 
I just don't understand why you couldn't build a SFF setup for space and get a cheap lightweight laptop. I'd reckon it would only be a few hundred more than the Air. I'd only get the Air if it was vital that I not be bothered by an additional 1.5 lbs during the course of a school day
 
Is the moral of the story here that a MBA could theoretically connect to an external PCI Express enclosure via Thunderbolt and have sufficient bandwidth for gaming?

I've reiterated why I want a MBA already; thin, light, capable, sexy.. and with thunderbolt, some of the best GPU/gaming available to a laptop, period.
 
Is the moral of the story here that a MBA could theoretically connect to an external PCI Express enclosure via Thunderbolt and have sufficient bandwidth for gaming?

I've reiterated why I want a MBA already; thin, light, capable, sexy.. and with thunderbolt, some of the best GPU/gaming available to a laptop, period.

yes in theory it could but now you gotta wait and see if anyone bothers to make such a product available.
 
Is the moral of the story here that a MBA could theoretically connect to an external PCI Express enclosure via Thunderbolt and have sufficient bandwidth for gaming?

I've reiterated why I want a MBA already; thin, light, capable, sexy.. and with thunderbolt, some of the best GPU/gaming available to a laptop, period.

I think the problem lays with profitability.

Apple isn't selling the Air for "potential gamers". In fact, the Air doesn't even have an ethernet port, hence Apple selling the Air for "ultra portable users".

The Air, even the new one, is not marketed to be sold to gamers of any kind.

It could be done and the tech is there, but Apple will not do this as it is outside of their target market.

It's not a technology limitation, it's a profitability limitation.
 
yes in theory it could but now you gotta wait and see if anyone bothers to make such a product available.

It's not a technology limitation, it's a profitability limitation.

I hear what you're saying--and I agree. The caveat, of course, is that other manufacturers are making the technology to implement this. This is the external PCIE enclosure (that includes power) that would house a GPU, and (hypothetically) only needs to be connected via Thunderbolt to act as an extension of the system:

Echo Express:
http://www.sonnettech.com/product/thunderbolt/index.html
 
Is the Air going to play nice with a video input coming in from Thunderbolt?
 
You're right, it probably equals less than an eighth for current cards.
Try doing a little google search and you might get facts.;)

http://www.techpowerup.com/reviews/AMD/HD_5870_PCI-Express_Scaling/1.html
TechPowerUp said:
PCI-Express 2.0 x4 is where the Radeon HD 5870's discomfort is slightly notable, with a 5% drop, and even more surprisingly, on PCI-Express 2.0 x1, big as it seems, the performance drop is "only" 25% overall. Considering that you rob the card most of its data transfer potential, leaving only a 1/16th of the optimum bandwidth, it is still impressive that it can deliver 75% of its performance.

Hardocp did a few of these tests too, only with SLI:
http://www.hardocp.com/article/2010/08/25/gtx_480_sli_pcie_bandwidth_perf_x16x16_vs_x4x4
http://hardocp.com/article/2010/08/23/gtx_480_sli_pcie_bandwidth_perf_x16x16_vs_x8x8/
http://hardocp.com/article/2010/08/16/sli_cfx_pcie_bandwidth_perf_x16x16_vs_x16x8/
 
I hear what you're saying--and I agree. The caveat, of course, is that other manufacturers are making the technology to implement this. This is the external PCIE enclosure (that includes power) that would house a GPU, and (hypothetically) only needs to be connected via Thunderbolt to act as an extension of the system:

Echo Express:
http://www.sonnettech.com/product/thunderbolt/index.html

I did notice that the video card would have to be under 75W meaning lower end cards, or under 150W for the upgraded model allowing mid-range cards. But either's still an improvement for any non gaming laptop. Looks like external graphics cards are gaining acceptance. If I read it right its x4 rated meaning only about a 5% performance drop and more interesting is that they can be daisy chained. I dont think that means SLI/CrossfireX as designed, but maybe someone could hack together a bridge and make it so!
 
This is quite feasible, technically. I think the main limiting factor will be when Asus / MSI / Acer etc. broadly adopt Thunderbolt, because they are the most likely (IMHO) to create a generic "Video Card Dock" with a full physical 16x PCIe slot in it (probably 4x electrically) when there is enough of an infrastructure for it. These companies would likely not release such a product for just the MBA. I think we still need to see a mobile chipset with integrated Thunderbolt before it becomes mainstream, or for intel to release an extremely cheap and easy to implement thunderbolt IC.
 
Back
Top