We haven't even seen a good hub yet, or even a simple drop in hard drive enclosure. I wish manufacturers would jump on this because there is demand out there.
Well, the average consumer is using a hard drive (not SSD, not a RAID array), which will run perfectly well on USB 3.0, despite its inefficiencies.
Sure, there will be T-Bolt RAID arrays (and those already exist with eSATA, through port expanders). But be prepared to spend plenty of $$$ for your niche component.
I have tested a fast 6Gb/s SATA to 5GB/s USB3 bridgeboard (Koutech ASU330) with the fastest USB3 host adapter (RocketU) and fastest 6Gb/s SSDs (OCZ Vertex3 Max IOPS, OWC Mercury Extreme Pro, etc.). The average large sequential transfer speed was 275MB/s.
I put the same SSDs in a Thunderbolt enclosure (Pegasus X4). The single SSD transfer speed was 466MB/s.
Then with a fast 6Gb/s SATA host adapter (RocketRAID 2744) in a 6Gb/s rated SATA enclosure, I got 510MB/s.
A pair of 6Gb/s SSDs in a stripe (RAID 0), USB3 = 504MB/s, Thunderbolt = 767MB/s, and 6Gb/s SATA = 1003MB/s.
Four 6Gb/s SSD in RAID 0, USB3 = 954MB/s, Thunderbolt = 855MB/s, and 6Gb/s SATA = 1864MB/s.
Not quite a fair fight in the two and four SSD RAID sets. The USB 3.0 and SATA setups had a dedicated data channel for each drive. Thunderbolt had one data channel shared by all drives.
Didn't Nvidia add a bandwidth compression system in the 258 drivers (and later) for a certain small subset of systems (Intel IGP, Optimus, x1 link)? I've seen a lot of benches from people (and in the GUS II thread someone linked a big list of DIY eGPU examples) that seem to indicate surprisingly good performance, even with higher end cards. Particularly if Nviida enabled their compression (TB more like an x2.5) it certainly looks as if, short of something SLI/multiscreen, performance could be pretty solid even with mid-range (560/6870-class say) cards (and naturally it would slaughter anything integrated). Having future CPUs that can also scale up their TDP while on mains/in a cooling dock would make it even better.
I do hope Intel changes their mind here, as long as they're maintaining backwards compatibility it'd preferable to see at least one speed bump as soon as feasible. Doubling the bandwidth would do a great deal to make this a non-issue. Even if they don't though it's still exciting.
Thunderbolt is a great technology. The idea of having a laptop/tablet to carry around, then being able to plug in a single cable at my desk and have a high speed gaming computer is quite appealing.
In my mind the biggest concern is that all controllers are coming from Intel. Until other manufacturers can produce competing TB controller, pricing will remain high, which will prevent widespread adoption.
Sony already has an implementation of this on their Vaio Z. They used a pretty mediocre GPU, which may have been due to the bandwidth limitations of this early thunderbolt interface:
From CNET: We tested the system's gaming chops both with the external GPU and without. Running only on the internal Intel HD 3000 graphics, Street Fighter IV ran at 11.7 frames per second (fps), at the native 1,920-by-1,080-pixel resolution. Connect the docking station and its AMD 6630M GPU, and we got 29fps on the same test. In the much more challenging game Metro 2033, we got 12.1fps with the external GPU, which is actually a pretty good score, considering we ran the game at 1,920-by-1,080 pixels in DirectX 11 mode with graphics set to high.
We'll probably need the full-speed version of Thunderbolt to have any chance at even mid-range gaming capabilities from an external GPU...
Better yet - you can now buy one external GPU and connect to the system that you are using with just plugging one cable. Think about it:
Buy light/small systems with integrated GPUs to do standard work. When needs be, bring out the external GPU and voila you have the performance - paying only once for the performance, instead of for each system you have just so that they have the needed performance for those few times.
This maybe crazy talk but Thunderbolt could be the new SATA. 10g on a low latency link could really work well for SSDs.
With ZFS, Windows Storage Pools and caching becoming more common .Very fast access to SSD storage is very much needed. If you have used a Tablet then you know it all about the instant access of IT and not having to wait for bootups.
Also I work in the AV industry and Fiber Thunderbolt could become the industry standard over VGA. (DVI does not look good over a 20 (60ft) meter cable.) We really need to cables that can do for 300m+ without costing thousands.
Intel have the Tech we just need other companies to get it all sorted.
The technology is not so limited, just the electronic cable. 10 Gbps is about as fast as we can expect to ever have from an electronic cable, and then only for short cable lengths. The photonic version will fix that issue. It's just that the silicon photonics chips needed to make optical Thunderbolt cheap enough to be viable are not yet ready, but they are coming.
Actually, it's just the opposite. At the moment, Thunderbolt host controllers are the limiting factor, not the cable, unless you need one longer than 3m (or 1m, I guess, since that's the longest one currently for sale.)
The only way you can currently max out a Thunderbolt cable is to daisy chain two Apple Thunderbolt displays and a Pegasus R6 full of SF-2281 based SSD's, and then perform a sequential write test using highly compressible data. This would result in 11.6Gbps of DisplayPort packets and 10Gbps of PCIe packets all heading in the same direction simultaneously and overwhelm the cable's total single-direction bandwidth of 20Gbps. This is an absolute corner case, and yet the performance impact would still be fairly minor.
Until Intel produces Thunderbolt controllers that can push more data, the electrical signaling will do just fine. In fact, current Thunderbolt ports can already be used with optical cables, but the only benefit for the increased cost would be longer cable lengths.
I've been thinking of the "one plug can carry video to display and high-speed-data to peripherals", but it just dawned on me the reverse can work, too. Just make a Thunderbolt GPU that sends it signal back through the cable to the host system. Portable graphics horsepower on the go! (It would work with the iMac, for example.)
An updated Thunderbolt (with DisplayPort 1.2 instead of DisplayPort 1.1a) is urgent and needed for "Retina" monitors: current Thunderbolt maxes out at 2560 x 1600 pixels (1280x800 @2x, that would be only a 13" Retina panel). 2014 is too far. So Intel should release an intermediate revision of Thunderbolt in 2013 with some DP 1.2 passthrough (Redwood Ridge). Two Thunderbolt ports may be used together to drive the insane resolution of a 27" Retina display.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
22 Comments
Back to Article
Dug - Thursday, January 12, 2012 - link
We haven't even seen a good hub yet, or even a simple drop in hard drive enclosure.I wish manufacturers would jump on this because there is demand out there.
mckirkus - Thursday, January 12, 2012 - link
The problem is cost. You can get a USB3 external enclosure for $14 that will handle bandwidth from all but the most expensive SSDs.USB3 = 5Gb/s
SATA3 = 6Gb/s
TBolt = 10Gb/s
iwod - Thursday, January 12, 2012 - link
You didn't factor in USB 3 has the worst bandwidth efficiency in your Three Listed IO Port.madmilk - Friday, January 13, 2012 - link
Well, the average consumer is using a hard drive (not SSD, not a RAID array), which will run perfectly well on USB 3.0, despite its inefficiencies.Sure, there will be T-Bolt RAID arrays (and those already exist with eSATA, through port expanders). But be prepared to spend plenty of $$$ for your niche component.
barefeats - Sunday, January 15, 2012 - link
I have tested a fast 6Gb/s SATA to 5GB/s USB3 bridgeboard (Koutech ASU330) with the fastest USB3 host adapter (RocketU) and fastest 6Gb/s SSDs (OCZ Vertex3 Max IOPS, OWC Mercury Extreme Pro, etc.). The average large sequential transfer speed was 275MB/s.I put the same SSDs in a Thunderbolt enclosure (Pegasus X4). The single SSD transfer speed was 466MB/s.
Then with a fast 6Gb/s SATA host adapter (RocketRAID 2744) in a 6Gb/s rated SATA enclosure, I got 510MB/s.
A pair of 6Gb/s SSDs in a stripe (RAID 0), USB3 = 504MB/s, Thunderbolt = 767MB/s, and 6Gb/s SATA = 1003MB/s.
Four 6Gb/s SSD in RAID 0, USB3 = 954MB/s, Thunderbolt = 855MB/s, and 6Gb/s SATA = 1864MB/s.
Not quite a fair fight in the two and four SSD RAID sets. The USB 3.0 and SATA setups had a dedicated data channel for each drive. Thunderbolt had one data channel shared by all drives.
zanon - Thursday, January 12, 2012 - link
Didn't Nvidia add a bandwidth compression system in the 258 drivers (and later) for a certain small subset of systems (Intel IGP, Optimus, x1 link)? I've seen a lot of benches from people (and in the GUS II thread someone linked a big list of DIY eGPU examples) that seem to indicate surprisingly good performance, even with higher end cards. Particularly if Nviida enabled their compression (TB more like an x2.5) it certainly looks as if, short of something SLI/multiscreen, performance could be pretty solid even with mid-range (560/6870-class say) cards (and naturally it would slaughter anything integrated). Having future CPUs that can also scale up their TDP while on mains/in a cooling dock would make it even better.I do hope Intel changes their mind here, as long as they're maintaining backwards compatibility it'd preferable to see at least one speed bump as soon as feasible. Doubling the bandwidth would do a great deal to make this a non-issue. Even if they don't though it's still exciting.
dcollins - Thursday, January 12, 2012 - link
Thunderbolt is a great technology. The idea of having a laptop/tablet to carry around, then being able to plug in a single cable at my desk and have a high speed gaming computer is quite appealing.In my mind the biggest concern is that all controllers are coming from Intel. Until other manufacturers can produce competing TB controller, pricing will remain high, which will prevent widespread adoption.
damianrobertjones - Thursday, January 12, 2012 - link
eGPU/ViDock... doneTrackSmart - Friday, January 13, 2012 - link
Sony already has an implementation of this on their Vaio Z. They used a pretty mediocre GPU, which may have been due to the bandwidth limitations of this early thunderbolt interface:From CNET:
We tested the system's gaming chops both with the external GPU and without. Running only on the internal Intel HD 3000 graphics, Street Fighter IV ran at 11.7 frames per second (fps), at the native 1,920-by-1,080-pixel resolution. Connect the docking station and its AMD 6630M GPU, and we got 29fps on the same test. In the much more challenging game Metro 2033, we got 12.1fps with the external GPU, which is actually a pretty good score, considering we ran the game at 1,920-by-1,080 pixels in DirectX 11 mode with graphics set to high.
We'll probably need the full-speed version of Thunderbolt to have any chance at even mid-range gaming capabilities from an external GPU...
cz - Thursday, January 12, 2012 - link
This option paves the road for an ultra light notebook with full featured docking station combo or even tablet with docking station.XZerg - Thursday, January 12, 2012 - link
Better yet - you can now buy one external GPU and connect to the system that you are using with just plugging one cable. Think about it:Buy light/small systems with integrated GPUs to do standard work. When needs be, bring out the external GPU and voila you have the performance - paying only once for the performance, instead of for each system you have just so that they have the needed performance for those few times.
Would be great for LAN parties too.
MGSsancho - Thursday, January 12, 2012 - link
Or have a PCIe slot on the back of the display and a thunder bolt like dock similar to the new apple displays.hadphild - Thursday, January 12, 2012 - link
This maybe crazy talk but Thunderbolt could be the new SATA. 10g on a low latency link could really work well for SSDs.With ZFS, Windows Storage Pools and caching becoming more common .Very fast access to SSD storage is very much needed. If you have used a Tablet then you know it all about the instant access of IT and not having to wait for bootups.
Also I work in the AV industry and Fiber Thunderbolt could become the industry standard over VGA. (DVI does not look good over a 20 (60ft) meter cable.) We really need to cables that can do for 300m+ without costing thousands.
Intel have the Tech we just need other companies to get it all sorted.
MGSsancho - Thursday, January 12, 2012 - link
http://www.matrox.com/graphics/en/products/kvm/avi... :)But as a ZFS user myself, I know exactly what you mean :D
cendrizzi - Thursday, January 12, 2012 - link
That's too bad, I was most excited for this application and didn't realize the limitations.Jaybus - Thursday, January 12, 2012 - link
The technology is not so limited, just the electronic cable. 10 Gbps is about as fast as we can expect to ever have from an electronic cable, and then only for short cable lengths. The photonic version will fix that issue. It's just that the silicon photonics chips needed to make optical Thunderbolt cheap enough to be viable are not yet ready, but they are coming.XZerg - Thursday, January 12, 2012 - link
I believe cendrizzi meant the power issue, not the bandwidth.anyhow i could see some makers providing external power to the card similar to how we have 4/6/8pin molex being attached to current high end cards.
repoman27 - Thursday, January 12, 2012 - link
Actually, it's just the opposite. At the moment, Thunderbolt host controllers are the limiting factor, not the cable, unless you need one longer than 3m (or 1m, I guess, since that's the longest one currently for sale.)The only way you can currently max out a Thunderbolt cable is to daisy chain two Apple Thunderbolt displays and a Pegasus R6 full of SF-2281 based SSD's, and then perform a sequential write test using highly compressible data. This would result in 11.6Gbps of DisplayPort packets and 10Gbps of PCIe packets all heading in the same direction simultaneously and overwhelm the cable's total single-direction bandwidth of 20Gbps. This is an absolute corner case, and yet the performance impact would still be fairly minor.
Until Intel produces Thunderbolt controllers that can push more data, the electrical signaling will do just fine. In fact, current Thunderbolt ports can already be used with optical cables, but the only benefit for the increased cost would be longer cable lengths.
MySchizoBuddy - Thursday, January 12, 2012 - link
similar to powered USB HUB you can have powered Thunderbolt HUB. Not a big issue i say.CharonPDX - Thursday, January 12, 2012 - link
I've been thinking of the "one plug can carry video to display and high-speed-data to peripherals", but it just dawned on me the reverse can work, too. Just make a Thunderbolt GPU that sends it signal back through the cable to the host system. Portable graphics horsepower on the go! (It would work with the iMac, for example.)Dug - Thursday, January 12, 2012 - link
That's what I would want!If you could add graphics power and use the same screen you already have on a laptop.
flux_capacitor - Sunday, June 24, 2012 - link
An updated Thunderbolt (with DisplayPort 1.2 instead of DisplayPort 1.1a) is urgent and needed for "Retina" monitors: current Thunderbolt maxes out at 2560 x 1600 pixels (1280x800 @2x, that would be only a 13" Retina panel). 2014 is too far. So Intel should release an intermediate revision of Thunderbolt in 2013 with some DP 1.2 passthrough (Redwood Ridge). Two Thunderbolt ports may be used together to drive the insane resolution of a 27" Retina display.