Thunderbolt Performance

The Eagle Ridge Thunderbolt controller is home to two Thunderbolt channels, each one is good for up to 10Gbps in either direction (up or downstream). That works out to be 20Gbps of bandwidth per channel or 40Gbps aggregate between the two. You can only send two channels worth of data down a single Thunderbolt cable, so there's no point to having more than two from a performance standpoint unless you have more than one port on your system.

If DisplayPort and PCIe traffic are indeed carried on separate channels, then the Thunderbolt Display by itself is eating up around 70% of the bandwidth of a single channel on its own (2560 x 1440 x 32bpp x 60Hz with 8b/10b encoding > 6.75Gbps). That leaves 10Gbps in each direction for PCIe traffic. If we look at our benchmarks from the previous section we see that we can hit just under 2Gbps with all of the auxiliary interfaces (GigE, FW800, USB2) running. Given our previous investigation with the Promise Pegasus we know that 8Gbps is feasible there as well. It's possible, that with everything running at once, we could actually run into bottlenecks with Thunderbolt.

To find out I ran a few tests. First I needed a baseline so I threw four SF-2281 SSDs into the Pegasus R6 chassis and configured them in a RAID-0 array. I ran a 2MB sequential read test (QD=16) and measured 909MB/s from the array. This value was obtained without the Thunderbolt Display connected, only the Pegasus R6.

Next I connected the Thunderbolt Display directly to my test MacBook Pro, and then connected the Pegasus to it. I repeated the test, this time getting 900MB/s. Thankfully the presence of the Thunderbolt Display doesn't seem to impact the max data rate I can get from the Pegasus.

For my third test I added a Gigabit Ethernet transfer from a file server to a local SSD using the GigE port on the display. During this test I was also playing back music using the Thunderbolt Display's internal audio codec and speakers. I re-ran the Pegasus test and got 855MB/s.

For my final test I re-ran the third test but added a FireWire 800 to USB 2.0 SSD transfer, both connected to the Thunderbolt Display. I also fired up the FaceTime HD camera on the display using Photo Booth and left it on during the test. The final performance score from the Pegasus was 817MB/s.

Apple Thunderbolt Display Performance

With everything running Thunderbolt performance took a 10% hit. Note that the standard Pegasus configuration isn't able to hit these data rates to begin with, so unless you've pulled out the 12TB of storage and stuck in your own SSDs you won't see any performance drop.

What this does tell me however is the ultra high end users that are looking to daisy chain multiple Thunderbolt storage boxes together may not want to do so. I only have a single Pegasus R6 on hand, but I'm guessing there will be significant performance drop off after the first box. Not that I'm complaining about being able to push nearly 1GB/s over a $49 cable from a notebook, I'm just trying to give a heads up to those who may have aspirations of even higher performance.

Testing the Pieces Display Testing - Color Quality & Uniformity
POST A COMMENT

275 Comments

View All Comments

  • Boopop - Friday, September 23, 2011 - link

    OK, so my Dell 2408WFP isn't as big, but in most if not all the tests it outperforms this new monitor. If I was a graphic designer (which I'm not!), even if I had a MBA I reckon I'd stick with a higher quality monitor, and put up with the extra cables.

    On the other hand, if I was the average Joe Bloggs with a MBA, this makes a great monitor for that specific laptop. I like where Apple are going with this, it's just a shame about the lack of many USB ports, and the average screen quality.
    Reply
  • IceDread - Friday, September 23, 2011 - link

    Normally when you test a display you also test the input lag which I find very important. I could not find info about input lag in this review. Reply
  • tipoo - Friday, September 23, 2011 - link

    Not really as important to the target demographic, I think. Most people who get these will be using them for professional tools, so things like colour accuracy are more important than reaction time. If someone is buying one of these and a mac to game on, they've made a pretty bad error, lol. Reply
  • jecs - Friday, September 23, 2011 - link

    You have the general idea right.

    For professionals not demanding the highest color accuracy for print or for broadcast production yes, the Apple monitor is a good choice. That is professionals who work on content creation like internet video, corporate videos or print material among others.

    Serious print houses, photographers or broadcast professionals will choose high end specialized monitors in the range of $3000+, not in the sub $1000.
    Reply
  • JasperJanssen - Saturday, September 24, 2011 - link

    But they will generally only choose that type of monitor for *one* display, where the guy sits who does final colour correction on the output -- not for all the content creation people. (fair enough, if you're large enough that's multiple people, but it's never going to be the majority of your staff). Reply
  • Anand Lal Shimpi - Friday, September 23, 2011 - link

    We normally test input lag by driving a CRT in parallel with the display being evaluated. I didn't have a good way of doing that with a Thunderbolt display unfortunately :-/

    Take care,
    Anand
    Reply
  • JasperJanssen - Saturday, September 24, 2011 - link

    Well, keep it on the list for when the Mac pro comes out, which will hopefully have a videocard with multiple thunderbolt outputs :)

    Come to think of it -- new iMac with dual thunderbolt out and one of them through a VGA dongle? Hmno. Those are active dongles, which mess up results.

    Two PCs, using NTP or something similar to sync up their internal clocks maximally, and one driving a VGA CRT with the other driving the thunderbolt display, each displaying very precisely the current system time in a large font, plus the usual fast shutter speed photography. Your accuracy would depend on the NTP-or-similar protocol. I wonder if you can get close enough with that, over a crossover Gigabit ethernet kept free of other traffic.

    If you do get something like that running, you can also compare input lag between:
    - Display port driving Displayport display
    - Thunderbolt port driving Displayport display
    - Thunderbolt port driving Thunderbolt display
    -Thunderbolt port driving thunderbolt display switched through another TB device or display (add to the chain as possible)

    And even whether displayport mac versus thunderbolt macs are different in this respect when running over the various dongles.

    I would expect a Thunderbolt port running in Displayport mode to be very slightly slower than a real displayport, would be interesting to see if that is the case, and how it compares to a TB port in TB mode, and whether other devices on the chain affect it.
    Reply
  • JasperJanssen - Saturday, September 24, 2011 - link

    "NTP v4 with kernel mods to support it, is capable of much better than 1ms accuracy, possibly as good as 1ns. According to his article, NTP v3 is accurate to 1-2ms in a LAN and 10s of ms in WAN nets. "

    Well, since what you need is ms range, I guess this could actually work.
    Reply
  • sheh - Tuesday, September 27, 2011 - link

    I was curious about that too. Regardless of who the target audience of the monitor is, it's a new technology so I'm curious about its performance vs. DP/DVI/VGA. But could be difficult to tell apart from the panel's logic own latency, at least until there are more TB displays. Reply
  • MrJim - Friday, September 23, 2011 - link

    The Youtube-video in this article, http://www.youtube.com/watch?v=LtAgkIE42jc&fea... , is private. Hard to see then :) Reply

Log in

Don't have an account? Sign up now