Overall impression: For a laptop, it's pretty darn amazing!
Compared to setting up with a triple head, it's smooth sailing. Each screen is recognized as an individual display, making screen arrangment and assigning via the advanced output a piece of cake. The choice to use only HDMI for the third connection is a little unfortunate, since you're forced to use a consumer connection. We had to fiddle quite a bit with our projectors before we had all three outputs recognized correctly. In all fairness that has more to do with the HDMI input on the projectors than the MBP, but it would have been better if three DVI or component connections had been available. Connecting to the computer monitors worked immediately.
Performance wise, it's quite astounding to see the amount of pixels that are being pushed without the computer complaining. Of course better performance can be gotten on high end media servers, but we're talking about a 15" laptop here. 6 layers of 1080p at 43 fps, across three monitors, while fitting in a backpack? Everyone who ever had to haul a server case across a muddy festival field or into a elevated FOH booth will recognize how awesome that is.
The fact that it includes a SSD by default makes the interface feel incredibly snappy.Triggering a column of 30 clips is gaspingly instantaneous. Deck switching is a fraction of the time as with moving drives.
Check below for all the facts and figures of this fanboy extravaganza!
One cheeseburger, please.
- 640x480 composition, 640x480 DXV content. Sent to three monitors, 2xMiniDP to DVI, 1xHDMI, 1920x1200 each. Layer 1 and 2 each get their own screen via layer routing, entire composition is sent to screen 3.
3 layers 58 fps, 34(!) layers 34-38 fps. At this point screen 3 was a big white soup and we stopped adding layers.
Wait! Make that a Big Mac, actually:
- 1920x1200 composition, 1920x1080 DXV content. Sent to three monitors, 2xMiniDP to DVI, 1xHDMI, 1920x1200 each. Same output routing as above.
2 layers 58 fps, 3 layers 55 fps, 6 layers 43 fps, 12 layers 25-27 fps, 17 layers 20-22 fps
- 5760x1200 composition, 5760x1200 DXV content. Sent to three monitors, 2xMiniDP to DVI, 1xHDMI, 1920x1200 each, one continuous image.
1 layer 49-52 fps, 2 layers 30-35 fps, 3 layers 22 fps.
But will it blend?
- 5760x1080 composition, 5760x1080 DXV content. Sent to three 1080p projectors, 2xMiniDP to DVI to HDMI (!), 1 x HDMI, 1920x1080 each. Edge blended to form one continuous screen with about 15% overlap between each projector.
1 layer 58 fps, 2 layers 39-41 fps, 3 layers 32-34 fps, 4 layers 20 fps. (The kids that are paying attention will have noticed that the test with edge blending actually ran faster than the test without. This can be accounted for by the fact that two edge blends of 15%, results in 30% less horizontal pixels that need to be calculated. Effectively this test was running at around 4000 by 1080.)
- Effects, sources, Quartz Composer patches and Flash content run fine, fps hits vary with the actual file or effect used, as is to be expected.
All tests ran for about two hours, GPU temperature stayed steady at 68-71º. The body gets hot in the usual places, but not much hotter than a late 2011 MBP with one monitor connected.
For the super wide content, the limiting factor for the fps was the disk access. When the framerate drops below 30, it was maxing out at 90%. GPU load never came above 80%. For the 480p and 1080p content, the GPU was the limiting factor, and it was maxing out at 95% when dropping below 30 fps. Disk access in these cases was not even touching 1%.
In general we felt that when the fps dropped below 30, it would start affecting performance. The output was still fluent, but the interface became more sluggish. We were pushing things quite a bit in these cases, and bringing things back to 'only' 6 layers of 1080p content brought the snappy feeling right back.
Overall CPU never strayed too far from 15%.
We used both graphic as well as photographic content. Performance was overall better when using graphic content, since DXV can compress this more efficiently, so disk access is less. One of the super wide edge blend test files can be downloaded here.
The main monitor was running at retina recommend size, which is the middle one of the scaling options (resolutions apparently are handled differently on retina MBPs). So we could have gone with a larger interface screen, but we were more interested in output performance than screen real estate on this test. This also was a comfortable viewing size for a live performance, without having to hunch over the screen more than we already would.
For temp and performance monitoring, we used atMonitor.