It’s no secret among those we work with that we love RED. The first digital cinema camera we used on a regular basis was the RED ONE, back when we worked together prior to starting Mystery Box. At that time we’d purchased the original REDROCKET to accelerate our R3D decodes.
Since then we’ve owned the Epic MX, Epic Dragon, Weapon Dragon 6K, Weapon Helium 8K, and the Raven, and used both the Scarlet MX and Scarlet Dragon - which covers almost the whole gamut of RED cameras. Needless to say, we’re fans. We love the compressed RAW format, the quality of image we can get for the low cost of the equipment, and its (relatively) compact size.
And yet, with all of these camera’s here at Mystery Box, we’ve never bought our own REDROCKET or REDROCKET-X. On occasion we’ve borrowed a REDROCKET for projects here or there and we regularly discuss whether we should get one or not. But we haven’t. Even after the upgraded REDROCKET-X was released in 2013, we were still on the fence as to whether it would actually accelerate our workflows.
But instead of arguing about what-ifs and maybes, we decided to use a couple of days near the end of last year to really put it to the test. We borrowed a friend’s REDROCKET-X and two full days of testing later, we had our results.
The TL;DR version of our results is that the the value of a REDROCKET-X depends significantly on your workflow. For some it’s definitely worth it, while for others (including us) it’s far less so.
Specifically, you should consider a REDROCKET-X when your workflow demands 1. Real-time or faster R3D decoding and 2. The bottleneck / choking point is the actual decoding process, and not another point in the workflow.
CPUs, GPUs, and the REDROCKET-X
REDRAW is an interesting image storage format, when compared to other ways of storing sequential images. Most RAW image formats use run length (RLE), Huffman coding, or LZW lossless compression for reducing the data size on their pre-debayer footage. These aren’t very efficient though. REDRAW, on the other hand, uses a lossy wavelet-based image compression algorithm directly on its RAW data.
While the actual mechanics of wavelet compression and how they’re implemented in REDRAW are well beyond what we want to talk about here, what we need to know about them is that they trade off complexity for data size. Compared to other image compression methods they’re really complicated to decode, but render either a better quality than something like the Discrete Cosine Transform (used in ProRes, DNxHD, H.264, etc) at the same data rate, or match the same quality as the DCT, but with smaller file sizes.
Both your computer’s CPU or GPU can be used to decode REDRAW, and how much power those processors have are the main limiting factors for how long it takes to render a single frame. If you don’t want to use either of those, or if they are significantly slower than you’d like, you can add a dedicated processing unit to handle the decode, like the REDROCKET / REDROCKET-X. So what are the tradeoffs of using each type of processor for decoding?
A computer’s CPU is built to handle a wide variety of tasks. While we often think of them as the primary indicator of a computer’s performance at tasks like web browsing or photo editing or any other program, they also handle all of the low-level processes for things like the core operating system - interpreting keypresses, managing disk access, etc. The fact that they can handle this kind of diversity makes them incredibly versatile, but not very efficient at doing lots of the same complicated task over and over again.
Essentially CPUs are built to handle lots of different kinds of tasks in sequence, which is why they sport high clock speeds (in the gigahertz range) but only have a few cores. GPUs on the other hand are a lot slower and have more limited sets of functions, but have a LOT more cores, usually thousands of them. And the set of functions they use is optimized for graphics operations, like scaling, color transformations, vector calculations, and discrete transform decoding and encoding. Including the math needed to decode wavelets.
While a single GPU core decodes a single wavelet slower than a single CPU core, overall, the GPU is quite fast at decoding the thousands of wavelets that make up a single frame because it has thousands of mini processors working at the same time, in parallel. Faster GPUs, higher core counts and clock speeds, and additional GPUs also help decrease decoding time, and with enough GPU power you can easily handle real time decoding of REDRAW at a variety of resolutions. But most computers don’t have anywhere near enough GPU power to do that. And in those cases, the best option for real time REDRAW decoding is a REDROCKET-X
At face value, the REDROCKET-X is similar to a graphics card. Both the REDROCKET-X and GPUs are built for hyperparallelism - the ability to do hundreds or thousands of small, similar tasks simultaneously. The big difference is that the REDROCKET-X does only one task: decode REDRAW. That’s it. It takes the wavelets found in REDRAW and turns them into rasterized pixels with only a small amount of color processing done before sending the decoded image to the host computer or an externally attached device. And because it’s built exclusively for REDRAW, it easily crushes R3D decoding in faster than real-time in almost every case.
But it’s still not right for every workflow.
Where it Works
The real value of the REDROCKET-X is its portability. You can attach it to a laptop using a Thunderbolt PCIe expansion box and suddenly give any computer R3D decoding superpowers. For location and destination shooting with small crews, where you’re keeping transportation and accessory costs to a minimum, you can turn almost any laptop into a DIT powerhouse in your hotel room at night and churn out proxies and quick grades as fast as you can store them, or work on real time editing in the field with the R3Ds themselves.
This type of speed and accessibility cuts turnaround time on location down by a lot. No waiting, just start cutting. This is the type of shooting and workflow that our friend who lent us his card does, and at our recommendation he’s pulled the card from a desktop and started bringing it with him to locations to cut down time on set.
But it’s not just run-and-gun type shoots that can benefit. For smaller post houses using all-in-one iMac type computers for editing and turnkey solutions instead of more powerful MacPro or Windows based towers, the REDROCKET-X is a powerful addition that can cut time costs and save money. DIT carts built on the card can get away with significantly less powerful computers while offering the same quality of service. In these cases, if your cost-benefit analysis outweighs the hefty price tag of the card, you should get one.
Where it Doesn’t Work
As soon as we start talking price for benefit, we uncover the biggest drawback of the REDROCKET-X, and the reason we don’t use it in our day-to-day workflows (though still recommended it for the use cases above!). With an MSRP of $6,750, plus the need for a powered external thunderbolt box (the main one available being the Sonnet Echo Express III-D at $900 MSRP), actually deploying a REDROCKET-X is quite pricey. Especially when you compare it to the alternatives.
Specifically, modern graphics cards.
The tests we did last December placed the decoding power of the REDROCKET-X at about 10% less powerful than an NVIDIA Quadro K6000 graphics card for decoding REDRAW, and about as fast as the Dual AMD D700s in the MacPro. Keep in mind that the Quadro K6000 is a 2-3 year old card, but will only run you about $2,000 used today. The more recent Quadro M6000 and P6000 are even more powerful, with the P6000 having about 2.5x the power of the cards I used for testing, but with only a $4,500-$5,000 price tag. A brand new 12-Core MacPro with Dual D700 graphics cards only runs about $5,200.
With the REDROCKET-X you’re essentially buying 4-5 year old technology with a price tag greater than a modern graphics card that has considerably more power. That’s why for the use cases mentioned above, portable or all-in-one type workstations where you can’t add more graphics power, the REDROCKET-X is a fantastic choice to unlock the workstation’s potential.
But if you’re using a Windows tower or a MacPro with D700s, you’re better off without it - you’ll get a far greater bang-for-your-buck with a new graphics card for your tower, or simply equal performance to the REDROCKET-X with the dual D700s out of REDCINE-X Pro.
Why that happens goes back to the original choice that the computer makes for selecting how to decode REDRAW, specifically in REDCINE-X. REDRAW is decoded on either the CPU, the GPU, or the REDROCKET-X, it can’t use more than one type at the same time. Which means that the REDROCKET-X is competing to the decode power of one of those other types, which is tough because if you’re using the GPU method it’ll be handled on as many GPUs as you have installed. And if the combined power of the GPUs is bigger than the REDROCKET-X’s, you’ll see a render speed penalty if you switch to the dedicated card.
On the other hand, programs that can leverage both the GPU and the REDROCKET-X do see some benefit with having the card installed. DaVinci Resolve, for instance, did see a performance boost when adding the REDROCKET-X to our dual K6000s, but that performance boost was quickly reduced the more correction nodes were added to the image, since the limiting factor was the speed of the color processing rather than the speed of R3D decoding.
Rendering DPX (essentially, uncompressed video) was essentially on par with all systems since that wasn’t limited by how quickly we could decode the REDRAW, but by the write speed of the RAIDs we’re using; generating ProRes or DNxHR proxies at various resolutions tended to be limited by the CPU power more than any other factor.
And since the majority of our workflows involve rendering R3Ds to DPX files, using REDCINE-X or DaVinci, adding a REDROCKET-X doesn’t help us here at Mystery Box. But that might not be the case for you.
And on a quick side note: when we ran our tests with the REDROCKET-X, it didn’t work with 8K footage in DaVinci Resolve 12.5.2-12.5.5: The footage simply appears as “media offline” when 8K runs through the REDROCKET-X. This may have been fixed in DaVinci Resolve 14, but we haven’t tested it.
The REDROCKET-X isn’t for everyone, but for individuals who need portable or flexible workstations it’s a fantastic piece of hardware. Despite its age, RED is constantly updating its capabilities through firmware enhancements and is finally now unlocking its full potential. And if after reading this article and you think it’ll help, it probably will and is worth testing to see how it can work for you.
Written by Samuel Bilodeau, Head of Technology and Post Production