Coming soon to an NLE near you: real-time 8K Redcode Raw workflow on a single-CPU, single-GPU desktop workstation.
At SIGGRAPH this month, most of the news around Nvidia’s Turing GPU architecture had to do with AI-driven advances that promised to make ray-traced rendering faster, and more interactive than ever before for applications in VFX and animation, virtual cinematography, VR, and more. But Nvidia said the new GPUs are also set to make real-time 8K workflow more accessible than ever, cutting the price of an workstation that can slice through 8K footage like butter by more than half.
“Red is passionate about getting high-performance tools in the hands of as many content creators as possible,” said Red Digital Cinema President Jarred Land in a prepared statement published in an Nvidia blog entry. “Our work with Nvidia to massively accelerate decode times has made working with 8K files in real time a reality for all.”
Red has enabled the option to offload the most computationally intensive portion of Redcode Raw workflow — specifically debayering and wavelet decoding of the encoded footage — to the GPU for CUDA-based processing. This will improve performance on any CUDA-enabled Nvidia GPU, not just the new generation of hardware, though users with the fastest Turing-based GPUs may see the most dramatic benefits.
In a post to reduser.net, Land said Red had recorded 24fps real-time 8K performance on a workstation equipped with a single $4,000 Nvidia Quadro P6000. But later in the thread, he also noted that some current CUDA cards were reaching real-time performance in testing. (Depending on a given system configuration, real-time effects layers may be possible as well, he said.) Further, consumer-grade cards like the upcoming $1,199 RTX 2080 Ti, with its 4,352 CUDA cores (compared to 4,608 in the P6000) may handle 8K nearly as well as their pricier Quadro counterparts.
An equivalent OpenCL solution may be available “in a year or two,” Land said, while emphasizing that CUDA is the current development priority.
According to Land, CUDA-accelerated Red workflow is expected to be integrated in NLE software starting in December 2018. It should make the Red Rocket-X, an optional $6,750 card for faster-than-real-time Redcode processing, obsolete.