How much storage does this shoot need?

What the tool answers

The question on the prep call is rarely “how big is ProRes”. It is: how many cards does the camera team need, how many shuttle drives does production buy, and does the DIT go home at midnight or at three. The tool works that out from what actually gets recorded — every camera with its own codec, resolution, frame rate and hours of footage per day.

The chart stacks the cameras day by day and draws a line wherever a shuttle drive is full, so you can see on which day the first drive goes to post. Below it every backup copy gets its own row of drives. The Sound tab sizes the recorder side, the Delivery tab the files that leave the edit.

How we calculate

Three kinds of codec behave differently. ProRes, DNxHR, XAVC-I and XF-AVC store a near-constant amount per frame, so the rate scales exactly with pixel count and frame rate: 50 fps is twice 25 fps, DCI 4K is 6.7 % more than UHD. ProRes figures come straight from Apple’s white paper. RAW formats are calculated from bytes per pixel times the pixels of the camera’s recording mode. Long-GOP codecs such as H.264 or H.265 at a set bitrate stay the same size whatever the frame rate.

All sizes are decimal, like the label on a card: 1 TB is 1,000 GB. Drives per copy include 10 % headroom. Offload time uses the sustained write rate of the slowest target, because Silverstack, ShotPut Pro and Hedge write every copy at once and wait for the slowest one; with verification on, every byte is read a second time.

GB per hour (frame-constant) = MB per UHD frame × (pixels ÷ 8,294,400) × fps × 3600 ÷ 1000 GB per hour (RAW) = pixels × bytes per pixel × fps × 3600 ÷ 10^9 offload per day (min) = GB per day × 1000 ÷ MB/s ÷ 60 × (2 with verification)

Where the tool is wrong

• Hours of footage are a guess. They swing more than any codec choice: a dialogue scene covered from three sizes with long takes produces twice the footage of the same scene shot in short pieces. Ask the director or DoP how they cover, and check a past job’s shooting ratio.
• RAW varies with the image. ARRIRAW is uncompressed and constant; REDCODE, X-OCN, BRAW and ProRes RAW are compressed and come in a little under the nominal value with calm images, close to it with grain, foliage or noise. We plan with the nominal value, because the card has to survive the busiest take.
• Crop and proxy modes. A 16:9 or anamorphic mode records fewer pixels than the full sensor. If the camera also writes proxies to a second card, those files are not in the numbers — add them as an extra row in a small codec such as ProRes Proxy.
• Offload speeds are field values. Card readers, cables, a transcode running at the same time or a nearly full HDD can halve the rate. Measure your own station once with a full card and enter the slower target.

From the schedule

Two ARRI Alexa 35 in 4.6K Open Gate ARRIRAW at 25 fps, each rolling about three hours a day, four shoot days. That is 2.13 TB per camera hour — a 2 TB Codex drive is full after 56 minutes — and 12.8 TB a day, 51 TB for the job, 154 TB with three copies: eight 8 TB drives per copy. Copying to NVMe SSDs over USB at 850 MB/s with verification takes 8 h 22 min a day: the DIT would still be offloading at breakfast.

Three levers fix it. A Thunderbolt RAID at about 1,400 MB/s brings it down to 5 h 5 min; offloading every full drive during the day leaves 19 minutes after wrap. Shooting the scenes that don’t need the full sensor in 4K 16:9 saves a third (1.38 TB per camera hour). Both decisions belong in prep, together with the drive budget — not on day two, when the second shuttle set is already in use.