A videographer uses an SSD for anything that touches the active workflow — camera media, ingest, editing scratch space, and project files — because random reads, sustained writes, and shock resistance all matter there. An HDD is for capacity-tier storage: finished projects, proxy archives, and backups, where sequential throughput is adequate and cost per terabyte dominates. The split is not about one drive being better; it is about which failure mode you can tolerate at each stage.

That distinction sounds simple until a 4 TB card offload stalls at 90 percent, or a project drive fills mid-render. The rest of this explainer goes one level deeper into why the hardware behaves this way, where the thresholds actually sit, and which assumptions break under real production load.

How each drive physically stores a frame

An SSD writes by trapping charge in NAND flash cells, organized into pages and erased in blocks. A controller maps logical addresses to physical cells and must erase an entire block before rewriting any page inside it. That erase-before-write rule is why SSD write speed collapses when a drive is nearly full or has no spare cells to shuffle data into. The drive needs free blocks as working room, not just as empty space.

An HDD writes by magnetizing regions of a spinning platter through a moving head. Sequential writes are fast because the head streams along a track; random writes are slow because the actuator must physically seek. A 7,200 rpm drive averages roughly 4–6 ms per random access, and at 120 accesses per second that is a hard ceiling no interface can lift.

Plain language: the SSD is limited by how it manages erased space, the HDD by how fast metal can move.

The numbers that decide the workflow

Camera codecs set the floor. ProRes 422 HQ at UHD runs near 880 Mbps, about 110 MB/s. Uncompressed or high-bitrate RAW can exceed 1,000 MB/s. A single stream rarely saturates modern hardware; the load arrives when you stack multicam angles, effects, and background renders.

SATA SSDs cap near 550 MB/s. NVMe drives over PCIe 4.0 reach 5,000–7,000 MB/s. A 7,200 rpm HDD sustains roughly 150–250 MB/s sequential but falls to under 2 MB/s under heavy random access. Those figures explain why an HDD feels fine copying a single file and unusable as a scratch disk.

Drive type Sequential read Random 4K read Typical use Relative cost per TB
SATA SSD ~550 MB/s 30,000–90,000 IOPS Ingest, proxies Mid
NVMe SSD (PCIe 4.0) 5,000–7,000 MB/s 500,000+ IOPS Scratch, multicam High
7,200 rpm HDD 150–250 MB/s ~75–150 IOPS Archive, backup Low
Portable bus-powered HDD 100–130 MB/s Low Field backup Lowest
RAID HDD array 400–1,000 MB/s Low per drive Nearline vault Mid-low

Cost per terabyte should be read as relative, not fixed; prices move and vary by region.

Timeline: where storage sits in a production

Stage Time window Dominant load Sensible drive
Field capture Shoot day Sustained writes from camera Fast card, not a drive
Offload and checksum 1–3 hours post-wrap Sequential write plus verify SSD
Proxy generation Same day Random read/write NVMe SSD
Edit and grade Days to weeks Random read, cache writes NVMe SSD
Delivery and review Days Sequential read SSD or fast array
Archive Months to years Sequential write, rare read HDD
Backup rotation Ongoing Sequential write HDD

The pattern is consistent: active work lives on flash, cold work lives on spinning media.

Edge cases where the rule inverts

Heat and power are the first inversion. An NVMe drive under sustained write can throttle once it exceeds roughly 70 °C, and bus-powered enclosures with no thermal path hit that wall during long card offloads. A slow, cool HDD can finish a 2 TB dump that a throttling pocket SSD abandons.

Sustained-write behavior is the second. Many consumer SSDs cache writes in fast SLC mode, then drop to native TLC or QLC speeds once the cache empties. A drive rated at 5,000 MB/s may fall to 400–800 MB/s after 100–200 GB of continuous writing. For a 1 TB ingest, that cliff is the difference between 20 minutes and an hour.

The counterintuitive professional caveat: for a single-stream 4K ProRes edit, a modern 7,200 rpm HDD is often genuinely adequate. Editors who insist on NVMe for every task are buying latency headroom they may not use. The SSD earns its place at ingest, multicam, and cache-heavy timelines, not on every seat.

Misconceptions worth retiring

SSDs do not wear out from reading. Endurance is rated in drive writes per day and total terabytes written; a 1 TB TLC drive rated at 600 TBW survives years of normal editing. The real risk is sudden controller failure, which arrives without the mechanical warning clicks an HDD gives.

HDDs are not obsolete for video. They remain the cheapest reliable cold tier and the standard for LTO-adjacent nearline storage. What they cannot do is serve as scratch space for layered timelines.

Red flags to watch for:

An SSD above 85–90 percent capacity, where write speed degrades sharply. A portable SSD with no DRAM cache or thermal pad, marketed on peak burst speed. An HDD used as a project scratch disk on a multicam timeline. A single drive holding the only copy of camera original media. Offloads performed without a checksum verification pass.

Practical actions and thresholds

Follow a three-tier rule. Keep camera originals on at least two SSDs until verified, edit from NVMe with 15–20 percent free space, and move finished projects to HDD pairs. Verify every offload with a checksum tool; a copy without verification is a hope, not a backup.

Set concrete limits. Keep scratch SSDs below 80 percent full. Replace or re-tier a drive that drops below 300 MB/s sustained during a large sequential copy. For archival HDDs, run a full surface scan quarterly and retire any drive showing reallocated sectors.

Match the drive to the stream count. One 4K ProRes stream needs about 110 MB/s; four multicam angles need roughly 440 MB/s before effects. That threshold decides whether an HDD array or an NVMe drive is the correct tool.

The plain version: buy flash for speed where you feel it, spinning disks for volume where you do not.

Limits of this guidance

These are general engineering principles, not benchmarks for your specific hardware. Controller firmware, enclosure thermals, filesystem choice, and cable quality can move real-world throughput by large margins. Codec bitrates quoted here are typical figures, not guarantees for every camera. The cost ratios shift with market conditions and should be checked at purchase time. No single article can substitute for testing your own ingest and edit pipeline under your own footage.

For any decision with financial or data-loss consequences, consult a qualified storage or IT professional who can assess your specific setup and redundancy needs.