RAID 0 Isn't Redundant. One Dead Drive Takes Every File.

Written and maintained by the Desert Data Recovery lab team. · Updated September 10, 2026

We already wrote about the controller's Rebuild button, and about a RAID 5 that loses a second drive. Those posts assume parity or a mirror — something left to reconstruct from. This is the other common home and DIY setup: two NVMes, two SATA SSDs, or two hard drives striped for speed. RAID is in the name. Redundancy is not in the design.

Striping is not a backup, and it is not "half the files"

RAID 0 takes each chunk of the volume — a stripe, often 16 KB to 128 KB, sometimes larger — and writes the next chunk to the next disk. Read and write bandwidth go up. Usable capacity is the sum of the members. Fault tolerance is zero.

That last part is what people miss, because the word RAID sounds like protection. RAID 1 keeps a copy. RAID 5 and 6 keep parity so the set can lose a disk and stay online. RAID 0 keeps neither. The RAID service page says it in one line: failure of even one drive results in complete data loss. This post is what that looks like on a desk, not a textbook row.

The surviving member does not hold "the files that lived on that drive." There is no such pile. A document, a Lightroom catalog, a VM, a Steam library — anything bigger than a stripe — is interleaved. Filesystem metadata is interleaved too: the NTFS MFT, the GPT, the directory tree. So you do not get a mountable volume with some folders missing. You get a disk full of fragments that do not form a volume, and a second disk that is not talking.

Unraid is the opposite layout — each file lives on one data disk — which is why an unmountable Unraid member is often that disk's filesystem, not the whole array. We covered that separately. Do not apply that mental model here. RAID 0 is the stripe.

Where this shows up at home

We see the same stripe under three different stickers. The failure is identical.

• Two NVMes striped for games or video. Motherboard RAID / Intel RST / AMD RAIDXpert, or a software stripe the builder turned on because CrystalDiskMark looked better. One M.2 dies — firmware lock, PMIC, a controller that stops enumerating — and the C: or the scratch volume is gone, not "the games on SSD 2."
• Windows Storage Spaces Simple. The UI calls it Simple (no resiliency). That is striping across columns. It is not a two-way mirror and not parity. People hear "Spaces" and assume Windows is protecting them. Simple means it is not. One dead physical disk typically takes the virtual disk with it. A Windows striped dynamic disk is the older version of the same idea.
• Motherboard / firmware RAID 0. Intel Rapid Storage, AMD RAID, the "RAID" toggle in BIOS that is really fakeRAID: a driver plus a metadata header on the disks. The controller dies, a CMOS reset drops the array, or someone loads "reset disks to non-RAID" to get Windows to see the drives. The members can still be readable. The volume is not, until the metadata and order are put back correctly — or until someone initializes them and that chance is gone.

This is not an Intel Optane H10. That stick is two devices on one M.2 — Optane plus QLC — not a RAID 0 you built. Different geometry. Different mistake. If you striped two separate NVMes in RST, you are in this post.

What not to do

Do not run a rebuild experiment. On RAID 5 or a mirror, a rebuild is a write that reconstructs one disk from the others. On RAID 0 there is nothing to reconstruct from. The controller or Storage Spaces "rebuild" is a new stripe or a metadata rewrite. Same class of mistake as the Rebuild we already warn about — except here the assumption of leftover redundancy was never true.

Do not initialize the disk. Windows Disk Management offering Initialize on the survivor (or on both members after a CMOS reset) writes a new signature and partition table. That is the start of the volume you still had. GPT at sector 0 / 1 is often how we confirm a virtual reconstruction later. Overwriting it is not a diagnostic.

Do not format the survivor. "I'll just use the good SSD as a single drive" formats the fragments that are still there. Format is a write. It does not migrate the files off first, because those files were never wholly on that disk.

Do not run desktop recovery software that writes. Recuva, R-Studio, GetDataBack, EaseUS, PhotoRec pointed at the live member and told to save output onto the same disk — or onto a "repaired" volume — is a scan plus a write. Those tools are looking at stripe pieces, not a filesystem. A chkdsk / repair-volume pass is the same problem. If you already started one, stop. Do not do a second pass.

Do not reorder members or reset firmware RAID metadata to "see if it mounts." Order, stripe size, and start offset are the geometry. Guessing on the original disks is how you write the wrong assembly over the only copies.

When a lab can still do something

We do not "rebuild the RAID 0" on your hardware. We do not promise the volume comes back because one LED is still on.

The job, when it is possible, is imaging plus geometry:

• Partial stripe imaging. The member that "died" is often not a brick. Firmware lock, a dropped PCIe lane, weak heads, a controller that will talk on a tool — those can still yield a full or partial image. Every readable stripe from that disk is a hole we do not have to leave empty. A member that will not read at all leaves those stripes gone. That is the honest limit, not a slogan.
• Member order and geometry. Stripe size, start offset, which disk is first. Wrong order looks like a successful assemble and produces garbage. We test that on copies, not on your SSDs.
• Controller metadata. Intel RST / IRSTe, AMD RAID, motherboard fakeRAID headers, Storage Spaces / interleave maps, Windows dynamic-disk LDM. Sometimes both members image cleanly and the volume is still "gone" because the metadata that tied them together was reset. That is a reconstruction problem, not a dead-NAND problem. Resetting the metadata on the originals to try again is how it becomes both.

If one member is mechanically or electrically finished and will not image, files that spanned the missing stripes do not come back from the survivor alone. We will say that after evaluation. We don't advertise a success rate, and we will not invent a RAID 0 case ID. The RAID jobs in the public case log are a Synology RAID 5 and an HP SmartArray RAID 6 + RAID 5 — parity arrays. Different math. Different leftover. This post is the stripe.

What to do right now

Do not initialize. Do not format. Do not rebuild. Do not make a new RAID 0 with a spare "so the PC will boot."

1. Power it down. If this is the boot stripe, do not keep trying BIOS / RST / Windows until it mounts. Every write is a stripe you do not get back.
2. Leave the members in order. Do not swap M.2 slots or SATA ports to experiment. Label them. Photograph the RST / RAIDXpert / Storage Spaces / Disk Management screen and the motherboard RAID banner if it is still there.
3. Note what you have: two NVMe, two SATA SSD, two HDD, or a mix; motherboard RAID vs Storage Spaces Simple vs a Windows striped dynamic disk vs a software stripe you built in Disk Management. Photos are not a diagnosis.
4. Get the set evaluated. Send every member, including the one that still enumerates and any spare you already inserted. The technician on your case will tell you whether this looks like a member we can still image, a metadata / order problem, or a stripe that is gone. Free evaluation, no data, no fee.

If you are mailing it in, pack both (or all) members like any other failed drive — how to ship a failed drive safely. For the array-level work, see RAID data recovery. If the dead member is an SSD that vanished from BIOS, the closer single-drive write-up is SSD not detected — then treat the volume as this stripe, not as a single disk you can format. If the array you actually have is RAID 5 with a second disk down, that is the RAID 5 post. If the screen is offering Rebuild on a NAS with parity, start with rebuild vs. recovery.