PCIe generations for SSDs

Each PCIe generation doubles what one lane carries. A newer drive works in an older slot at the older slot's speed, so the generation is never a compatibility problem. It is a question of whether you move enough large files to reach the headroom you are paying for.

What you get from a newer PCIe generation

PCIe carries data over lanes, and a lane is one differential pair in each direction. An M.2 socket carries at most four of them. The generation sets how fast each lane signals; the lane count sets how many signal at once. What the link carries is the product of the two.

Gen 3 signals at 8 gigatransfers per second per lane, Gen 4 at 16, and Gen 5 at 32. Not all of that is your data. PCIe packs every 128 bits of payload into a 130-bit block, which costs about one and a half percent of the wire. What survives is just under 1 GB/s per lane at Gen 3, just under 2 at Gen 4, and just under 4 at Gen 5. Over four lanes, that is just under 4, 8, and 16 GB/s. Each figure is one direction; the link carries as much again the other way at the same time.

From Gen 3 on, the doubling is exact rather than approximate, and that makes lanes and generations interchangeable. Two Gen 4 lanes carry what four Gen 3 lanes carry. Two Gen 5 lanes carry what four Gen 4 lanes carry.

These are ceilings on the link, not speeds a drive delivers. A drive's rating sits below its generation's ceiling, and what you see sits below its rating. What SSD speed numbers actually tell you covers the distance between those three.

Will a newer drive work in an older slot

Yes, and in the other direction too, with no adapter and nothing to configure. When a drive and a slot come up they negotiate, a process called link training, and settle on the fastest speed both ends support and the widest link both ends have wired. A Gen 5 drive in a Gen 4 slot runs at Gen 4. A Gen 3 drive in a Gen 5 slot runs at Gen 3, and the drive appears normally either way. Generation only decides how much of what you bought you get to use.

The case that catches people is width rather than speed. An M.2 socket carries at most four lanes, but how many the board routes to it is the board maker's decision, and the connector looks the same either way. Thin laptops and second M.2 slots are often wired with two. The drive works, at half the bandwidth the full-width link would carry. M.2 sizes and SSD form factors covers reading a socket's keying, and why the key states what a socket of that type may present rather than what sits behind it.

This is where the exact doubling earns its keep. On a two-lane link, a Gen 4 drive carries what a Gen 3 drive carries on four lanes. Buying a newer generation for a narrow slot buys back what the width took away, and no more. Read the documentation for the lane count as well as the generation. A slot described as Gen 4 x2 is telling you both.

Where the extra bandwidth actually shows up

In transfers long enough to hold the link open. Copying a large file between two fast drives, restoring a backup image, moving footage off a card: those are single streams that run at whatever the slowest link in the path allows, and raising the ceiling raises them.

It does not show up where something else is the limit, and usually something else is. Small scattered reads are bounded by latency rather than bandwidth. The far end of a copy is often the constraint — a network share, a USB enclosure, a hard disk — and no slot changes what the other end supplies. A long write can drop to the drive's own sustained rate partway through and sit well under any ceiling.

Treat the ceiling as a permission rather than a promise. The drive, the workload, and the other end of the transfer decide what happens under it.

Why a faster drive runs hotter and slows itself down

Signalling twice as fast costs power at both ends of the link, and the controller marshalling it has a budget of its own. The power leaves as heat, from a package the size of a postage stamp on a board with no fan. Newer generations run hotter for that reason, and each one widens the gap.

Controllers defend themselves by slowing down when they get too hot, which is thermal throttling. A short transfer finishes before the drive has warmed up, so throttling surfaces in exactly the long sustained copies that were the reason to buy the bandwidth. Under that load a hot drive can finish behind a cooler, slower-rated one.

That is why fast drives and heatsinks travel together, and why many boards now supply an M.2 cover. Whether a given drive needs one depends on the drive and the air around it. Whether the heatsink it ships with fits your machine is a separate question, and the one that sends drives back.

When a newer generation is worth paying for

Work backwards from the transfers you actually run. If you move tens of gigabytes at a time between two fast drives most weeks, the link is the ceiling on how long that takes, and a higher one buys the time back. If your largest routine transfer is a game arriving over a download, the download is the limit.

Price is the other half. The newest generation on sale carries a premium for being newest, and the generation below it is where volume and sharp pricing sit. That relationship travels rather than staying with a number: whatever is at the top today becomes the sensible middle later.

The comparison worth making is against capacity, because the two compete for the same budget, and capacity is the harder one to revisit — changing it means moving an operating system. A larger drive one generation down usually beats a smaller drive at the top of the range. How to choose an SSD puts fit, capacity, speed, and price in the order to decide them.

If the slot you are filling is a generation or two behind, the question mostly answers itself. A newer drive works, at the slot's speed, and buying one is a bet on the machine after this one.

Every SSD we track is on one page.