A PCIe 5.0 SSD reads sequentially about twice as fast as a PCIe 4.0 drive. In head-to-head game loading tests across generations, the gap stayed under 0.2 seconds. Both statements hold because sequential transfer is the only SSD metric that actually doubles when the interface generation changes.

Start with the published specifications. The Samsung 9100 PRO is rated at 14,800MB/s sequential read; its predecessor, the Gen4 990 PRO, at 7,450MB/s. Random performance tells a different story: 2.2 million IOPS read versus 1.6 million, and 2.6 million IOPS write versus 1.55 million. Those figures come from the manufacturer specs compiled by StorageReview.

메인보드 M.2 슬롯에 드라이브를 나사로 고정하는 손

Where Gen5 is genuinely double: theoretical bandwidth versus measured

PCIe 4.0 runs at 16GT/s per lane, PCIe 5.0 at 32GT/s. NVMe drives typically use four lanes, and everything since PCIe 3.0 uses 128b/130b encoding. That gives:

  • PCIe 4.0 x4 = 16GT/s × 4 × (128/130) ÷ 8 = about 7,877MB/s
  • PCIe 5.0 x4 = 32GT/s × 4 × (128/130) ÷ 8 = about 15,754MB/s

Layering the published sequential read figures on top and computing efficiency against theoretical maximum produces the table below.

MetricPCIe 4.0 x4PCIe 5.0 x4
Per-lane rate (GT/s)1632
Theoretical x4 bandwidth (MB/s)7,87715,754
Reference drive sequential read (MB/s)7,450 (990 PRO)14,800 (9100 PRO)
Efficiency vs theoretical (%)94.693.9
Random read (million IOPS)1.602.20
Random write (million IOPS)1.552.60

Both generations extract roughly 94% of their theoretical ceiling, meaning the interface is already close to saturated with about 6% of headroom left. Random figures, meanwhile, improve by only 1.38x on reads and 1.68x on writes. Booting, launching applications and streaming game assets all live in that random regime, not the sequential one. Interface bandwidth failing to translate into felt performance is the same trap covered in the piece on reading USB4 and Thunderbolt's 40 and 80Gbps ratings.

What doubles across a generation is interface bandwidth, not perceived speed. The remaining bottleneck usually sits outside the drive entirely.

Why the game loading gap stops at 0.2 seconds

HotHardware covered benchmarks running PCIe 3.0, 4.0 and 5.0 drives side by side in Forspoken, a DirectStorage title. Gen5 drives generally came out on top, but no generation was ever more than a couple of tenths of a second apart, and even the slowest load finished in about three seconds.

The reason lies in the transfer volume itself. The game rarely pushed past 3GB per second even at its busiest — roughly 40% of a Gen4 drive's theoretical ceiling. HotHardware attributed this to Forspoken skipping GPU decompression, the central feature of DirectStorage, leaving the CPU to unpack assets and become the bottleneck.

The ordering that actually determines load time, then, is decompression throughput first, random read performance second, sequential bandwidth last. Sequential bandwidth had already cleared the requirement back in the Gen3 era. Raise the generation without moving the first two stages and what remains is a rounding difference.

원룸 바닥에 놓인 데스크톱 PC 케이스 내부

Heat: 8 degrees just from the slot

What Gen5 unambiguously increased is thermal output. In Quasarzone's measurements of the 9100 PRO, idle temperatures sat at 52–56°C and sustained load pushed them to 66–68°C. The same drive ran at about 58°C in a PCIe 5.0 slot and about 50°C in a PCIe 4.0 slot — an 8-degree spread from the slot alone.

That is why Crucial recommends a heatsink on Gen5 drives. A heatsink governs not peak speed but how long that speed holds. If throttling kicks in during a long bulk transfer, the 14,800MB/s on the spec sheet is only true for the first few dozen seconds.

It also adds a step before purchase. Whether the motherboard's own M.2 heatsink physically conflicts with the drive's, and whether either clears a large CPU cooler or graphics card backplate, becomes a build-time problem. Power draw rises too: the 9100 PRO's active read consumption runs 7.6–9.0W depending on capacity. Sizing total system headroom is covered in the piece on calculating PSU capacity from summed TGP and ATX 3.1 margin.

Endurance: how many GB per day the TBW rating allows

Endurance is set by NAND and capacity, not generation. Dividing the 9100 PRO's TBW ratings by the five-year warranty period (1,825 days) gives a daily write allowance and DWPD figure:

CapacityTBW (TB)Daily write allowance (GB)DWPD
1TB6003290.33
2TB1,2006580.33
4TB2,4001,3150.33
8TB4,8002,6300.33

DWPD comes out identical at 0.33 across every capacity, meaning the manufacturer scaled TBW strictly in proportion to size. Neither the 1TB model nor the 8TB one is relatively weaker. The absolute allowances differ sharply, though: 329GB a day on the 1TB model is within reach of downloading a few large games, while the 4TB model's 1,315GB is hard to hit under normal patterns.

Warranty ends at whichever comes first, the TBW or the term. Writing 100GB a day leaves even the 1TB model with roughly 16 years of headroom, so TBW is effectively a non-issue there. Move hundreds of gigabytes of video masters daily and capacity choice becomes lifespan choice.

원룸 바닥에 앉아 PC 케이스를 들여다보는 20대 후반 남성

Who it fits, and who can wait

Judged only on published specs and measurements, the line is reasonably clear.

  • Gen5 earns its price — 8K and RAW video editing that repeatedly reads and writes tens of gigabytes, local AI inference or training that loads whole datasets, environments where large drive-to-drive copies are routine. Sequential bandwidth is a real bottleneck here
  • Gen4 is enough — gaming, office work, web development, ordinary photo editing. Double the bandwidth never gets recovered in workloads separated by two tenths of a second
  • Reasons to wait — no PCIe 5.0 M.2 slot on the motherboard, or a slot that shares lanes with the graphics card. A Gen5 drive in that configuration runs at Gen4 speed and leaves only the heat
  • The heatsink is not optional — if you go Gen5, confirm the drive ships with one or that the motherboard heatsink covers that M.2 slot

Sources

늦은 밤 스탠드 불빛 아래 책상 위 키보드와 꺼진 모니터