Samsung Portable SSD P9 Review: 4,000MB/s USB4 Reads in a Metal Shell That Peaks at 40C

extract

The Samsung Portable SSD P9 is the company’s first USB4 portable drive, rated at 4,000MB/s read and 3,800MB/s write. That’s double the T9 it sits above. It ships in 1TB, 2TB, 4TB, and 8TB capacities; the 8TB is Samsung’s first USB4 portable SSD at that size, and the 2TB unit we tested sells for $649.99 on Amazon (affiliate link), matching Samsung’s list price. The P9 arrives with portable SSDs split into two camps: 20Gbps USB 3.2 Gen 2×2 drives that have topped out near 2,000MB/s for three years, and a USB4 class that runs near internal NVMe speeds when the host has the port. We ran the P9 through the first full outing of our new external storage test, alongside the SanDisk Extreme Pro USB4 2TB, the LaCie Rugged SSD4 4TB, the Samsung T9 4TB, and the Lexar ES5 2TB. In short, the P9 delivers on its headline numbers. At low queue depth, only the SanDisk reads faster. The P9 also runs cooler under load than every other drive here. Long writes are the trade-off and where the LaCie pulls ahead. Samsung Portable SSD P9 Specifications Design and Build The P9 is a compact, slim slab at 90 x 65 x 8.3mm and 96g, noticeably thinner and lighter than the T9 at 88 x 60 x 14mm and 122g, and the bigger change is the shell. The T9 wrapped its drive in a thick rubberized cover; the P9 moves to a metal enclosure that gives up some drop forgiveness for a case that pulls heat out of the NVMe controller and NAND inside. Samsung rates the P9 for a 2-meter drop against 3 meters for the T9, a step down on paper, though 2 meters still clears a desk or a shoulder bag. That design change is likely thermal, because a USB4 drive moving 4GB/s dissipates more power than a 20Gbps drive, and rubber is an insulator. Our temperature logs back it up: the P9 peaked at 40C across the whole suite and sat at 36 to 38C through a 30-minute sustained write, while the SanDisk Extreme Pro USB4 reached 62C, the LaCie 60C, and the Lexar 70C on the same tests. The drive is well built, and for anyone who will put it to work, a few scratches on a metal case are a fair price for the headroom. The connector is USB-C, the drive supports hardware AES-256 encryption managed through Samsung Magician, and Samsung’s warranty footnote covers even the LED for five years, with typical use defined as four hours a day at 40C or below. The 8TB model shares the same dimensions as the 1TB, which puts a lot of capacity potential in a device that slips into a pocket. How We Tested This is the first review on our external storage test, a Windows suite built from the same sr-client tooling we use for client systems and pointed at one external drive. It runs nine tests against the drive in order, with a five-minute rest between each (longer if the drive is still warm), using direct I/O and incompressible data throughout. Every drive in this review was formatted exFAT and left on Windows’ Quick removal policy, which disables the device write cache, how most buyers run a portable drive. The tests are a spec-sheet check that replicates CrystalDiskMark 8’s four default rows in fio; our standard four corners at 5GiB and 25GiB footprints; a queue-depth sweep of 4K random and 128K sequential I/O from QD1 to QD32 with latency; real file copies with robocopy (a 53.7GB file, a 1,000-file camera card, and 50,000 small files, each way). The suite closes with a fixed-rate capture ladder at 25 to 1,000MB/s for 10 minutes per step; a sustained write of half the drive’s capacity (capped at 30 minutes) followed by write bursts after 5, 15, and 30 minutes idle; and a fill test that repeats the corners and a 100GB write at 0, 50, and 90 percent full. We ran the P9, SanDisk, LaCie, and Lexar on a Dell Pro Max 14 Premium (Core Ultra 9 285H, 64GB LPDDR5-8400) via its Thunderbolt 5 port. The T9 would not stay attached to that machine, so it ran on our consumer SSD test bench, a Ryzen 7 9850X3D desktop, on an ASMedia USB 3.2 Gen 2×2 port; its results are included for the generational comparison with that caveat, and it reset once during the 25GiB random write test before the suite resumed. One more note on the Lexar ES5: its exFAT volume ended up with a 512-byte cluster size, compared with 512KB to 1MB on the other drives, and the suite flagged at the start of the run that its copy and fill results would measure the format more than the drive. We have left the Lexar out of the file copy charts and treat its small-block write numbers with that in mind; its sequential and read results are unaffected. Spec-Sheet Check The spec-sheet test replicates CrystalDiskMark’s defaults (1GiB file, five 5-second passes) in fio, reporting the median pass. The P9 hit 4,048MB/s sequential read at QD8 and 3,769MB/s write, 1.01x and 0.99x of Samsung’s ratings, and 2x the T9’s 2,027MB/s read, which is the comparison Samsung makes on the box. The LaCie matched the P9 on reads at 4,048MB/s and trailed by 61MB/s on writes; the SanDisk came in at 3,811 and 3,665MB/s. At queue depth 1, where a single file copy lives, the P9 read at 2,551MB/s and wrote at 1,748MB/s. The SanDisk led reads at 2,830MB/s, and the LaCie led writes by a wide margin at 2,517MB/s, the first hint that the two USB4 competitors are tuned differently from the P9. The 4K random rows tell the same story: the P9 posted 94,933 IOPS at QD32 read and 13,674 at QD1, both ahead of the LaCie and behind the SanDisk, but its QD1 random write of 4,547 IOPS at 219us is a third of the T9’s figure and a seventh of the LaCie’s 32,387. Queue-Depth Sweep The queue-depth sweep is where a bridge chip or a slow controller shows up, because every request at QD1 pays the full round trip. On 4K random reads, the P9 averaged 97us at QD1, 101us at QD4, and 233us at QD32, matching the LaCie within a few microseconds at every step and finishing at 94,135 IOPS. The SanDisk was quicker still at 66us and 107,371 IOPS. The T9 sat at 196us at QD1, and the Lexar at 114us; both topped out below 70,000 IOPS, which is the 20Gbps generation showing its age. Random writes are the P9’s weak spot, and at QD1 it managed 4,575 IOPS at 218us, versus 31,141 IOPS at 31us for the LaCie and 13,545 at 74us for the T9, with the SanDisk close behind at 4,811 IOPS. The gap closes with depth: at QD32, the P9 reached 65,022 IOPS at 431us mean latency, 16 percent behind the LaCie’s 77,363, and its 99.99th percentile write latency was under half the T9’s 16.9ms. For a drive that will mostly see large sequential files, this matters less than it looks, but it is the first of several write-side areas where the T9 holds its own. File Copies The copy test uses robocopy with the Windows file cache flushed first and the clock running until the data is on the drive. Copying a 53.7GB file to the P9 ran at 1,992MB/s (27 seconds), compared with 3,128MB/s for the LaCie, 1,929MB/s for the SanDisk, and 1,493MB/s for the T9 on its 20Gbps host. The 1,000-file camera card hit 1,549MB/s, and the 50,000 small files hit 84MB/s, where the exFAT allocator and the P9’s QD1 write latency compound; the T9 handled the same set at 180MB/s and the LaCie at 150MB/s. Reading back is the P9’s better direction, with the large file coming off at 1,898MB/s and the camera card at 1,909MB/s, the fastest in the group on both, with the LaCie a few percent behind. The small-file read ran at 350MB/s, matching the LaCie and SanDisk and trailing the T9’s 662MB/s on its desktop host. The SanDisk’s large-file and camera-card reads landed at 405 and 416MB/s in this run, a fifth of its write speed in the same test; nothing else in its results is that slow, so we are reporting it as measured without a conclusion. Fixed-Rate Capture Cameras and recorders write at a fixed rate and drop frames when the drive falls behind, so this test writes at 25, 50, 100, 220, 500, and 1,000MB/s for 10 minutes each and passes a rate only if the average reaches 99 percent of it, no second drops below 90 percent, and no single write takes longer than half a second. Every drive held 500MB/s, which covers ProRes 422 HQ 4K60 at about 220MB/s with room to spare. At 1,000MB/s, 600GB in 10 minutes, only the LaCie matched the rate, averaging 999.6MB/s. The P9 averaged 561MB/s over the 1,000MB/s step, the lowest in the group, with the SanDisk at 709MB/s, the Lexar at 611MB/s, and the T9 at 946MB/s on its desktop host. The P9 held 500MB/s cleanly, with a slowest second of 231MB/s and a longest single write of 137ms, so a camera with a buffer of a second or so would never notice. Samsung’s materials list 12K video recording among the P9’s supported workflows; based on our numbers, that’s a safe 500MB/s recording target, and anything at or above 1,000MB/s sustained is LaCie’s territory in this group. Sustained Write and Cache Recovery The sustained test writes half the drive’s capacity in one pass, logged every second. The P9 opened at about 1,950MB/s and held it for 149GB, then dropped to a 540MB/s plateau; from minute 19, it fell into a sawtooth between roughly 240 and 500MB/s, averaging 366MB/s through the end of the 30-minute window. The drive stayed at 36 to 38C the entire time, so this is the controller flushing data from its cache, not thermal throttling. The SanDisk followed the same shape at a slightly higher level, 1,830MB/s burst to a 533MB/s plateau, and ran 24 degrees hotter doing it. The LaCie is the outlier in the other direction, writing its full 2TB test at 1,740 to 1,820MB/s with a 54GB cache and no visible steady-state penalty. The T9 wrote 1.7TB at a flat 945MB/s on its host. Samsung’s figure for the P9 is 2,000MB/s after Intelligent TurboWrite, based on internal testing, and our test measures something different: a continuous half-capacity write on an exFAT volume that Windows would not retrim between tests, after the capture ladder had already put about 870GB through the drive. The two numbers aren’t like-for-like; the useful comparison is between drives on the same test, where the P9 and SanDisk share a profile, and the LaCie doesn’t. Recovery is where the P9 pulls ahead of the SanDisk. Five minutes after the sustained write ended, the P9 pulled a full 50GB burst at 1,283MB/s on the second camera card after a coffee break; the SanDisk managed under 1GB at burst speed in the same window and 650MB/s overall. After 15 and 30 minutes, the P9’s burst settled to 27GB at 742 and 843MB/s, while the LaCie and T9 took the whole 50GB at every interval. The bursts run in that order, so each later burst follows another 50GB of writes, and we don’t have a clear explanation for why the longer idle periods performed worse. Performance as the Drive Fills The fill test repeats the corners and a 100GB write at 0, 50, and 90 percent full. Sequential reads didn’t move: the P9 read 3,990MB/s empty and 3,903MB/s at 90 percent, and QD1 random reads dropped from 10,360 to 8,553 IOPS. Sequential writes also dropped, from 3,487MB/s empty to 1,528MB/s at half full and 1,281MB/s at 90 percent, as the cache shrinks with less free space. The SanDisk held about 3,600MB/s at every level, the LaCie dropped from 3,687 to about 2,100MB/s, and the T9 halved to 948MB/s. The outlier in the fill test is the P9’s 4K random write at QD32, which came in at about 650 IOPS at every fill level with a mean latency near 195ms, against 65,022 IOPS in the queue-depth sweep earlier in the same run. The fill test is the last of the nine, after the capture ladder, the sustained write, and the fill writes themselves, and Windows could not issue a retrim to the exFAT volume between any of them. Read it as what the drive looks like after about three terabytes of writes with no idle time and no TRIM, which a heavy week of shooting and offloading could approach; a format is the reliable remedy; idle time alone did not restore it in our recovery test, and we would like to see the drive recover on its own. Conclusion The Samsung Portable SSD P9 does what the box says. It reads at 4,000MB/s and writes at 3,770MB/s on a USB4 port, doubles the T9 on sequential throughput, keeps low-queue-depth read latency under 100us, and does all of it while staying cooler than any other drive in this group, which is the benefit of the move from a rubber jacket to a metal shell. For a photographer or editor pulling camera cards and working off the drive, the read-side numbers and the 500MB/s capture result matter most, and the P9 is as good as anything in its class on both. The reservations are all on long writes: the P9’s cache runs out after about 150GB, and the drive settles to 540MB/s and then drops; it can’t sustain 1,000MB/s, and its small random writes are the slowest in the group at queue depth 1. The LaCie Rugged SSD4 is the drive to buy if sustained write speed is the job, at the cost of a bulkier, heavier package, and the SanDisk Extreme Pro USB4 matches the P9’s write profile while leading at low queue depth and running a lot hotter. At $649.99 for 2TB, the P9 looks expensive against our review-time prices for its rivals, $279 for the SanDisk Extreme Pro USB4 2TB and $479.99 for the LaCie Rugged SSD4 4TB, but both have moved with everything else in tech: the SanDisk is out of Amazon’s own stock and starts at $519 from other sellers, and the LaCie 4TB is $1,169.99 at B&H as this is written. Against those, and the T9 2TB at $436.96, the P9 sits at the top of a class that has moved up with it, and what you are paying for is the thermal design, Samsung’s software and warranty, and the fastest file-copy reads in the field. Leaderboard: The Samsung Portable SSD P9 holds Coolest-Running USB4 SSD on our Best Portable SSDs leaderboard, the first drive ranked there on our new external storage test. Samsung Portable SSD P9 at Amazon (affiliate link)