Crucial, who makes the MX500, our all-time favorite SATA SSD, has always been lagging behind the competition slightly when it comes to M.2 NVMe SSDs. Back in 2018, they released the QLC-based Crucial P1, which impresses with excellent price/performance, but can't compete with the various high-performance TLC offerings on the M.2 storage market. This changes with the Crucial P5, which is based on a completely new controller design that's been developed in-house at Micron, Crucial's parent company. The new controller, just called "DM01B2" for the lack of a better marketing name, is a modern 8-channel design with support for TLC and pseudo SLC caching. It also has full hardware-based encryption and reports data from two thermal sensors back to the host system—nearly all other M.2 NVMe SSDs only have one. As expected, the flash chips are provided by Micron. We've seen these 96-layer B27A 3D TLC chips on many SSDs before. Last but not least, 1 GB of fast DDR4-4266 provides storage for the SSD mapping tables, which helps with random write performance.

Our synthetic test results are surprisingly unimpressive, especially random IO seems a bit low. While the maximum "up to" rates reach over 400,000 IOPS, our more real-life-oriented mix of QD1 thru QD4 is much lower. Sequential read and write rates are very good though, peaking at well over 3 GB/s for reads and writes. The 512K sequential 50% read/write speed drops a bit for queue depths above four, which I suspect to be an effect of the controller architecture—this test case barely matters for real-life usage anyway.

What's much more impressive is real-life performance, where the Crucial P5 does much better than what you would expect from the synthetic test results. On average, the drive is 5% faster than the Crucial P1, matching drives like the Intel 760p, WD Black 2018, Kingston A2000, and Samsung 970 EVO. The fastest drives in our test group, like the HP EX950 and Kingston KC2000, are 4% faster, which is not a whole lot.

Just like all other modern SSDs, the Crucial P5 has a pseudo SLC cache that operates a portion of the drive's capacity in single-bit mode, which is much faster to write to and soaks up bursts of write activity. With 93 GB, the SLC cache is medium-sized by today's standards, roughly 10% of the whole drive. Around 100 GB should be sufficient for nearly all workloads. You may see some slowdowns only while restoring a full drive backup. When the SLC cache is exhausted, performance drops to 1 GB/s, which is a good result. It's still more than what's needed when data is coming from a slower source, like a SATA drive or USB.

Thermal performance of the Crucial P5 is slightly problematic as the drive can thermally throttle both from reads and writes. The vast majority of M.2 SSDs will throttle only with writes and run unconstrained no matter the read activity. When sitting in a case with zero airflow and hit with a tough non-stop load, the P5 will gradually heat up and start throttling after 200 seconds during reads and 130 seconds with writes. That's around 400 GB of data written in a very short timeframe—something that won't happen very often, if ever. While scientifically interesting, I wouldn't call it a dealbreaker. It would still be nice to see Crucial include a small heatsink in their package to appeal to enthusiast users.

The Crucial P5 is widely available, retailing at $150 for the tested 1 TB variant. While I can see how they justify that price point due to the product's positioning, I still find it a bit expensive. The HP EX950 and ADATA SX8200 Pro are both faster than the P5 and $10 more affordable. Also worth mentioning is the Kingston A2000 at $120 (20% cheaper) with roughly the same performance. What I definitely wouldn't buy is the Samsung 970 EVO, which is way too expensive at its current price point of $170. Crucial's own P1 is only $105 and only 5% slower on average—a great deal if you can live with the lower write performance. Considering the Crucial P5 only consists of in-house components, something nobody but Samsung can achieve, I'm sure there's plenty of room for price reductions here—below $120 doesn't seem unrealistic to me.