Samsung 970 EVO 1TB M.2 NVMe SSD Review

Brent_Justice

Administrator
Staff member
Joined
Apr 23, 2019
Messages
773
Points
93
banner1-2.png




Introduction



Today we have another SSD on our test bench to review.  This will be our second Solid State Drive SSD review.  We began with a review of the XPG SX8100NP 2TB M.2 NVMe SSD.  That was our first SSD review on our website.  As such, that review did not have any comparison data at the time.  Now that we have that review in the bag, we are able to add its data to our comparisons.  As we continue to build out and do more SSD reviews we will accumulate more data we can add to our comparisons.This review today is of the Samsung 970 EVO 1TB M.2 NVMe SSD.  This is not a brand new SSD, but it is a popular and well-known consumer SSD.  The...

Continue reading...
 
For those wondering at the difference between this and the 970 EVO Plus: 64L vs. 96L (92L) TLC. You can't go by set layers as there will be dummy layers to reduce wordline program disturb etc. In any case, the program time of these two is about the same (500µs for 2-bit MLC, 200µs+ for TLC in SLC mode, 800-1100µs for TLC) so the speed difference implies Samsung went from two-plane to four-plane flash which has diminishing returns (~50% gains) and higher TLC program latency. TLC read times of ~75µs ensure maximum seq. read (3400 MB/s) even at 250GB (8 dies x 2 planes = 16-way interleaving). The 970 EVO Plus thus has higher IOPS and sequential write performance in both SLC and TLC modes. Controller is the same, both have DRAM, similar firmware, same SLC cache design (static + dynamic).

As the review mentions this uses the Phoenix controller - this is a penta-core ARM Cortex-R5 design. The R5 is the successor to the R4 and is intended for specialized use, that is it a microcontroller (RISC by the old definition) optimized for low-latency, real-time operations as you see with storage (it's also used on HDDs). Samsung built this off the R4-based UAX/UBX which was a tri-core design with cores for read, write, and host (user); the penta-core, R5 iteration is in a 2/2/1 configuration instead. The previous drive generation (960 EVO) used the similar but lower-clocked Polaris. (clocks can vary, the older designs were typically 400 MHz but tend to be 500-700 MHz now as in the SM2262/EN and Phison E12 at 575-650 and 667 MHz respectively; the SX8100 compared is a dual-core design)

0-70C is a typical operating range but this is the composite temperature, ARM controllers can get far hotter internally (125C). Actually Samsung lists two temperatures, one of which is for NAND, and yes - NAND programs better with heat. They will throttle ~70C which is most commonly invoked with sustained, sequential writes but can also derive from heavy, mixed workloads. While most NVMe drives do not require a driver, Samsung's drives do perform better with one. RAPID mode should never be used as it just adds an additional layer of RAM caching. Likewise SED (self-encryption drive) tends to not be secure.

Just some commentary I want to add to the review after looking through it!
 
Last edited:
Nice review, and I appreciate the editorialized opinion piece at the end. Makes sense and is well reasoned.
 
Become a Patron!
Back
Top