Original Link: https://www.anandtech.com/show/13761/the-samsung-970-evo-plus-ssd-review
The Samsung 970 EVO Plus (250GB, 1TB) NVMe SSD Review: 92-Layer 3D NAND
by Billy Tallis on January 22, 2019 10:00 AM ESTThe first retail consumer SSDs to be updated with new 96-layer 3D NAND flash memory are the Samsung 970 EVO Plus. The 970 EVO Plus will be replacing the 970 EVO as Samsung's mainstream consumer NVMe SSD. Today we are reviewing two of the launch drives: the 970 EVO Plus at 250GB, and the 970 EVO Plus at 1TB. These two drives are Samsung's attack on both the mainstream and high capacity markets, while the new NAND has a focus more on power than driving down overall costs per GB. Performance per watt is a key focus of these drives.
Samsung's offerings for the 970 EVO Plus will range from 250GB to 2TB.
Samsung 970 EVO Plus Retail Units | ||||||
Capacity | Form Factor | Serial | MSRP | Price per GB |
||
250GB | 2280 M.2 | MZ-V7S250BW | $90 | 36¢ | ||
500GB | 2280 M.2 | MZ-V7S500BW | $130 | 26¢ | ||
1TB | 2280 M.2 | MZ-V7S1T0BW | $250 | 25¢ | ||
2TB | 2280 M.2 | MZ-V7S2T0BW (?) | April | - |
Capacities up to 1TB will be available from today, with the 2TB model launching in April. Pricing for that part has not yet been announced, however it is likely to be at a similar price density per GB as the 1TB.
Transitioning to 92-Layer (9xL) NAND
Last fall, Samsung outlined their plans for transitioning from 64L to 9xL NAND. Most of Samsung's SSDs that use TLC NAND will be updated to 9xL NAND while keeping the same SSD controllers as current products. While the consumer products will remain PCIe 3.0 for the time being, Samsung's top enterprise drives are getting an update to support PCIe 4.0. This means that the Samsung Phoenix controller that is at the heart of their retail, OEM and low-end datacenter NVMe product lines will be sticking around for another year, including in the new 970 EVO Plus and relatives like the PM981a and PM983a.
Samsung's fifth generation, 9x-layer 3D NAND was first announced at Flash Memory Summit in 2017, and mass production began in July 2018 with 256Gb TLC dies. However, last year saw NAND flash memory prices crash as good yields and high production volumes of 64L 3D NAND created an oversupply. The major manufacturers have been taking steps to slow their transition to 9xL NAND in order to avoid making that situation even worse for their profit margins, but they have not entirely halted the process. There's still strong incentive to provide annual updates to retail products, so it isn't surprising to see the 970 EVO Plus showing up now.
Samsung's 9xL 3D NAND process doesn't provide huge density increases to further drive costs down, but it does provide numerous performance and power efficiency improvements that make it a good way for Samsung to improve a high-end product like the 970 EVO. The new generation of 3D NAND upgrades the interface between the controller and NAND to Toggle-mode DDR 4.0, increasing the interface speed from 800Mbps to 1400Mbps while reducing the voltage from 1.8V to 1.2V. This gives Samsung the lead for another generation, as 96L NAND from Intel, Micron and SK Hynix will only support up to a 1200Mbps interface, and Toshiba/SanDisk 96L NAND is only at 800Mbps.
Within each NAND flash die, Samsung has made improvements to read and program latency, now down to about 50µs and 500µs respectively, improving by about 30%. Samsung is the only NAND manufacturer still increasing the layer count without resorting to manufacturing using string stacking, which Samsung has achieved by making each layer 20% thinner, but this has not affected the write endurance ratings for the drives. Samsung has notably fallen slightly behind in total layer count, with this generation only using 92 active layers compared to the expected 96 layers achieved by all their competitors.
Samsung 970 EVO Plus Specifications | ||||||
Capacity | 250 GB | 500 GB | 1 TB | 2 TB | ||
Form Factor | M.2 2280 single-sided | |||||
Controller | Samsung Phoenix | |||||
LPDDR4 DRAM | 512 MB | 1 GB | 2 GB | |||
NAND Flash | Samsung 92-layer 3D TLC | |||||
SLC Cache | Min | 4 GB | 4 GB | 6 GB | TBD | |
Max | 13 GB | 22 GB | 42 GB | TBD | ||
Sequential Read | 3500 MB/s | TBD | ||||
Sequential Write (SLC) | 2300 MB/s | 3200 MB/s | 3300 MB/s | TBD | ||
Sequential Write (TLC) | 400 MB/s | 900 MB/s | 1700 MB/s | 1750 MB/s | ||
4KB Random Read | QD1 | 17k IOPS | 19k IOPS | TBD | ||
QD128 | 250k IOPS | 480k IOPS | 600k IOPS | 620k IOPS | ||
4KB Random Write | QD1 | 60k IOPS | TBD | |||
QD128 (SLC) | 550k IOPS | 560k IOPS | ||||
QD128 (TLC) | 100k IOPS | 200k IOPS | 400k IOPS | 420k IOPS | ||
Power | Read | 5.0 W | 5.5 W | 5.5 W | TBD | |
Write | 4.2 W | 5.8 W | 6.0 W | TBD | ||
Idle | 30 mW | TBD | ||||
L1.2 Idle | 5 mW | TBD | ||||
Encryption | AES 256, TCG Opal 2.0, IEEE 1667 | |||||
Warranty | 5 years | |||||
Write Endurance | 150 TB 0.3 DWPD |
300 TB 0.3 DWPD |
600 TB 0.3 DWPD |
1200 TB 0.3 DWPD |
||
MSRP | $89.99 (36¢/GB) |
$129.99 (26¢/GB) |
$249.99 (25¢/GB) |
Launching in April |
Lately, Samsung has been giving more detailed performance specs than any other consumer SSD vendor, including precise SLC cache sizes and write performance after the cache is full, and random IO specifications at both QD1 and at unrealistically high queue depths. On the other hand, we don't have full specs yet for the 2TB model because it won't be shipping until April, but the press release suggests a few small advantages over the 1TB model. The smaller models use 256Gb TLC dies but the 2TB model needs the 512Gb parts in order to fit enough flash on a single-sided M.2 card.
Compared to the original 970 EVO, the 970 EVO Plus improves performance at every capacity and on every metric, but unevenly: the biggest gains are to write speeds, especially sequential writes. The 1TB model is now rated for 3.3GB/s writes to SLC cache compared to 2.5GB/s, and 1.75GB/s after the cache is full compared to 1.25GB/s. Random read speeds are about 20% better, and sequential reads see the smallest benefit with an increase from 3.4GB/s to 3.5GB/s; the PCIe 3.0 x4 interface is preventing anyone from doing much better than that.
It's interesting to note that the 970 EVO Plus performance specifications all now exceed those of the MLC-based 970 PRO, except when the SLC write cache is filled. In those cases, the 970 PRO is still rated for an advantage of 25–255% depending on which capacity is under consideration and whether the writes are sequential or random.
Hardly anything has changed about the construction of the 970 EVO Plus; the PCB layout is unchanged save for the adjustment of a few of the smallest passive components. The part numbers for the NAND packages have changed in one digit to reflect the new NAND generation and the date codes on all the major components are about a year newer. The label on the bottom of the drive still has a layer of copper foil as a token heatspreader while the label on the top is still just plastic.
Samsung did not provide price information for the full lineup before launch, but the $89.99 MSRP for the 250GB 970 EVO Plus is only slightly above the current street prices for the 970 EVO and matches what Samsung sells it for direct, so we don't expect any significant changes in the near term for the other capacities. In the long run however, NAND prices are still in decline and Samsung has had to respond to that. The original 970 EVO was initially announced for around 45¢/GB but this was cut to 40¢/GB before it even hit the shelves. Now, it's going for around 25¢/GB. Further price cuts to the 970 EVO [Plus] are likely over the coming months, but the transition from one model to another will cause some fluctuations depending on which one is in stock at more retailers on any given day.
Where's the 970 PRO Plus?
(Not to be confused with caffeine pills - Ian)
Conspicuously absent from Samsung's 9xL roadmap last fall was any mention of MLC-based drives, and in particular Samsung has not mentioned a 970 PRO Plus. The lack of new MLC drives for the datacenter or OEMs is no surprise given how thoroughly those markets have shifted over to TLC. But on the consumer side, Samsung has been one of the few remaining holdouts offering MLC-based SSDs for enthusiasts and prosumers with their PRO drives.
The later launch date of the 2TB 970 EVO Plus suggests that the only 9xL NAND Samsung is ready to release to the retail market is their 256Gb TLC; the 512Gb TLC and QLC parts are probably still in the sampling phase for either the memory itself, or the OEM or datacenter drives are still sampling rather than in full production. If Samsung plans a regular MLC part at 92 layers, it's almost certainly the lowest priority to move off 64L, after they get both TLC parts and both QLC parts out the door.
We can't rule out the possibility of a '970 PRO Plus' sometime this year, but it seems far more likely that we won't see another PRO NVMe drive until 2020 with a new controller that may also bring PCIe 4.0 support. Until then, the 970 EVO Plus takes over the performance flagship position for Samsung's retail/consumer product line and the 970 PRO is only relevant for workloads that involve sustained writes of hundreds of GB per day. For workloads that merely consist of tens of GB per day or include some disk idle time for the SLC cache to recover, the 970 EVO Plus is now likely to offer both better performance and better pricing than the 970 PRO.
(Samsung does plan to eventually introduce a MLC version of the second generation of their specialized low-latency Z-NAND, but they are still in the process of rolling out the SLC version, and Z-NAND's layer count is lagging behind their general-purpose V-NAND 3D NAND. Samsung has not mentioned any plans to use Z-NAND outside of their datacenter product lines. At the other end of their NVMe product spectrum, they've disclosed plans for a 980 QVO with QLC NAND, but we aren't sure when to expect that to launch, or whether it will introduce a new controller or continue with the Phoenix controller.)
This Review
For the review today, we are testing the 250GB and 1TB versions of the Samsung 970 EVO Plus.
AnandTech 2018 Consumer SSD Testbed | |
CPU | Intel Xeon E3 1240 v5 |
Motherboard | ASRock Fatal1ty E3V5 Performance Gaming/OC |
Chipset | Intel C232 |
Memory | 4x 8GB G.SKILL Ripjaws DDR4-2400 CL15 |
Graphics | AMD Radeon HD 5450, 1920x1200@60Hz |
Software | Windows 10 x64, version 1709 |
Linux kernel version 4.14, fio version 3.6 | |
Spectre/Meltdown microcode and OS patches current as of May 2018 |
- Thanks to Intel for the Xeon E3 1240 v5 CPU
- Thanks to ASRock for the E3V5 Performance Gaming/OC
- Thanks to G.SKILL for the Ripjaws DDR4-2400 RAM
- Thanks to Corsair for the RM750 power supply, Carbide 200R case, and Hydro H60 CPU cooler
- Thanks to Quarch for the XLC Programmable Power Module and accessories
- Thanks to StarTech for providing a RK2236BKF 22U rack cabinet.
Whole-Drive Fill
This test starts with a freshly-erased drive and fills it with 128kB sequential writes at queue depth 32, recording the write speed for each 1GB segment. This test is not representative of any ordinary client/consumer usage pattern, but it does allow us to observe transitions in the drive's behavior as it fills up. This can allow us to estimate the size of any SLC write cache, and get a sense for how much performance remains on the rare occasions where real-world usage keeps writing data after filling the cache.
Samsung obviously hasn't changed anything significant about the SLC caching behavior for the 970 EVO Plus: the cache initially runs out right on schedule, and the 1TB model still jumps back up to SLC speed for a short while when the drive is a little less than half full. The change to the underlying NAND does provide a performance boost to the write speeds before and after the cache fills up. The 240GB ADATA XPG SX8200 is an outlier among the small drives: it has a very large variable size SLC cache so it maintains high performance longer than the other drives in that capacity class, but once that cache is full it ends with one of the slowest write speeds.
Average Throughput for last 16 GB | Overall Average Throughput |
The improvement to post-SLC sequential write speed isn't quite as big as the 970 EVO Plus specifications promise, but it's still plenty to make it the fastest TLC drive for this test, clearly surpassing the performance of even last week's new WD Black SN750. At the low end of the capacity range, the 250 GB model is well ahead of any other small TLC drive we have tested. It also looks like we are getting quite close to the point where the post-SLC write speed of a small TLC drive can saturate a SATA link, so if Samsung releases another generation of MLC SATA drives for the consumer market it may be impossible to measure any performance advantage over TLC NAND.
BAPCo SYSmark 2018
BAPCo's SYSmark 2018 is an application-based benchmark that uses real-world applications to replay usage patterns of business users, with subscores for productivity, creativity and responsiveness. Scores represnt overall system performance and are calibrated against a reference system that is defined to score 1000 in each of the scenarios. A score of, say, 2000, would imply that the system under test is twice as fast as the reference system.
SYSmark scores are based on total application response time as seen by the user, including not only storage latency but time spent by the processor. This means there's a limit to how much a storage improvement could possibly increase scores, because the SSD is only in use for a small fraction of the total test duration. This is a significant difference from our ATSB tests where only the storage portion of the workload is replicated and disk idle times are cut short to a maximum of 25ms.
AnandTech SYSmark SSD Testbed | |
CPU | Intel Core i5-7400 |
Motherboard | ASUS PRIME Z270-A |
Chipset | Intel Z270 |
Memory | 2x 8GB Corsair Vengeance DDR4-2400 CL17 |
Case | In Win C583 |
Power Supply | Cooler Master G550M |
OS | Windows 10 64-bit, version 1803 |
Our SSD testing with SYSmark uses a different test system than the rest of our SSD tests. This machine is set up to measure total system power consumption rather than just the drive's power.
The SYSmark Responsiveness test shows the 970 EVO Plus delivering slightly higher performance than any previous TLC-based SSD, but not quite enough to reach the level of the Intel Optane SSD 900P. Oddly, the 250GB 970 EVO Plus came out a bit ahead of the 1TB model. The other two scenarios—Creativity and Productivity—don't depend on storage performance enough to even show a meaningful difference between the 860 EVO SATA SSD and the fastest NVMe SSDs.
Energy Use
The SYSmark energy use scores measure total system power consumption, excluding the display. Our SYSmark test system idles at around 26 W and peaks at over 60 W measured at the wall during the benchmark run. SATA SSDs seldom exceed 5 W and idle at a fraction of a watt, and the SSDs spend most of the test idle. This means the energy usage scores will inevitably be very close. A typical notebook system will tend to be better optimized for power efficiency than this desktop system, so the SSD would account for a much larger portion of the total and the score difference between SSDs would be more noticeable.
The 970 EVO Plus required a bit less energy to complete a SYSmark run than the original 970 EVO did, but they're still relatively power-hungry compared to other NVMe SSDs, and good SATA SSDs use significantly less power. These differences don't seem like much in the context of our desktop testbed's total power draw, but in a laptop the performance of the 970 EVO Plus does come at some cost to battery life.
AnandTech Storage Bench - The Destroyer
The Destroyer is an extremely long test replicating the access patterns of very IO-intensive desktop usage. A detailed breakdown can be found in this article. Like real-world usage, the drives do get the occasional break that allows for some background garbage collection and flushing caches, but those idle times are limited to 25ms so that it doesn't take all week to run the test. These AnandTech Storage Bench (ATSB) tests do not involve running the actual applications that generated the workloads, so the scores are relatively insensitive to changes in CPU performance and RAM from our new testbed, but the jump to a newer version of Windows and the newer storage drivers can have an impact.
We quantify performance on this test by reporting the drive's average data throughput, the average latency of the I/O operations, and the total energy used by the drive over the course of the test.
The 1TB Samsung 970 EVO Plus has a significantly higher average data rate on The Destroyer than its predecessor, or any other TLC-based SSD we've ever tested. It's about 15% slower overall than the Intel Optane SSD 900P that costs about four times as much per GB. The smallest 250GB 970 EVO Plus is little less than half as fast as the 1TB model, but that still puts it above the 1TB HP EX920 and anything else in the 250GB capacity class.
The average and 99th percentile latency scores for the 970 EVO Plus are a clear improvement over its predecessors, but it doesn't come out ahead of all the competition. The Phison E12-based Corsair MP510 outperforms the 970 EVO Plus on both metrics, and the WD Black SN750 has better 99th percentile latency.
The 970 EVO Plus has a faster average read latency on The Destroyer than any of its competition, but the Corsair MP510 still holds the top spot for average write latency thanks to its very fast write cache. Among the slightly different mix of smaller drives, the 250GB 970 EVO Plus has a much more commanding lead in average write latency than average read latency.
The 1TB 970 EVO Plus has better 99th percentile read and write latency scores than its predecessor, but it can't entirely catch up to the best competitors we've seen in recent months. The smaller drives have vastly higher 99th percentile latencies, but the 250GB 970 EVO Plus comes out ahead among that group with a small write QoS lead and a larger read QoS lead.
The 1TB 970 EVO Plus uses slightly less energy to complete The Destroyer than its predecessor. However, both capacities are still very inefficient compared to the best drives in their capacity class. At 1TB there are options that are both very efficient and very fast, but the only particularly efficient small NVMe drive we've tested is the entry-level MyDigitalSSD SBX.
AnandTech Storage Bench - Heavy
Our Heavy storage benchmark is proportionally more write-heavy than The Destroyer, but much shorter overall. The total writes in the Heavy test aren't enough to fill the drive, so performance never drops down to steady state. This test is far more representative of a power user's day to day usage, and is heavily influenced by the drive's peak performance. The Heavy workload test details can be found here. This test is run twice, once on a freshly erased drive and once after filling the drive with sequential writes.
The Samsung 970 EVO Plus does not significantly improve on the best data rates we've seen from TLC SSDs on the Heavy test, and the 250GB model in particular is way behind the 240GB ADATA SX8200. However, the 970 EVO Plus does deliver class-leading full drive performance.
(Our 1TB 970 EVO persistently underperforms expectations when the Heavy test is run on an empty drive, with unusually high read latencies. The drive does not report any media or data integrity errors, and the other 970 EVO and 970 EVO Plus drives do not exhibit this behavior.)
The 1TB 970 EVO Plus turns in great average and 99th percentile latency scores, with the averages on par with the Intel Optane 900P and 99th percentile latency that clearly beats the Optane SSD when the test is run on an empty drive. The 250GB 970 EVO Plus compares favorably with most of the other drives in its capacity class, but the ADATA SX8200 comes out well ahead when the test is run on an empty drive, at the cost of much worse full-drive performance.
Both capacities of the 970 EVO Plus turn in great average read latency scores, but the average write latencies show a very clear division between the 1TB and 250GB classes. The 1TB 970 EVO Plus leads its class for average write latency while the 250GB model is overshadowed by the empty-drive performance of the ADATA SX8200.
The 99th percentile read and write latency scores both show relatively clear separation between the 1TB and 250GB NVMe drives. The 1TB 970 EVO Plus has excellent 99th percentile write latency but somewhat worse 99th percentile read latency than most of its competition. The 250GB model again looks much better than most of its competition.
The 970 EVO Plus uses a bit more energy during the Heavy test than its predecessor, keeping it at the bottom of the list while the WD Black SN750 only needs half the energy and several other competitors are well ahead of Samsung.
AnandTech Storage Bench - Light
Our Light storage test has relatively more sequential accesses and lower queue depths than The Destroyer or the Heavy test, and it's by far the shortest test overall. It's based largely on applications that aren't highly dependent on storage performance, so this is a test more of application launch times and file load times. This test can be seen as the sum of all the little delays in daily usage, but with the idle times trimmed to 25ms it takes less than half an hour to run. Details of the Light test can be found here. As with the ATSB Heavy test, this test is run with the drive both freshly erased and empty, and after filling the drive with sequential writes.
The Samsung 970 EVO Plus has the highest average data rates on the Light test, and even the 250GB Plus is faster than the 1TB original 970 EVO. Ignoring its predecessor, the 970 EVO Plus is more than 20% faster than the next fastest competitors. The lead isn't quite as pronounced when looking at full-drive performance, but the 970 EVO Plus is still on top.
The Samsung 970s turn in the best average and 99th percentile latency scores on the Light test, though the 250GB model's 99th percentile latency is strongly affected when the drive is full. Aside from that, the 250GB model's latency is not meaningfully higher than the 1TB model, and none of the other drives in that capacity class offer such low latency.
The 970 EVO Plus tops the charts for both average read and write latency scores from the Light test. The average write latency of the 250GB model does show that the smaller drive's performance suffers when the test is run on a full drive, but the other small TLC drives are all worse off.
The 970 EVO Plus is joined by several other drives in having 99th percentile write latencies under 100µs, but the 250GB model can't maintain that when the test is run on a full drive. The 99th percentile read latencies are great, but the full-drive read QoS for both capacities of the 970 EVO Plus is nothing special for this product segment.
The high performance of the 970 EVO Plus once again comes at the cost of power efficiency, with energy usage that is significantly above the competition and twice that of the WD Black SN750. The performance potential of the 970 EVO Plus is largely wasted on a workload this light, and that leads to wasted power.
Random Read Performance
Our first test of random read performance uses very short bursts of operations issued one at a time with no queuing. The drives are given enough idle time between bursts to yield an overall duty cycle of 20%, so thermal throttling is impossible. Each burst consists of a total of 32MB of 4kB random reads, from a 16GB span of the disk. The total data read is 1GB.
The QD1 burst random read performance of the Samsung 970 EVO Plus is an improvement over its predecessor, but not as large as was promised by the spec sheet. The Samsung drives are also still lagging far behind the fastest combinations of IMFT NAND and Silicon Motion controllers.
Our sustained random read performance is similar to the random read test from our 2015 test suite: queue depths from 1 to 32 are tested, and the average performance and power efficiency across QD1, QD2 and QD4 are reported as the primary scores. Each queue depth is tested for one minute or 32GB of data transferred, whichever is shorter. After each queue depth is tested, the drive is given up to one minute to cool off so that the higher queue depths are unlikely to be affected by accumulated heat build-up. The individual read operations are again 4kB, and cover a 64GB span of the drive.
On the longer test of random read performance that also brings in some higher queue depths, the Samsung 970 EVO Plus is more competitive and comes out tied or ahead of last year's best. But not shown here are the preliminary results from our tests of SM2262EN drives (next in the review queue) which are substantially faster than the 970 EVO Plus.
Power Efficiency in MB/s/W | Average Power in W |
The 970 EVO Plus uses more power during the random read test than any other drives in this bunch, leaving it with some of the worse efficiency scores despite the good performance.
The random read performance of the 250GB 970 EVO Plus begins to fall behind the 1TB model as queue depths increase, and the smaller drive seems to be close to saturating by QD16. At QD32 the 250GB model's random read throughput is less than half that of the 1TB model. The 250GB 960 EVO delivered substantially higher random read performance at high queue depths than the small 970 EVO Plus manages.
The relatively high power consumption of the 970 EVO Plus is apparent when plotting its test results against the rest of the database. At lower queue depths it delivers speeds that are achievable by SATA SSDs but uses more power than almost all of them. Once the queue depth has increased sufficiently to take the 970 EVO Plus beyond the SATA limit, it still uses more power than most of the remaining competition.
Random Write Performance
Our test of random write burst performance is structured similarly to the random read burst test, but each burst is only 4MB and the total test length is 128MB. The 4kB random write operations are distributed over a 16GB span of the drive, and the operations are issued one at a time with no queuing.
The burst random write performance of the 970 EVO Plus takes a slight step backwards from its predecessor, leaving it far behind the best competitors, which are currently led by the Phison E12-based Corsair MP510, 67% faster than the 970 EVO Plus.
As with the sustained random read test, our sustained 4kB random write test runs for up to one minute or 32GB per queue depth, covering a 64GB span of the drive and giving the drive up to 1 minute of idle time between queue depths to allow for write caches to be flushed and for the drive to cool down.
When the longer random write test brings in some higher queue depths, the 1TB 970 EVO Plus regains a high-end standing but doesn't quite catch up to the WD Black SN750. The 250GB model is a bit further behind the lead thanks to the fast write cache of the ADATA SX8200, which performs identically to that of the related 1TB HP EX920.
Power Efficiency in MB/s/W | Average Power in W |
The 970 EVO Plus improves upon the random write power efficiency of its predecessors and ranks very highly, but the WD Black SN750 provides 26% better performance per Watt on this test, and the ADATA SX8200 beats the 250GB 970 EVO Plus by 21%.
The Samsung 970 EVO Plus reaches full random write speed by QD4 and is reasonably steady through the rest of the test. The 1TB model saturates at a slightly higher level than the WD Black SN750, but its overall score suffered due to the slower QD1 performance. The ADATA SX8200 that provides excellent low-QD random write performance at low capacities does run out of SLC cache for part of the test and drops below the 250GB 970 EVO Plus's performance level for a while before recovering.
Comparing the 970 EVO Plus against all the results in the database shows the 1TB drive starts out with QD1 performance just beyond the limit for SATA drives, and power requirements don't increase much as throughput climbs. At higher queue depths the 970 EVO Plus falls roughly in the middle of the normal power consumption range for high-end drives.
Sequential Read Performance
Our first test of sequential read performance uses short bursts of 128MB, issued as 128kB operations with no queuing. The test averages performance across eight bursts for a total of 1GB of data transferred from a drive containing 16GB of data. Between each burst the drive is given enough idle time to keep the overall duty cycle at 20%.
The QD1 burst sequential read performance of the Samsung 970 EVO Plus is a clear step up from its predecessors, but the Silicon Motion-based HP EX920 and ADATA SX8200 are still a few percent faster.
Our test of sustained sequential reads uses queue depths from 1 to 32, with the performance and power scores computed as the average of QD1, QD2 and QD4. Each queue depth is tested for up to one minute or 32GB transferred, from a drive containing 64GB of data. This test is run twice: once with the drive prepared by sequentially writing the test data, and again after the random write test has mixed things up, causing fragmentation inside the SSD that isn't visible to the OS. These two scores represent the two extremes of how the drive would perform under real-world usage, where wear leveling and modifications to some existing data will create some internal fragmentation that degrades performance, but usually not to the extent shown here.
On the longer sequential read test with some higher queue depths, the Samsung 970 EVO Plus is tied for first place when reading contiguous data. When reading back data that was written randomly and is likely to be fragmented on the NAND itself, the 970 EVO Plus's performance has regressed but is still ahead of almost all of the competition, especially the SMI-based drives that score so well on contiguous data.
Power Efficiency in MB/s/W | Average Power in W |
The 970 EVO Plus takes the clear lead for power efficiency when reading contiguous data, but doesn't stand out when the data is fragmented.
The 970 EVO Plus significantly increases sequential read performance at queue depths from 2 to about 16. It doesn't quite hit full speed at middling queue depths, but it gets much closer than its predecessors, which don't improve much on their QD1 performance until around QD16.
During the sustained sequential read test the 970 EVO Plus stays entirely within the upper half of the NVMe performance range, where the power efficiency competition isn't quite as strong. Below 2GB/s there are quite a few drives that are much more efficient than the 970 EVO, but above 3GB/s its power consumption doesn't stand out from the crowd.
Sequential Write Performance
Our test of sequential write burst performance is structured identically to the sequential read burst performance test save for the direction of the data transfer. Each burst writes 128MB as 128kB operations issued at QD1, for a total of 1GB of data written to a drive containing 16GB of data.
The burst QD1 sequential write performance of the Samsung 970 EVO Plus is slightly lower than its predecessor's, but the 1TB model is still clearly in high-end territory. The 250GB model looks quite slow compared to the two generation old 250GB 960 EVO, but that drive uses 128Gb TLC dies and no current-generation NAND is available in capacities that low. The reduced parallelism available to the modern 250GB drives prevent them from delivering the full performance that their older or larger counterparts provide.
Our test of sustained sequential writes is structured identically to our sustained sequential read test, save for the direction of the data transfers. Queue depths range from 1 to 32 and each queue depth is tested for up to one minute or 32GB, followed by up to one minute of idle time for the drive to cool off and perform garbage collection. The test is confined to a 64GB span of the drive.
On the longer sequential write test that includes some higher queue depths, the 1TB Samsung 970 EVO Plus substantially improves upon the 970 EVO's lead. The smaller 250GB 970 EVO Plus is completely outclassed by the 240GB ADATA SX8200.
Power Efficiency in MB/s/W | Average Power in W |
In line with its excellent performance, the 1TB 970 EVO Plus tops the power efficiency ranking. The smaller 250GB model uses about 1W less power than the 240GB SX8200 but ends up worse off on the power efficiency rating due to the huge performance gap.
The Corsair Force MP510 is the only drive in this batch that shows serious performance variation suggestive of the SLC cache running out intermittently. The rest of the drives are either not filling their SLC write caches entirely, or fill them part way through each phase of the test but have no trouble flushing the cache during the idle time between phases (mostly the latter). The 240GB ADATA SX8200 has such a strong advantage over the 250GB 970 EVO Plus because the cache on the Samsung drive runs out quite quickly while the SX8200's ~72GB cache is plenty large enough to handle the 32GB written during each phase of the test.
Of all the drives in our database that get close to 3GB/s during the sequential write test, none use less power than the 1TB 970 EVO Plus. There are several much more efficient drives that top out at around 2.6GB/s, but the very fastest drives all require 5-6W.
Mixed Random Performance
Our test of mixed random reads and writes covers mixes varying from pure reads to pure writes at 10% increments. Each mix is tested for up to 1 minute or 32GB of data transferred. The test is conducted with a queue depth of 4, and is limited to a 64GB span of the drive. In between each mix, the drive is given idle time of up to one minute so that the overall duty cycle is 50%.
The 1TB Samsung 970 EVO Plus slightly improves on its predecessor's performance on the mixed random IO test, to widen Samsung's lead over other TLC-based drives. The 250GB model fares reasonably well, but is again no match for the oversized SLC cache on the 240GB ADATA SX8200.
Power Efficiency in MB/s/W | Average Power in W |
The WD Black SN750 holds on to another power efficiency win and the Samsung 860 EVO SATA SSD is more or less tied with the 1TB 970 EVO Plus. Most of the 250GB-class drives have substantially lower power efficiency scores due to spending much more time working with a full SLC cache.
The 1TB 970 EVO Plus steadily picks up speed as the workload shifts to be more write heavy. The smaller 250GB model's performance flattens out during the middle half of the test as its smaller SLC write cache starts to get in the way, but toward the very end it too speeds up.
Mixed Sequential Performance
Our test of mixed sequential reads and writes differs from the mixed random I/O test by performing 128kB sequential accesses rather than 4kB accesses at random locations, and the sequential test is conducted at queue depth 1. The range of mixes tested is the same, and the timing and limits on data transfers are also the same as above.
The 970 EVO Plus is a bit faster than the original 970 EVO on the mixed sequential IO test, breaking Samsung's own record. The 250GB model finishes ahead of even the 240GB ADATA SX8200 and is only a few percent slower overall than the 1TB HP EX920.
Power Efficiency in MB/s/W | Average Power in W |
The WD Black SN750 and Corsair Force MP510 have the top two power efficiency scores on the mixed sequential IO test, but the Samsung 970 EVO Plus and original 970 EVO are next in line. Both capacities of the 970 EVO Plus require more power than most of their competition, but they put it to good use.
The key to the high overall performance scores from the 970 EVO Plus seems to be that the Samsung drives do not lose performance as quickly when writes are first added to the workload. Many drives have pretty good pure read speed but at 90% or 80% reads they may be only half as fast, while the Samsung 970 EVOs don't see a steep performance drop until around the middle of the test.
Power Management Features
Real-world client storage workloads leave SSDs idle most of the time, so the active power measurements presented earlier in this review only account for a small part of what determines a drive's suitability for battery-powered use. Especially under light use, the power efficiency of a SSD is determined mostly be how well it can save power when idle.
For many NVMe SSDs, the closely related matter of thermal management can also be important. M.2 SSDs can concentrate a lot of power in a very small space. They may also be used in locations with high ambient temperatures and poor cooling, such as tucked under a GPU on a desktop motherboard, or in a poorly-ventilated notebook.
Samsung 970 EVO Plus NVMe Power and Thermal Management Features |
|||
Controller | Samsung Phoenix | ||
Firmware | 1B2QEXM7 | ||
NVMe Version |
Feature | Status | |
1.0 | Number of operational (active) power states | 3 | |
1.1 | Number of non-operational (idle) power states | 2 | |
Autonomous Power State Transition (APST) | Supported | ||
1.2 | Warning Temperature | 85 °C | |
Critical Temperature | 85 °C | ||
1.3 | Host Controlled Thermal Management | Supported | |
Non-Operational Power State Permissive Mode | Not Supported |
The Samsung 970 EVO Plus doesn't bring any changes to the power or thermal management features supported by the 970 EVO, but the declared power limits for each power state have been increased, with the full-performance PS0 state now allowing for up to 7.8W compared to 6.2W for the original 970 EVO.
Samsung 970 EVO Plus NVMe Power States |
|||||
Controller | Samsung Phoenix | ||||
Firmware | 1B2QEXM7 | ||||
Power State |
Maximum Power |
Active/Idle | Entry Latency |
Exit Latency |
|
PS 0 | 7.8 W | Active | - | - | |
PS 1 | 6.0 W | Active | - | - | |
PS 2 | 3.4 W | Active | - | - | |
PS 3 | 70 mW | Idle | 0.21 ms | 1.2 ms | |
PS 4 | 10 mW | Idle | 2 ms | 8 ms |
Note that the above tables reflect only the information provided by the drive to the OS. The power and latency numbers are often very conservative estimates, but they are what the OS uses to determine which idle states to use and how long to wait before dropping to a deeper idle state.
Idle Power Measurement
SATA SSDs are tested with SATA link power management disabled to measure their active idle power draw, and with it enabled for the deeper idle power consumption score and the idle wake-up latency test. Our testbed, like any ordinary desktop system, cannot trigger the deepest DevSleep idle state.
Idle power management for NVMe SSDs is far more complicated than for SATA SSDs. NVMe SSDs can support several different idle power states, and through the Autonomous Power State Transition (APST) feature the operating system can set a drive's policy for when to drop down to a lower power state. There is typically a tradeoff in that lower-power states take longer to enter and wake up from, so the choice about what power states to use may differ for desktop and notebooks.
We report two idle power measurements. Active idle is representative of a typical desktop, where none of the advanced PCIe link or NVMe power saving features are enabled and the drive is immediately ready to process new commands. The idle power consumption metric is measured with PCIe Active State Power Management L1.2 state enabled and NVMe APST enabled if supported.
The 970 EVO Plus brings modest improvements to both active idle and deep idle power consumption, likely due to the reduced voltage of the Toggle DDR 4.0 interface between the controller and the new 96L 3D NAND. However, the 970 EVO Plus is still a fairly power-hungry drive when its sleep states are disabled.
The idle wake-up latency of the 970 EVO Plus is about half that of the original 970 EVO. The 970 EVO Plus is now almost an order of magnitude faster to wake up than the Silicon Motion SM2262-based drives, but the Phison E12 controller used in the Corsair MP510 provides good power management and wakes up several times faster than Samsung's NVMe drives.
Conclusion
The only technical change the Samsung 970 EVO Plus brings relative to last year's original 970 EVO is an upgrade from 64-layer 3D NAND to 92-layer 3D NAND. Fortunately, this change brings quite a few positive effects for end-users.
Samsung isn't the first to ship 9x-layer 3D NAND but they are the first to make it readily accessible to consumers in retail SSDs. This is Samsung's fifth generation 3D NAND and they have continued to improve not just the density and cost per GB, but also performance and power consumption. Those latter effects are more immediate, since the 970 EVO Plus is launching at similar prices to the 970 EVO it replaces. We do expect the 970 EVO Plus to continue the downward price trend that made 2018 such a great year for SSD buyers, but in the meantime the performance and power improvements are more interesting.
The 970 EVO Plus uses the same controller as the last generation of products, but it doesn't seem to get in the way of taking advantage of the new NAND's capabilities. With the 970 EVO Plus, Samsung retakes or strengthens their lead in some of our most challenging benchmarks. This is most noticeable in the tests that fill up the drive's SLC write cache, because the 970 EVO Plus has the best post-cache write speed we've measured from a TLC drive. This higher post-cache write speed is especially important given that Samsung's SLC caches are somewhat conservatively sized, at least in comparison to drives like the ADATA SX8200 that allow their SLC caches to grow to use almost the entire drive's array of flash memory.
The contrast between the SX8200's maximum size SLC cache strategy and the more reserved 970 EVO Plus cache sizes tends to allow the ADATA drive to perform better in ideal conditions, but hurts performance severely when the cache runs out. For the smallest drives in the 240GB–250GB range, ADATA's approach is probably a bit too aggressive because 240GB drives are fairly easy to fill up these days. A somewhat less aggressive SLC cache more like Samsung's mitigates the worst-case performance for the models that are most likely to actually hit that worst-case scenario in the real world.
Since the 970 EVO Plus does not introduce a new SSD controller, it still reflects Samsung's prioritization of performance over power efficiency. Western Digital and Toshiba have been demonstrating repeatedly that NVMe drives don't have to be power-hungry to hit the highest levels of performance, and Samsung is definitely lagging here in the overall picture. The 970 EVO Plus does score some efficiency wins, but only where it builds on the original 970 EVO's greatest strengths. For notebook use, the WD Black SN750's occasionally lower performance is absolutely worth the huge power savings.
240-280GB | 480-512GB | 960GB-1TB | 2TB | |
Samsung 970 EVO Plus (MSRP) | $89.99 (36¢/GB) | $129.99 (26¢/GB) |
$249.99 (25¢/GB) |
|
Samsung 970 EVO | $85.00 (34¢/GB) | $129.99 (26¢/GB) | $247.99 (25¢/GB) | $499.99 (25¢/GB) |
Samsung 970 PRO | $167.99 (33¢/GB) | $399.99 (39¢/GB) | ||
Western Digital WD Black SN750 (MSRP) | $79.99 (32¢/GB) | $129.99 (26¢/GB) | $249.99 (25¢/GB) | $499.99 (25¢/GB) |
Western Digital WD Black (2018) | $84.99 (34¢/GB) | $119.98 (24¢/GB) | $234.99 (23¢/GB) | |
ADATA XPG SX8200 Pro | $74.99 (29¢/GB) | $114.95 (22¢/GB) | $199.95 (20¢/GB) | |
HP EX920 | $57.99 (23¢/GB) | $89.99 (18¢/GB) | $174.99 (17¢/GB) | |
HP EX950 | $119.99 (23¢/GB) | $229.99 (22¢/GB) | $399.99 (20¢/GB) | |
Mushkin Pilot | $59.99 (24¢/GB) | $99.99 (20¢/GB) | $189.99 (19¢/GB) | $399.99 (20¢/GB) |
MyDigitalSSD BPX Pro | $54.99 (23¢/GB) | $99.99 (21¢/GB) | $189.99 (20¢/GB) | $519.99 (27¢/GB) |
Corsair Force MP510 | $74.99 (31¢/GB) | $113.99 (24¢/GB) | $266.05 (28¢/GB) | $475.99 (25¢/GB) |
With Samsung launching the 970 EVO Plus at the same prices as the 970 EVO, they're taking one of the fastest TLC drives and making it a slightly better deal.
There aren't a lot of options for stepping up in performance from the 970 EVO Plus. The 970 PRO hasn't been updated to 96L NAND and has limited capacity options, and the Intel Optane SSD 900P is even more expensive. Both drives suffer from difficulty providing any tangible benefit over fast TLC drives for lighter workloads, and on heavier workloads the PCIe 3.0 x4 bottleneck becomes an issue. They also don't consistently beat the top TLC drives on synthetic benchmarks. Given what today's best TLC drives can do, MLC NAND and 3D XPoint memory both need to be regarded as niche options that cannot automatically be assumed to offer better real-world performance.
For most purposes, the 970 EVO Plus can now be regarded as Samsung's flagship consumer SSD, and it deserves that title. Its primary competition comes from NVMe drives that are much cheaper but offer similar real-world performance with lower worst-case synthetic benchmark performance.