Original Link: https://www.anandtech.com/show/9433/adata-xpg-sx930-120gb-240gb-480gb-ssd-review
ADATA XPG SX930 (120GB, 240GB & 480GB) SSD Review: JMicron JMF670H Debuts
by Kristian Vättö on July 16, 2015 10:00 AM EST
At Computex ADATA had a variety of new SSDs on display. While most were based on upcoming technologies such as TLC NAND and the PCIe/NVMe interface, the company also displayed an XPG SX930, which is an update to ADATA's high-end XPG lineup. The series was in need of a refresh because the SX900 dates back to 2012, so with the SX930 ADATA is hoping to breath new life into its enthusiast SSDs.
ADATA markets the SX930 as a gaming SSD with a five-year warranty and in order to attract more gamers the company had to reconsider its branding and design. The hummingbird logo was considered not to be very "street-proof" among gamers who are often looking for something a little flashier, hence the old logo is now gone and replaced by a new flame design. Personally I think flames are quite a cliché in gamer marketing, but I guess a portion of the gaming crowd may appreciate a design that's more than just a metal chassis. Whether the look of an SSD inside a case matters is a whole new question, but I'll let everyone use their own justification on that.
On the hardware side the SX930 uses JMicron's brand new JMF670H controller. JMicron has never really been known for high performance SSD controllers and while the JMF670H isn't aimed to take a jab at Samsung and Marvell based drives, JMicron believes it is competitive against Silicon Motion's SM2246EN and Phison's S10 controllers.
At the silicon level the JMF670H is very similar to its predecessor. JMicron employs a single 32-bit ARM968 core, which is the smallest and lowest power member of ARM's ARM9 family and is mostly aimed for embedded applications such as SSD, USB and networking controllers. JMicron prefers not to disclose the frequency, but told us that the frequency is the same in both JMF667H and JMF670H.
Only the ECC circuitry sees an enhancement to support BCH ECC of 72 bits per 1KB (i.e. can correct up to 72 bits in 1KB of data), whereas the JMF667H was only capable of correcting up to 40 bits. Improving ECC is necessary for supporting the latest 15nm and 16nm NAND nodes because as NAND scales down the error rate increases as cells become more vulnerable to cell-to-cell interference and electron leakage. JMicron doesn't have any RAID5-like parity scheme in the JMF670H, so the BCH ECC engine is solely responsible for error correction.
While Silicon Motion and Phison both support TLC in their latest controllers, JMicron won't be supporting TLC until next year when the JMF680H ships. That's certainly a disadvantage compared to the competition, but I'm no longer that bullish on TLC after Samsung's issues and the marginal price cuts that OEMs are promising. I'm now looking forward to 3D TLC because it will enable planar MLC-like performance and endurance, along with hopefully larger price cuts because the market in general will shift more towards TLC, meaning higher production and scale benefits. At this time, we wait for JMicron's solution in this space.
| ADATA XPG SX930 Specifications | |||||
| Capacity | 120GB | 240GB | 480GB | ||
| Controller | JMicron JMF670H | ||||
| NAND | Micron 16nm 128Gbit MLC | ||||
| Sequential Read | 550MB/s | 550MB/s | 540MB/s | ||
| Sequential Write | 460MB/s | 460MB/s | 420MB/s | ||
| 4KB Random Read | 70K IOPS | 75K IOPS | 75K IOPS | ||
| 4KB Random Write | 45K IOPS | 70K IOPS | 72K IOPS | ||
| Slumber Power | 140mW | 140mW | 140mW | ||
| Read Power | 1.38W | 1.39W | 1.48W | ||
| Write Power | 1.90W | 3.05W | 4.38W | ||
| Encryption | N/A | ||||
| Warranty | Five years | ||||
| MSRP | $80 | $110 | $200 | ||
The retail package includes all the typical ADATA accessories: 3.5" adapter, 9.5mm spacer and eight mounting screws (four for the drive and another four for the desktop adapter). All buyers can download a Disk Migration Utility from ADATA's website, which was co-developed with Acronis that supplies migration software to nearly all SSD vendors.
The SX930's maximum capacity is 480GB because of the JMF670H's limitations. With the DRAM controller in the JMF670H only supporting 512MB of DDR3, the NAND capacity maxes out at 512GB because modern NAND mapping table designs typically require about 1MB of DRAM per every 1GB of NAND. The next generation JMF680H will overhaul the DRAM controller and support up to 2GB of DRAM, enabling capacities as high as 2TB. While 1TB-class SSDs are certainly still a small niche, it comes across a little confusing that ADATA's high-end SX930 does not have a 1TB model, but the more value-oriented Premier SP610 and Premier Pro SP920 (with 3-year warranties) carry 1TB SKUs. For an end-user this is mostly negligible, but I'm not convinced this is the best product positioning strategy.
ADATA refers to the NAND in the SX930 as "Enterprise MLC+". It's certainly not eMLC, but merely higher binned normal MLC to support the five-year warranty ADATA is offering (compared to the standard 3-year). As ADATA does NAND binning and packaging in-house, it has the ability to sort dies and save the highest quality ones for SX930 and enterprise SSDs, while the lower quality dies end up in other client SSDs, USB sticks and SD cards depending on the quality level. ADATA doesn't give separate endurance rating for the SX930, but I was told the NAND endurance is at least 3,000 P/E cycles, which should give a rough idea of the expected lifetime.
| Capacity | 120GB | 240GB | 480GB |
| SLC Cache Size | 4GB | 8GB | 16GB |
While the hardware side of the new JMF670H doesn't differ much from the old JMF667H controller, the firmware has been upgraded. The JMF670H firmware carries a feature called Write Booster, which is JMicron's SLC cache implementation. Even though the JMF670H doesn't support TLC NAND, JMicron believes that an SLC cache can still improve performance for MLC, especially when combined with 15nm or 16nm NAND with higher program/erase latencies. Write Booster caches all IOs regardless of their size, and JMicron does some write optimizations when moving data from the SLC cache to MLC cache to reduce write amplification for higher endurance.
Write Booster works with both normal NAND and pseudo-SLC NAND. As many of you may know, it's possible to program just one bit per cell to MLC NAND by only using the lower page (i.e. larger voltage distribution), but NAND vendors also have special pseudo-SLC dies. Unfortunately, all vendors are relatively quiet about what exactly happens inside a pseudo-SLC die, but JMicron told us that there is an improvement in read performance when using proper pseudo-SLC instead of simply writing to lower pages. In the case of SX930, ADATA is using real pseudo-SLC NAND from Micron, which does carry a price premium over normal NAND, but given the higher-end focus of the SX930 that makes sense.
| AnandTech 2015 SSD Test System | |
| CPU | Intel Core i7-4770K running at 3.5GHz (Turbo & EIST enabled, C-states disabled) |
| Motherboard | ASUS Z97 Deluxe (BIOS 2205) |
| Chipset | Intel Z97 |
| Chipset Drivers | Intel 10.0.24+ Intel RST 13.2.4.1000 |
| Memory | Corsair Vengeance DDR3-1866 2x8GB (9-10-9-27 2T) |
| Graphics | Intel HD Graphics 4600 |
| Graphics Drivers | 15.33.8.64.3345 |
| Desktop Resolution | 1920 x 1080 |
| OS | Windows 8.1 x64 |
- Thanks to Intel for the Core i7-4770K CPU
- Thanks to ASUS for the Z97 Deluxe motherboard
- Thanks to Corsair for the Vengeance 16GB DDR3-1866 DRAM kit, RM750 power supply, Hydro H60 CPU cooler and Carbide 330R case
Performance Consistency
We've been looking at performance consistency since the Intel SSD DC S3700 review in late 2012 and it has become one of the cornerstones of our SSD reviews. Back in the days many SSD vendors were only focusing on high peak performance, which unfortunately came at the cost of sustained performance. In other words, the drives would push high IOPS in certain synthetic scenarios to provide nice marketing numbers, but as soon as you pushed the drive for more than a few minutes you could easily run into hiccups caused by poor performance consistency.
Once we started exploring IO consistency, nearly all SSD manufacturers made a move to improve consistency and for the 2015 suite, I haven't made any significant changes to the methodology we use to test IO consistency. The biggest change is the move from VDBench to Iometer 1.1.0 as the benchmarking software and I've also extended the test from 2000 seconds to a full hour to ensure that all drives hit steady-state during the test.
For better readability, I now provide bar graphs with the first one being an average IOPS of the last 400 seconds and the second graph displaying the IOPS divided by standard deviation during the same period. Average IOPS provides a quick look into overall performance, but it can easily hide bad consistency, so looking at standard deviation is necessary for a complete look into consistency.
I'm still providing the same scatter graphs too, of course. However, I decided to dump the logarithmic graphs and go linear-only since logarithmic graphs aren't as accurate and can be hard to interpret for those who aren't familiar with them. I provide two graphs: one that includes the whole duration of the test and another that focuses on the last 400 seconds of the test to get a better scope into steady-state performance. These results are for all drives 240GB and up.
Steady-state performance is fairly good, although given that the SX930 employes 12% over-provisioning it's expected to perform better than drives with only 7%.
Consistency, on the other hand, isn't that good. The good news is that the JMF670H is more consistent than the SM2246EN in the BX100, but there's a long way to achieve Samsung-level consistency.
 |
|||||||||
| Default | |||||||||
| 25% Over-Provisioning | |||||||||
As the SX930 isn't very consistent, we saw performance dropping below 1,000 IOPS. It seems that consistency is one of the areas that truly separate Samsung and Marvell drives from the rest because JMicron, Silicon Motion and Phison based drives all have trouble sustaining steady performance. Basically, the baseline performance is about 1,000 IOPS for all controllers with frequent peaks of +5,000 IOPS, whereas Samsung and Marvell drives have very little variation in performance. I suspect the controller design itself has something to do with this because Samsung and Marvell controllers are all multi-core, but JMicron and Silicon Motion rely on single-core designs for higher cost efficiency.
 |
|||||||||
| Default | |||||||||
| 25% Over-Provisioning | |||||||||
AnandTech Storage Bench - The Destroyer
The Destroyer has been an essential part of our SSD test suite for nearly two years now. It was crafted to provide a benchmark for very IO intensive workloads, which is where you most often notice the difference between drives. It's not necessarily the most relevant test to an average user, but for anyone with a heavier IO workload The Destroyer should do a good job at characterizing performance. For full details of this test, please refer to this article.
The SX930 doesn't perform too well in our The Destroyer trace. If this was a value-oriented drive, I would say the performance is decent, but any drive that is focusing on the higher-end segment should outperform the BX100 and 850 EVO to have any chance of being competitive.
The number of high latency IOs isn't particularly large, but again the SX930 is only competitive against the value drives.
Power consumption is fairly average for the 480GB model, but the 240GB consumes substantially more due to its lower performance.
AnandTech Storage Bench - Heavy
While The Destroyer focuses on sustained and worst-case performance by hammering the drive with nearly 1TB worth of writes, the Heavy trace provides a more typical enthusiast and power user workload. By writing less to the drive, the Heavy trace doesn't drive the SSD into steady-state and thus the trace gives us a good idea of peak performance combined with some basic garbage collection routines. For full details of the test, please refer to the this article.
In our Heavy trace the SX930 is an average performer. To be frank, the JMF670H doesn't seem to have enough juice to go head to head with the 850 EVO and BX100, which is a shame because it leaves price as the only competitive advantage ADATA can have.
Surprisingly the 240GB model experiences more >10ms IOs in our Heavy trace than in The Destroyer. I would speculate this is due to the SLC cache filling up because the Heavy trace is more peak IO focused, whereas The Destroyer is built more for sustained performance, so it may be that in The Destroyer the controller just bypasses the SLC cache at some point and writes directly to MLC. In any case, optimizations are certainly needed to improve performance.
Power consumption, on the other hand, is relatively good. It's not BX100-level, but compared to some other high-end SSDs the SX930 has a more efficient power profile.
AnandTech Storage Bench - Light
The Light trace is designed to be an accurate illustration of basic usage. It's basically a subset of the Heavy trace, but we've left out some workloads to reduce the writes and make it more read intensive in general. Please refer to this article for full details of the test.
It's been common knowledge for quite some time that for light client workloads, the differences between SATA SSDs are mostly negligible and the SX930 doesn't change the story. It's a decent performer, but can't offer any advantage over its alternatives.
The SX930 is quite power efficient, although it can't challenge the BX100 in this area.
Random Read Performance
For full details of how we conduct our Iometer tests, please refer to this article.
Random read performance is quite average and only Samsung has a notable advantage in this field, which is partly due to 3D V-NAND and its lower latencies.
Power consumption is on par with the BX100 and even though the BX100 has a slight performance advantage, the SX930 is very power efficient in random reads.
 |
|||||||||
Performance scaling could be a little more aggressive especially at QD4 and QD8, but overall random performance appears to be suited for the product.
Random Write Performance
Random write performance is mediocre and especially the 480GB SKU could use a boost to be more competitive. Fortunately power consumption is quite low, so the overall efficiency is decent.
Given the pseudo-SLC cache, one would expect performance scaling to be more aggressive from low to high, especially at QD4 and above, but there leaves an element of wanting compared to other high-end drives on the market.
 |
|||||||||
Sequential Read Performance
For full details of how we conduct our Iometer tests, please refer to this article.
Sequential read performance is surprisingly marked down. While we are only talking about a 10-20% difference, given the longevity of the push towards sequential data rates, I wouldn't expect any modern controller to have trouble with sequential read performance anymore.
Power is also low, though, resulting in efficient, yet a bit low performance.
 |
|||||||||
The reason for the poor sequential read performance lies in very poor performance scaling. Typically all modern drives reach ~550MB/s at QD2, but the JMF670H requires QD8 to reach its maximum performance. This has a knock on effect for any non-prosumer or gaming scenario which is where the SX930 series is aimed at.
Sequential Write Performance
Sequential write performance is a bit better and here the SX930 is a quite average drive, although the 480GB model should again be faster since it's slightly behind its competitors.
Scaling with queue depth is minimal, which is normal for low capacity drives but at 240GB and 480GB there should be enough NAND bandwidth available for higher throughput which we don't see here, perhaps indicating some of the limitations of the controller.
 |
|||||||||
Mixed Random Read/Write Performance
For full details of how we conduct our Iometer tests, please refer to this article.
The 480GB and 120GB do fairly well in mixed performance, but the 240GB is behind its peers. Although it is worth noting that the 480GB does also consume twice the power, making it not very efficient for use in battery driven platforms.
There is nothing spectacular in the scaling graphs. The performance scaling could be more aggressive with writes, but the SX930 doesn't have good write performance to begin with.
 |
|||||||||
Mixed Sequential Read/Write Performance
The SX930 does well in mixed sequential performance and the power efficiency is good for this test.
The reason lies in good read-centric performance, which is surprising given that sequential read performance is the Achilles' heel of the SX930. Despite this it performs well with writes thrown into the mix, although again the performance when the distributions become write-heavy are nothing to write home about.
 |
|||||||||
ATTO - Transfer Size vs Performance
ATTO is a handy tool for quickly measuring performance across various transfer sizes and it's also freeware that can easily be run by the end-user.
 |
|||||||||
AS-SSD Incompressible Sequential Performance
Similar to ATTO, AS-SSD is freeware as well and uses incompressible data for all of its transfers, making it a valuable tool when testing drives with built-in compression engines (e.g. SandForce).
Idle Power Consumption
Since we truncate idle times to 25µs in our Storage Bench traces, they don't give a fully accurate picture of real world power consumption as idle power consumption is not taken properly into account. Hence I'm still reporting idle power consumption as a separate benchmark because it's one of the most critical metrics when it comes evaluating an SSD for mobile use.
Unfortunately I still don't have a way to test DevSleep power consumption due to lack of platform support, but my testbed supports HIPM+DIPM power commands (also referred to as Slumber power), so the results give a rather accurate picture of real-world idle power consumption.
Compared to the reference samples, the SX930 isn't as low power as it could be. I suspect this is due to ADATA's optimizations, which could be to increase compatibility as some systems have issues with power saving states. Still, I would like to see the power saving settings in place because as it stands the SX930 consumes more power at idle than the competing drives with HIPM+DIPM support.
TRIM Validation
The move from Windows 7 to 8.1 introduced some problems with the methodology we have previously used to test TRIM functionality, so I had to come up with a new way to test. I tested a couple of different methods, but ultimately I decided to go with the easiest one that can actually be used by anyone. The software is simply called trimcheck and it was made by a developer that goes by the name CyberShadow in GitHub.
Trimcheck tests TRIM by creating a small, unique file and then deleting it. Next the program will check whether the data is still accessible by reading the raw LBA locations. If the data that is returned by the drive is all zeros, it has received the TRIM command and TRIM is functional.
Final Words
To be frank, reviewing a SATA MLC SSD has gotten rather unexciting over the past year or so. On the performance side there are barely any areas where one can get enthusiastic about because the SATA 6Gbps interface and AHCI driver stack are both so saturated. It feels like the purpose of my testing is mostly to make sure that someone didn't totally screw up the product design because other than that the performance differences between modern SATA 6Gbps controllers are getting negligible. Only Samsung and SanDisk can provide performance that's distinguishably better than others, which leaves JMicron, Silicon Motion and Phison based drives fighting over the value segment.
This brings us to the SX930 and JMF670H. If I had to pick one word to describe the two, that would be 'average'. There is nothing that truly separates the SX930 and JMF670H from the drives and controllers that are already available. Performance wise the JMF670H is fairly similar to Silicon Motion's SM2246EN, but at the end of the day the SM2246EN wins in both performance and power efficiency, which makes it difficult for ADATA and JMicron to compete in areas other than price or features.
While utilizing higher binned MLC NAND (or "enterprise-grade" as ADATA calls it) could be considered as a differentiating feature, I don't consider NAND endurance to be a significant issue for client usage, so even though the NAND is likely higher quality than what you would find inside a BX100 for instance, it's not going to have any impact on the end-user. A five-year warranty is definitely a welcome addition, but that alone doesn't provide enough value to make the SX930 stand out, especially with Samsung offer a five-year warranty for the 850 EVO.
| Amazon Price Comparison (7/16/2015) | |||
| 120/128GB | 240/250/256GB | 480/500/512GB | |
| ADATA XPG SX930 (MSRP) | $80 | $110 | $200 |
| ADATA Premier SP610 | $60 | $100 | $188 |
| Crucial MX200 | - | $103 | $180 |
| Crucial BX100 | $66 | $90 | $178 |
| OCZ Trion 100 | $60 | $90 | $180 |
| OCZ ARC 100 | $54 | $89 | $170 |
| OCZ Vector 180 | $80 | $130 | $250 |
| Samsung 850 EVO | $72 | $98 | $178 |
| Samsung 850 Pro | $97 | $140 | $251 |
| SanDisk Ultra II | $63 | $95 | $182 |
| SanDisk Extreme Pro | - | $135 | $220 |
| Transcend SSD370 | $58 | $90 | $176 |
Since the SX930 is ADATA's high-end drive, the prices aren't exactly cheap. You are looking at about $20 premium over the BX100, which is hard to justify given that the BX100 actually provides better performance. While street pricing tends to be lower than MSRPs, it's clear that the SX930 needs to be about $20 cheaper to be competitive. At equal pricing with the BX100, I might lean towards the SX930 and take a marginal hit in performance for two years of additional warranty, but I wouldn't pay $20 for the warranty alone because of the rapid developments in SSD performance and prices dropping about 20% year over year.
I did let JMicron know about my performance concerns when I tested the JMF670H reference design samples because sequential read performance in particular was below the average. JMicron promised an improvement through an upcoming firmware update and told me that the initial firmware mostly focused on optimizing performance for benchmarks such as CrystalDiskMark and AS-SSD, which typically use higher IO sizes and queue depths to extract the maximum performance out of an SSD. With a firmware better optimized for low queue depths and real world workloads, I see potential in the SX930 and JMF670H, but nevertheless it still needs to be more competitive in price in order to tackle the BX100 and 850 EVO.
Facebook
Twitter
RSS
