Introduction, Specifications and Packaging
What's better than an 18-channel NVMe PCIe Datacenter SSD controller in a Half Height Half Length (HHHL) package? *TWO* 18-channel NVMe PCIe Datacenter controllers in a HHHL package! I'm sure words to this effect were uttered in an Intel meeting room some time in the past, because such a device now exists, and is called the SSD DC P3608:
The P3608 is essentially a pair of P3600's glued together on a single PCB, much like how some graphics cards merge a pair of GPUs to act with the performance of a pair of cards combined into a single one:
What is immediately impressive here is that Intel has done this same trick within 1/4 of the space (HHHL compared to a typical graphics card). We can only imagine the potential of a pair of P3600 SSDs, so lets get right into the specs, disassembly, and testing!
Subject: Storage | September 22, 2015 - 02:39 AM | Allyn Malventano
Tagged: vnand, V-NAND, ssd, Samsung, pcie, NVMe, M.2 2280, M.2, 950 PRO, 512GB, 256GB
Samsung’s new product launching will be called the 950 PRO. This will be an M.2 2280 form factor product running at PCIe 3.0 x4. Equipped with Samsung’s 32-layer V-NAND and using the NVMe protocol enabled by a new UBX controller, the 950 PRO will be capable of up to an impressive 300,000 random read IOPS. Random writes come in at 110,000 IOPS and sequential throughputs are expected to be 2.5 GB/sec reads and 1.5 GB/sec for writes. Available capacities will be 256GB and 512GB.
- 256GB - $199.99 ($0.78/GB)
- 512GB - $349.99 ($0.68/GB)
- 1TB - (early next year with the switch to 48-layer V-NAND)
The 950 PRO will be shipping with a 5-year warranty rated at 200 terabytes written for the 256GB model and 400 TBW for the 512GB. That works out to just over 100GB per day for both capacities.
These hit retail in October and we currently have samples in hand for testing.
(for those curious, both capacities only have components on the front side of the PCB)
Subject: Storage | September 21, 2015 - 11:32 AM | Allyn Malventano
Tagged: vnand, Summit, ssd, Seoul, Samsung, M.2, Korea, Global, 2015
As I hinted during last week's podcast, I am in Seoul, Korea to cover an upcoming press conference.
..and with a Samsung SSD Global Summit comes product announcements. Those don't happen until tomorrow (late tonight for you folks back in the states), but I did notice a clue on the cover of our itinerary folder:
See it? Here, let me help:
A VNAND powered M.2 (presumably NVMe) SSD is *exactly* the thing I have been waiting for Samsung to unleash into the wild ever since we reviewed their NVMe SM951. Given that Samsung's prior M.2 offerings gave the Intel SSD 750 a run for its money all while consuming half the power, and did so with Samsung's older 2D Planar NAND, you can bet a VNAND version will be something to behold. Let's hope this new model is released as a consumer product and doesn't end up as OEM-channel unobtanium like the NVMe SM951 was!
Keep an eye out for additional posts from our coverage of the 2015 Samsung SSD Global Summit!
Subject: Storage | August 19, 2015 - 09:41 PM | Allyn Malventano
Tagged: IDF 2015, ocz, revodrive, RevoDrive 400, M.2, HHHL, pcie, NVMe, ssd
While roaming around at IDF, Ryan spotted a couple of new OCZ parts that were strangely absent from Flash Memory Summit:
You are looking at what is basically a Toshiba NVMe PCIe controller and flash, tuned for consumer applications and packaged/branded by OCZ. The only specific we know about it is that the scheduled release is in the November time frame. No specifics on performance yet but it should easily surpass any SATA SSD, but might fall short of the quad-controller-RAID RevoDrive 350 in sequentials.
As far as NVMe PCIe SSDs go, I'm happy to see more and more appearing on the market from every possible direction. It can only mean good things as it will push motherboard makers to perfect their UEFI boot compatibility sooner rather than later.
More to come on the RevoDrive 400 as November is just around the corner!
Subject: Storage | August 19, 2015 - 09:26 PM | Allyn Malventano
Tagged: ssd, pcie, NVMe, kingston, IDF 2015
**Edit** There was some speculation about which controller was in this SSD. It has since been solved. Here's a shot of the top of the PCB:
Now lets compare that with a shot I caught at FMS 2015 last week:
...from the Phison booth. I hadn't wirtten up my Phison post yet but this new Kingston SSD is most certainly going to be using the Phison E7 controller. Here's the placard stating some high level specs:
We saw a draft copy of Kingston’s HyperX Predator at CES 2014. That demo unit was equipped with a SandForce 3700 series controller, but since SandForce never came through on that part, Kingston had to switch gears and introduce the HyperX Predator with a Marvell 88SS9293 controller. The Marvell part was very capable, and the HyperX Predator turned out to be an attractive and performant PCIe SSD. The one catch was that Marvell’s controller was only an AHCI part, while newer NVMe-based SSDs were quickly pushing the Predator down in our performance results.
Kingston’s solution is a newer generation PCIe SSD, this time equipped with NVMe:
We have very little additional information about this new part, though we can tell from the above image that the flash was provided by Toshiba (toggle mode). They also had Iometer running:
We were not sure of the exact workload being run, but those results are in line with the specs we saw listed on Silicon Motion’s SM2260, seen last week at Flash Memory Summit.
We’ll keep track of the development of this new part and hope to see it in a more disclosed form at CES 2016. Kingston's IDF 2015 press blast appears after the break.
Subject: Storage | August 18, 2015 - 02:20 PM | Allyn Malventano
Tagged: XPoint, ssd, Optane, Intel, IDF 2015
Just three weeks ago, we reported 3D XPoint Technology. This was a 2-layer stack of non-volatile memory that couples the data retention of NAND flash memory with speeds much closer to that of DRAM.
The big question at that time was less about the tech and more about its practical applications. Ryan is out covering IDF, and he just saw the first publically announced application by Intel:
Intel Optane Technology is Intel’s term for how they are going to incorporate XPoint memory dies into the devices we use today. They intend to start with datacenter storage and work their way down to ultrabooks, which means that XPoint must come in at a cost/GB closer to NAND than to DRAM. For those asking specific performance figures after our earlier announcement, here are a couple of performance comparisons between an SSD DC P3700 and a prototype SSD using XPoint:
At QD=8, the XPoint equipped prototype comes in at 5x the performance of the P3700. The bigger question is how about QD=1 performance, as XPoint is supposed to be far less latent than NAND?
Yes, you read that correctly, that’s 76k IOPS at QD=1. That means only issuing the SSD one command at a time, waiting for a reply, and only then issuing another command. Basically the worst case for SSD performance, as no commands are stacked up in the queue to enable parallelism to kick in and increase overall throughput. For comparison, SATA SSDs have a hard time maintaining that figure at their maximum queue depths of 32.
Exciting to see a follow-on announcement so quickly after the announcement of the technology itself, but remember that Intel did state ‘2016’ for these to start appearing, so don’t put off that SSD 750 purchase just yet.
More to follow as we continue our coverage of IDF 2015!
Subject: Storage | August 14, 2015 - 04:44 PM | Allyn Malventano
Tagged: FMS 2015, silicon motion, SM2260, SM2256, SM2246EN, pcie, NVMe, ssd, controller
We’ve reviewed a few Silicon Motion SSDs in the past (Angelbird | Corsair Force LX | Crucial BX100), and I have always been impressed with their advances in SSD controller technology. Their SM2246EN SATA controller was launched two years ago, and strived to be a very efficient and performant unit. Based on our reviews that turned out to be true, and this allowed Silicon Motion to slide into the void left by SandForce, who repeatedly delayed their newer developments and forced the many companies who were sourcing their parts to look elsewhere.
The many SSDs using Silicon Motion’s SM2246EN controller.
Silicon motion pushed this further with their SM2256, which we first saw at the 2014 Flash Memory Summit and later saw driving SLC/TLC hybrid flash at this past Consumer Electronics Show. While the SM2256 makes its way into more and more products, I was glad to see an important addition to their lineup at this year’s FMS:
Finally we see Silicon Motion doing a PCIe controller! This is the SM2260, seen here in the M.2 form factor…
…and here in SATA Express. While the latter will likely not be as popular due to the more limited PCIe lanes present in SATA Express, I’m sure we will see this controller appearing in many PCIe devices very soon. The stated performance figures may be a bit shy of currently comparing SSDs like the Intel SSD 750 and Samsung SM951, but with the recent introduction of Z170 motherboards and RST PCIe RAID, it is now easier to RAID a smaller capacity pair of these devices, increasing the performance of slower units. Further, the point of the SM2260 is likely to get a low cost NVMe PCIe SSD controller into the hands of SSD makers, which can only mean good things for those looking to make the move away from SATA.
I’ve included Silicon Motion’s FMS press blast after the break.
Subject: Storage | August 13, 2015 - 08:12 PM | Allyn Malventano
Tagged: FMS 2015, ssd, sata, SAS, pcie, NVMe, novachips, HLNAND, flash
It turns out Samsung wasn’t the only company to have 16TB SSDs at Flash Memory Summit after all:
Now that I’ve got your attention, Novachips is an SSD company that does not make their own flash, but I would argue that they make other peoples flash better. They source flash memory wafers and dies from other companies, but they package it in a unique way that enables very large numbers of flash dies per controller. This is handy for situations where very large capacities per controller are needed (either physically or logically).
Normally there is a limit to the number of dies that can communicate on a common bus (similar limits apply to DRAM, which is why some motherboards are picky with large numbers of DIMMs installed). Novachips gets around this with an innovative flash packaging method:
The 16-die stack in the above picture would normally just connect out the bottom of the package, but in the Novachips parts, those connections are made to a microcontroller die also present within the package. This part acts as an interface back to the main SSD controller, but it does so over a ring bus architecture.
To clarify, those 800 or 1600 MB/sec figures on the above slide are the transfer rates *per ring*, and Novachips controller is 8-channels, meaning the flash side of the controller can handle massive throughputs. Ring busses are not limited by the same fanout requirements seen on parallel addressed devices, which means there is no practical limit to the number of flash packages connected on a single controller channel, making for some outrageous amounts of flash hanging off of a single controller:
That’s a lot of flash on a single card (and yes, the other side was full as well).
The above pic was taken at last years Flash Memory Summit. Novachips has been making steady progress on controller development as well. Here is a prototype controller seen last year running on an FPGA test system:
…and this year that same controller had been migrated to an ASIC:
It’s interesting to see the physical differences between those two parts. Note that both new and old platforms were connected to the same banks of flash. The newer photo showed two complete systems – one on ONFi flash (IMFT Intel / Micron) and the other on Toggle Mode (Toshiba). This was done to demonstrate that Novachips HLNAND hardware is compatible with both types.
Novachips also had NVMe PCIe hardware up and running at the show.
Novachips was also showing some impressive packaging in their SATA devices:
At the right was a 2TB SATA SSD, and at the left was a 4TB unit. Both were in the 7mm form factor. 4TB is the largest capacity SSD I have seen in that form factor to date.
Novachips also makes an 8TB variant, though the added PCB requires 15mm packaging.
All of this means that it is not always necessary to have huge capacity per die to achieve a huge capacity SSD. Imagine very high capacity flash arrays using this technology, connecting a single controller to a bank of Toshiba’s new QLC archival flash or Samsung’s new 256Gbit VNAND. Then imagine a server full of those PCIe devices. Things certainly seem to be getting big in the world of flash memory, that’s for sure.
Even more Flash Memory Summit posts to follow!
Subject: Storage | August 11, 2015 - 08:40 PM | Allyn Malventano
Tagged: toshiba, ssd, FMS 2015, flash, BiCS, Archive, Archival, 3d
We occasionally throw around the '3-bit MLC' (Multi Level Cell) term in place of 'TLC' (Triple Level Cell) when talking about flash memory. Those terms are interchangeable, but some feel it is misleading as the former still contains the term MLC. At Toshiba's keynote today, they showed us why the former is important:
Photo source: Sam Chen of Custom PC Review
That's right - QLC (Quadruple Level Cell), which is also 4-bit MLC, has been mentioned by Toshiba. As you can see at the right of that slide, storing four bits in a single flash cell means there are *sixteen* very narrow voltage ranges representing the stored data. That is a very hard thing to do, and even harder to do with high performance (programming/writing would take a relatively long time as the circuitry nudges the voltages to such a precise level). This is why Toshiba pitched this flash as a low cost solution for archival purposes. You wouldn't want to use this type of flash in a device that was written constantly, since the channel materials wearing out would have a much more significant effect on endurance. Suiting this flash to be written only a few times would keep it in a 'newer' state that would be effective for solid state data archiving.
The 1x / 0.5x / 6x figures appearing in the slide are meant to compare relative endurance to Toshiba's own planar 15nm flash. The figures suggest that Toshiba's BiCS 3D flash is efficient enough to go to QLC (4-bit) levels and still maintain a higher margin than their current MLC (2-bit) 2D flash.
More to follow as we continue our Flash Memory Summit coverage!
Subject: Storage | August 6, 2015 - 06:37 PM | Allyn Malventano
Tagged: SSD 750, ssd, pcie, NVMe, Intel
A new 800GB SKU of the Intel SSD 750 Series of PCIe SSDs was hinted at with the Skylake launch press materials, and it appears to have been a reality:
They may not be on the shelves yet, but appearing on ARK is a pretty good indicator that these are coming soon. We don't have pricing yet, but I would suspect a cost/GB closer to the 1.2TB model than to the 400GB model, which should come in at around $700. Performance sees a slight hit for the 800GB model, likely since this is an 'uneven' number of dies for the design of the SSD DC P3500 line it was based on.
Which would you prefer - a single 800GB or a pair of 400GB SSD 750's in a RAID (now that it is possible)?