As mentioned, this article is less about the OEM Samsung PM961 and more about previewing the performance we should see when Samsung releases an NVMe-based consumer SSD with TLC V-NAND. Later this month most of the established storage media will converge in Seoul, South Korea for, well..something. We don't know with any certainty what Samsung will announce this year, but the rumors are flowing like wine at a wedding. We expect to see a 960 EVO SSD based on the PM961 we tested, and I would be very surprised if it isn't part of the agenda.
Samsung's 48-layer TLC V-NAND is able to deliver up to 1600 MB/s with 128KB blocks in SLC mode and only two NAND packages. In native TLC mode, the performance drops to around 650 MB/s. In contrast, the Intel 600p 512GB achieves 550 MB/s with simulated SLC and fluctuates wildly after the cache is full. The Intel 600p does not use a direct-to-die write technique (which circumvents the SLC layer when it is full) so all of the data written to the drive must pass through the SLC buffer and then fold into the TLC. Intel's caching scheme makes it difficult to determine its native TLC performance. The 600p's wild teeter-totter during the 128KB sequential write workload fluctuates between 450 MB/s and 20 MB/s. No, I didn't miss a zero there--20 MB/s is accurate.
The Intel 600p plays a big role in the 960 EVO's destiny. Intel released the NVMe drive with aggressive pricing, but we don't think Intel planned to have the 600p 512GB selling at just $175 this early in the product lifecycle. The performance clearly keeps the drive out of the high-performance class. Be it intentional or not, the 600p ushers in a new segment for entry-level PCIe-based NVMe SSDs.
Phison is pushing the next wave of products that will fall into the same value category. Two PS5007-E7 products are shipping, and more will follow. Phison has yet to prove that a retail E7 product can hang with the big dogs, like the Samsung 950 Pro, SM961 and Intel SSD 750. Surprisingly, we do have a prototype E7 with 1TB of MLC flash running entirely in SLC mode (it is the fastest SSD I have in my lab). We recently learned of a new firmware update for retail E7-based drives that will increase performance. Samples are on the way and should arrive just after this review hits the site.
It won't be long before Phison pairs the E7 with Toshiba's BiCS 3D flash running in TLC mode, and we may see very low cost solutions come from the pair. Until then, most companies will utilize economies of scale and pair the E7 with Toshiba planar MLC NAND. Just like with SATA, these devices will need to leverage low pricing to move products in volume. It will be difficult, and margins will be slim, but the E7 products are coming from more than one big name in the consumer space.
Speaking of difficult times, Ballistix by Micron (Crucial) canceled the TX3 after a Computex reveal last June. The drive used a variant of the Silicon Motion, Inc (SMI) SM2260, which is the same base controller Intel used in the 600p. Micron planned to pair the drive with MLC flash, but pulled it back just days before the samples went to media.
We've been hard on IMFT's new 3D NAND. The flash scales capacity well, but we haven't seen a single high-performance product yet from either Intel or Micron, much less the customers that buy the flash to use in their own products. Intel has the DC P3520, which is the highest performance SSD with the new IMFT 3D NAND, but if you follow Intel's data center product line you already know the 3500 series is the lowest model in the food chain. It slots in under the 3600 and flagship 3700 Series, so we really have to wonder if IMFT's new 3D NAND is as good as Intel/Micron says it is.
With that long preface, let's circle back to Samsung's 3D TLC NAND and what it could mean for retail SSD shoppers. If the price is too low, Samsung will not be able to keep store shelves stocked, just as we've seen with the SM961. There are other factors in play right now, too. We are in the middle of a NAND shortage, partly due to the new Note 7 and Apple iPhone soaking up a lot of flash in very large capacities. The slow transition from planar to 3D for IMFT and Toshiba hasn't helped thus far, either.
When Samsung releases the 960 products, we expect them to sit on the market for a long time. Instead of Samsung progressing the product-naming scheme, we expect to see it increase capacities roughly every nine months. The 960 EVO should launch with 48-layer NAND, and the larger capacities will come to market with new 60-layer NAND in Q1 or Q2 2017.
Currently, the PM961 is not a product we recommend outside of a manufacturer upgrade in an OEM notebook. The few rare PM961s that are available cost an arm and a leg, and the sellers don't offer favorable terms like we get from RamCity with the SM961 SSDs. We've seen some of the PM961 drives pulled out of Lenovo Carbon X1 Gen 4 systems and sold online, and many sell for more than what Lenovo charges for the upgrade. A few sellers have also started calling them the 960 Pro and 960 EVO, which is a bit misleading. We are too close to Samsung's announcement, and potential retail availability, to advise readers to pull the trigger on the Samsung PM961.