Samsung Finally Goes QLC!
Samsung has opted to name this new product 'QVO'. The Q presumably stems from the use QLC flash, which can store four bits per cell.
While QLC writes are far slower than what we are used to seeing from a modern SSD, SLC caching is the answer to bridging that performance gap. The 860 QVO employs Samsung's Intelligent TurboWrite, which has a minimum 6GB static cache plus a dynamic cache of up to 72GB. This dynamic cache varies based on available QLC area which can be reconfigured to operate in SLC mode. Do note the 'After TubroWrite' speeds of 80 and 160 MB/s – that's the raw QLC speeds that you will see if the cache has been exhausted during an extended write period.
The rest of the specs are about what we expect from a SATA SSD, but I do have a concern with those QD1 4KB random read ratings of only 7,500 IOPS. This is on the low side especially for Samsung, who typically dominate in low QD random read performance.
Opening the SSDs we find comically small PCBs that are not even fully populated with flash packages.
Both the 1TB and 4TB models had only half of their flash locations populated, meaning the 1TB PCB can easily support 2TB, and the 4TB can easily support 8TB (though there is no 8TB model announced).
Jumping right into the test results (details of how the suite works here), we see weighted random read results coming in close to the 7,500 IOPS spec, which is disappointing given that is half of the performance of the 860 EVO 1TB. I've included the 660P and P1, which are NVMe QLC SSDs. While NVMe/PCIe does allow for higher throughputs, it does not work miracles for QD1 performance, meaning Samsung's QLC appears to have higher read latency than Intel / Micron QLC flash.
Here is a closer look at the low QD read performance. Note how much lower the two QVO SSDs fall compared to the rest of the pack.
Things are a bit better with sequential performance. Note this is a burst test that hits the cache, so we won't see the 'After TurboWrite' speeds here (we will later).
Respectable performance for a SATA SSD during the Mixed Burst test.
No surprises during the TRIM test, either.
Now for what I view as an important part of any QLC SSD evaluation – cache performance. We need the cache to be as effective as possible since running out of it means hitting that 80/160 MB/s brick wall. This test is performed with the SSD half full, meaning there should have been some dynamic cache available. That appeared to not be the case during the first run, where we saw 6GB of cache followed by that 80 MB/s QLC speed. Things changed after that first run (and an additional 5-minute idle period), and every subsequent run saw full speed for subsequent 60s writes.
We're not exactly sure what happened with the 4TB here, as the cache behavior was radically different from the 1TB model. There appeared to be no dynamic caching at all, never exceeding 6GB. The cache became more inconsistent during the runs towards the center of the chart (this is expected as there are shorter idle periods). The 4TB SSD was only half full during this test, so there should have been more than enough available capacity to absorb 60s passes at full speed, but that just didn't happen here.
Endurance is naturally lower than that of the EVO, but still plenty for typical client usage. Warranty comes in at 3-years.
- 1TB – $150 ($0.15/GB)
- 2TB – $300 ($0.15/GB)
- 4TB – $600 ($0.15/GB)
Now we're talking! 15 cents per GB MSRP for *any* decent SSD is certainly music to my ears. Sure the performance is not Pro-like, but at those prices, it's not hard to justify some concessions. I would like to see caching performance improve on that 4TB model though, and if Samsung hopes to push more QLC product, they will need to improve those low QD random read latencies.