Thursday, August 29th 2024
Disabled SLC Cache Tested on M.2 SSD, Helps Performance in Some Cases
Gabriel Ferraz, maintainer of the TechPowerUp SSD database and content creator, recently published an article that shows the relationship between SLC (Single-Level Cell) cache technology and its performance impact on SSDs. Using a Pichau Aldrin Pro 2 TB SSD featuring an Innogrit IG5236 controller and YMTC 128-layer TLC NAND, Gabriel has uncovered both the advantages and potential drawbacks of this feature. The article reveals that with SLC cache enabled, which acts as a high-speed buffer, the SSD achieved remarkable write speeds of up to 6.5 GB/s, but only until 691 GB had been written. Beyond that, speeds dropped to 2.2 GB/s and then to 860 MB/s as the drive filled up.
Disabling the SLC cache delivers more consistent performance results that are 2.1 GB/s across the whole capacity of the SSD, but with lower peak performance. Testing also examined the impact on power consumption and efficiency. With the SLC cache active, the SSD consumed approximately 5 W of power while achieving over 3000 MB/s bandwidth. Disabling the cache reduced power consumption but at the cost of halving the bandwidth to around 1900 MB/s, resulting in lower overall efficiency. Maximum power consumption with cache enabled peaked at 7.3 W, compared to a lower figure when operating in constant TLC mode. Below, you can see some performance benchmarks published on The Overclock Page.
Interestingly, in real-world scenarios such as game loading times and Windows boot speeds, the difference between cached and non-cached performance was minimal. Synthetic game benchmarks and Windows boot tests showed negligible variations, suggesting that current software may not be fully optimized to leverage the speed offered by SLC cache, likely due to the prevalence of random 4K operations demanded by software, which NAND flash is not optimal for, rather being ideal for sequential operations. File transfer tests, however, tell a different story. Copying large files and game installations took more than twice as long with the cache disabled, highlighting the significant advantage of SLC cache in data-intensive sequential tasks.
For complete benchmarks and in-depth explanation, check out the original article by Gabriel.
Source:
The Overclock Page
Disabling the SLC cache delivers more consistent performance results that are 2.1 GB/s across the whole capacity of the SSD, but with lower peak performance. Testing also examined the impact on power consumption and efficiency. With the SLC cache active, the SSD consumed approximately 5 W of power while achieving over 3000 MB/s bandwidth. Disabling the cache reduced power consumption but at the cost of halving the bandwidth to around 1900 MB/s, resulting in lower overall efficiency. Maximum power consumption with cache enabled peaked at 7.3 W, compared to a lower figure when operating in constant TLC mode. Below, you can see some performance benchmarks published on The Overclock Page.
44 Comments on Disabled SLC Cache Tested on M.2 SSD, Helps Performance in Some Cases
We already knew the addition of more levels per cell degrades latency and such. It's the #1 reason I still think QLC is pointless. Barely increases capacity over TLC, but comes with all the drawbacks.
Sorry, I couldn't resist.
For example: in M.2 Gen4 4TB (where QLC should logically be the cheapest due to increased capacity and lower prices) the cheapest model costs 200€. The next model after that costs 270€. With TLC there are 10 options between 225€ and 270€. So you go with QLC and take a massive hit to both performance and longevity to save 25€ or 11%?
I looked at 1TB and 2TB too and there the prices are pretty much the same or even skewed towards TLC due to number of models available and availability.
At 8TB there are ZERO QLC M.2 Gen4 models. Or Gen3 models even. Only 12 SATA models. While they are between 150-200€ cheaper than M.2 Gen4 TLC there it's still 1/10th the models and all of them with SATA speed limits and 4+ year old releases.
If manufacturers want to convince people to get QLC then make M.2 Gen4 or Gen5 even (only with new 6nm low power passive controllers) models that crush TLC at every price point in terms of price per GB. 1TB for 30€ instead of 60€. 2TB for 55€ instead of 110€. 4TB for 110€ instead of 200€ and finally 8TB for below 300€ instead of old SATA junk for 500€+
THEN i might at least consider it.
The market has spoken and it wants more capacity for less money, not faster speeds for more money.
www.techpowerup.com/forums/threads/gigabyte-reveals-the-aorus-gen5-12000-ssd.312995/post-5090926
Regarding the possibilities, we have NVMe Zoned Namespaces (ZNS), which allows the user to have some control over "zones" (partitions) and their behaviour, including bits per cell. At least the standard makes this possible, implememtations are another matter.
But jokes aside, it would be nice to have the option.
www.tomshardware.com/news/nvme-ssd-for-audiophiles
Also many of TPU QLC reviews have unfavourable pricing as well.
SourceZNS is pretty much dead, but a similar concept that the industry is pretty much moving towards is (Flexible Data Placement) FDP.
Seeing how the cache behaves when the SSD is, say, 80% full would also be really nice. But I suspect it's impossible to get consistent results in such a benchmark because they would depend on the current state of internal fragmentation, which can't be detected by the OS.
But that does not mean that if the drive sits idle at 1tb used it will no longer have slc cache, just if you write it in one go.
That being said I do wonder how drives benchmark if you assume a default state of 60-80% used as a starting point for all these operations. For me that seems like a typical drive utilization. At that point the cache will be much smaller. The drive will slow down sooner at that point.