In the recent release of some new SSD products, especially popular and entry-level new products, we will see a somewhat unfamiliar word appear frequently, that is, HBM cache. Some responsible vendors will specifically mention that this is a technology that uses system memory as a cache. We often said SSD onboard cache and analog SLC cache before, what is the HBM that came out this time? Why not use powerful HBM memory? The result?
First of course, Xiaobian should explain that the full name of the HBM cache is Host Memory Buffer (host memory buffer) and HBM memory (High Bandwidth Memory, high bandwidth memory) is only abbreviated the same, purely coincidental. In addition, it must be mentioned that this technology is actually not new, in the NVMe 1.2 standard, the current has long entered the NVMe 1.3 + PCIe 3.0 era M.2 SSD is almost all supported, but it is not open, not publicized.
The function of HBM cache is actually similar to the on-board cache and SLC analog cache, in addition to the data cache, it also stores ftL tables (Flash Translation Layer), which is used to convert from the logical address of the system to the physical address on the SSD, which is convenient for the master to quickly locate and access data. It has recently become a promotional point for entry-level SSDs for a simple reason, these entry-level SSDs have begun to use PCIe 3.0/4.0 high-speed interfaces, and the communication speed with memory is faster, so it is more practical.
In terms of performance, we should pay attention to the fact that the vendor's publicity will mention that there is an HBM cache, so how much IOPS is reached, not to say that the storage speed is increased, from this point of view, the main content of the HBM cache should be the FTL cache that has the greatest impact on the IOPS, and the data cache should rely on the virtual SLC cache built into these SSDs. So we mainly use IOPS to investigate, the model selected is Toshiba RC100 240GB and 480GB.
HBM's improvement in SSD performance is still obvious, but after the amount of data is higher than 32GB, its performance quickly disappears. The 480GB model has higher performance after turning off HBM, which should borrow a small amount of built-in cache of the main control chip, which is of course much faster than the system memory. Since the HBM cache settings of Windows and SSDs are opaque, no one knows how much is actually used, and the performance of different capacities can refer to the official data released by Western Digital.
From here, the HBM cache for FTL is higher than 16MB The effect is very good, and then increase the capacity is not much changed, I believe that most of the SSDs and Windows can still reach this level, so it seems that HBM can no longer improve performance after 32GB, it should not be a problem of cache capacity, but it is more likely to be the limitation of the master controller or even the NVMe standard. Hopefully, with the improvement of the ability of the PCIe channel, the future operating system and SSD can set a higher cache to bring stronger performance.