Drive caching

[ASXCyphin]

I'm a bit confused about the actual performance of caches on modern HDDs, SSDs , and NVMes.  Does anybody feel like shedding some light on the topic for me?

I understand the concept.  A drive has a cache to catch a bunch of information until it can be written to the HDD or SSD or NVMe.  The cache operates at speeds faster than the non-volatile memory or spinning platter operates at, but eh cache is limited in size and is volatile memory.  If the drive loses power, all the information in the cache is gone.

Are there memory standards for drive caches?

Does the volatile cache memory in a SATA III 5400RPM HDD have to meet a minimum speed requirement to be considered SATA III compliant?  Is that requirement different for a 7200 RPM drive?

Does the volatile cache memory in a SATA III SSD have to meet a minimum speed requirement to be considered SATA III compliant?  

Does the volatile cache memory in a NVMe have to meet a minimum speed requirement to be considered PCI-E compliant?  

Are there standards for drive caches?  I can't find any written standards myself, and I'm wondering if my wording is wrong and that's why I can't find what speeds drive caches work at, or if that sort of information is "insider" information, and is too niche for most users to even think about, thus the question hasn't been answered  yet.

[ASXCyphin]

Is this an LTT labs question?

[Electronics Wizardy]

28 minutes ago, ASXCyphin said:

A drive has a cache to catch a bunch of information until it can be written to the HDD or SSD or NVMe.  The cache operates at speeds faster than the non-volatile memory or spinning platter operates at, but eh cache is limited in size and is volatile memory.  If the drive loses power, all the information in the cache is gone.

I wouldn't say a bunch. This really depends on the drive, but there are drives with less than 1 mB of cache, and they work fine. Most of the dram on a ssd is used for the table that connects the logical sector to the location on nand.

 

30 minutes ago, ASXCyphin said:

Does the volatile cache memory in a SATA III SSD have to meet a minimum speed requirement to be considered SATA III compliant?  

Nope, no requirement, it just has to be able to communicate following the sata III protocol, same with nvme. The protocol doesn't care about the cache on the drive

 

 

33 minutes ago, ASXCyphin said:

Are there standards for drive caches? 

How exactly, and why would a standard matter. It only talks to the controller, so it doesn't matter how it connects.

 

 

[ASXCyphin]

On 8/18/2022 at 7:52 PM, Electronics Wizardy said:

I wouldn't say a bunch. This really depends on the drive, but there are drives with less than 1 mB of cache, and they work fine. Most of the dram on a ssd is used for the table that connects the logical sector to the location on nand.

 

Nope, no requirement, it just has to be able to communicate following the sata III protocol, same with nvme. The protocol doesn't care about the cache on the drive

 

 

How exactly, and why would a standard matter. It only talks to the controller, so it doesn't matter how it connects.

 

 

It would solve a matter of personal curiosity for me, that's all.  Say, for instance, the cache on one older drive uses... DRAM similar to LPDDR3 @2133mhz, and a new drive comes out using a cache operating like DDR4 @ 3200mhz, one would think there would be a small increase in performance for the drive with the higher operational frequency, and be better for transferring lots of small files.

It's just something you never really see discussed and of course would explain why Samsung NVMe's and SSD's are so good since Samsung RAM also tends to be some of the highest performing ram on the market.

[Kilrah]

Would be pretty much impossible to infer the performance of the drive by what type of cache it uses, hence why it's not mentioned. It's specs and tests of the whole drive that matter...

 

Last thing we need is another opportunity for manufacturers to misguide by just slapping "DDR4 Cache!!1!1" on an SSD that's actually terrible to make you think it's good.