Is CPU cache different from SSD caching?

[Xdrone]

1 Isn't the cache of a cpu what feeds information into it? Is this correct?

2 And how is this different from SSD caching found on SSHDs and Intel optane?

3 How do SSHDs work? Is the cache (DRAM) of an SSHD the same as SSD caching or CPU caching (feeding info to the HDD part).

[JaegerB]

1 and 2 I'm not an expert on, so I'll leave that to the experts. for 3, an SSHD is basically a normal HDD with a small SSD on it that uses an algorithm to figure out what files you use the most. These files are kept on the SSD for quicker access, while more uncommonly used files are left on the HDD. 

[Gameking002]

cache of a cpu hold whatever the cpu is processing at the moment.

the cache of a ssd holds data that is moved around. ssds move data around to faster parts of the ssd to keep performance higher.

sshds move the data that is most used by the user to a ssd inside the sshd so that data can be accesed faster so applications the user uses alot will work faster than others

DRAM is not the cache of an sshd, it is the data that a cpu needs to access for the current application, but is not processing at the current moment.

[Xdrone]

4 minutes ago, Gameking002 said:

cache of a cpu hold whatever the cpu is processing at the moment.

the cache of a ssd holds data that is moved around. ssds move data around to faster parts of the ssd to keep performance higher.

sshds move the data that is most used by the user to a ssd inside the sshd so that data can be accesed faster so applications the user uses alot will work faster than others

DRAM is not the cache of an sshd, it is the data that a cpu needs to access for the current application, but is not processing at the current moment.

Oooh, so the act of SSD caching is moving data to keep performance high. And in the case of an SSHD - moving more frequent files to the SSD portion of the drive, which is the equivalent of HDD + optane (HDD + optane = SSHD), everything here correct?

[Jurrunio]

1. Cache is like a short term quick storage for the CPU that holds information on commands that will be needed within a short time in the future. So yes, cache feeds the CPU with info, but it's not the only one doing so.

 

2. CPU cache and SSD cache in SSHDs work on different components of a computer, though they do similar things: Hold the commonly used stuff for faster access. Intel Optane is more like a super fast SSD with low capacity, rather than a cache.

 

3. SSHD has a small SSD inside that holds commonly used info for faster access. CPU dont send info to the HDD or SSHD directly. It creates / receives info and dumps into the RAM, then that sends info into the HDD / SSHD.

[Gameking002]

Just now, Xdrone said:

Oooh, so the act of SSD caching is moving data to keep performance high. And in the case of an SSHD - moving more frequent files to the SSD portion of the drive, which is the equivalent of HDD + optane (HDD + optane = SSHD), everything here correct?

yes, thats correct

[Gameking002]

this will maybe make it clearer (if it wasnt clear enough yet), you have these levels of storage. the input sources are basically mouse and keyboard. i wont be focusing on those at the moment

then you have the SSD/HDD (ROM storage). whenever you start an application, data gets pulled from the ROM into the RAM for quicker access by the CPU.

then the cache is what the CPU is processing at that specific moment. (for example a few lines of coding). then part by part those lines of coding get fed into the CPU register for the CPU to process. once it is done processing it sends that data back where it needs to be. and eventually when you save a file and/or exit, the saved file will get back to the ROM (SSD/HDD) for the next time you start that specific application. also options and preferences get overwritten from the previous data that was there for the application so you always have the options/preferences that you set.

 

[xentropa]

1.)

 

Cache refers to any memory within the CPU.  There are 3 levels.  L1-L3.

 

To the best of my knowledge, L3 cache is shared memory across all cores.  This might be where information from the DRAM is directly queued.  

 

L2 cache is specific to each core, where I think it holds information prior to the x86/x64 decoding.

 

L1 maybe the latches or the registers in the execution core.

[Indus Monk]

44 minutes ago, Xdrone said:

And how is this different from SSD caching found on SSHDs and Intel optane?

The only difference i know of, would be that CPU cache is volatile, but SSD  cache and optane memory is not volatile

 

[NumLock21]

2 hours ago, xentropa said:

1.)

 

Cache refers to any memory within the CPU.  There are 3 levels.  L1-L3.

 

To the best of my knowledge, L3 cache is shared memory across all cores.  This might be where information from the DRAM is directly queued.  

 

L2 cache is specific to each core, where I think it holds information prior to the x86/x64 decoding.

 

L1 maybe the latches or the registers in the execution core.

Cpu has the fastest access to L1 cache,  when it goes to L2, it gets slower and even more so when it reaches L3. L1 is also the most expensive to manufacture, which is why up to now, L1 cache is always the least amount. Back then you can actually run test to see how cpu cache affects performance. The bios has the option to enable or disable them. 

 

[Mira Yurizaki]

5 hours ago, Xdrone said:

1 Isn't the cache of a cpu what feeds information into it? Is this correct?

Yes and no. Cache holds data that the CPU frequently asks for from main memory (or RAM). The idea is that the more often the CPU asks for the same data, the closer the access time and transfer speed gets to cache speed.

Quote

2 And how is this different from SSD caching found on SSHDs and Intel optane?

SSD caching is still considered to be in secondary memory. The processor will not touch it unless what it wants can't be found in main memory. However, the idea is still the same: the more times you access the data from the slower secondary memory, the closer its access time and transfer speed gets to the SSD's performance.

Quote

3 How do SSHDs work? Is the cache (DRAM) of an SSHD the same as SSD caching or CPU caching (feeding info to the HDD part).

The drive monitors what data you've been frequently accessing and makes a decision based on some algorithm to either keep the data in the SSD portion or not.

 

Basically the whole gist of this is caching is a way to add smaller amounts of faster memory to improve the performance of slower memory. The amount of data you actually need is tiny compared to the amount of data you can store permanently. So by prioritizing the small bits of data you use often, you can get much faster performance for not a whole lot more in cost. There's also the fact that if your cache size is too large, the amount of time you spend looking if the data is even in cache is too much and it would've been comparatively as fast to just look it up in the slower memory anyway.

[System Error Message]

a cache is a cache regardless of where. It holds information for faster access for where it is placed. A drive cache holds information for the use of the drive. It could hold information about where different pieces of data is so its faster to find or even data mostly used on the drive. A CPU cache holds data for CPUs to use when needed without having to wait. For example if you have multiple cores working on the same thing its faster to draw from the cache than it is to wait for the data to be sent from one core to ram and for another core to get it from ram.

 

RAM is not cache, it is a storage device. a CPU is flexible, it can use data from anywhere, be it ram, drives, even network as modern CPUs have flexible data controllers to manage data.

 

A practical example would be a physics simulation. The algorithm is the same, never changing so it can be stored in cache. The objects are reused many times so that too can be cached as long as there is space otherwise the ram can continously keep sending that data to the cpu asynchronously so the CPU isnt kept waiting. The CPU can then keep working the loop despite not having enough registers and storage for all the pieces, not to mention that the results will have to go somewhere like back to ram, display or even storage. This is one of the easier applications where cache use is helpful and less chance of a cache miss. Another example would be code compilation. Any application that reuses code and data many times will benefit from cache for processors (even on GPUs too which nvidia strips from consumer cards) which is why AMD cards do general purpose compute faster. These applications benefit from having a bigger CPU cache (some games too)

 

Some applications like networking dont benefit from caching except for tables. In networking packets have to get from one place to another as fast as possible so caching data or packets are a bad idea but its a good idea to cache connections and tables of routes. Though on the internet these things get so big that CPU caches cant fit everything so ram and storage are used too hence why on the professional level they use x86 servers or routers with DIMM slots so you can upgrade the ram or get the ram amount you need because certain applications like BGP store internet routes which would be good to cache but is just too big for any processor cache (it can run to gigabytes for small companies). This is why consumer routers have kilobytes of CPU cache for the CPUs they use as they only need to store the code for processing packets and not any data. This sort of use doesnt benefit from having huge cache, even for mobile CPUs where power efficiency is more important than instant access.

 

a processor cache is a lot faster than ram in latency. Ram bandwidth is on par with the highest level cache of the CPU but not in latency. In drives RAM can be used as cache since it is a lot faster but also a lot cheaper than CPU cache per capacity. So theoratically you could have a hard drive with a sodimm slot and have many gigabytes of ram as cache but with drives you risk of running into the issue of not enough time to dump the cache in case of power failure if the cache is too big.

 

CPU cache can have many levels not just 3 (reffering to above posts). Sparc had 4 or more levels. I think 6 levelsis the highest known to CPUs. Theres no such thing as a limit to cache levels or size, but there are limitations such as power and cost. A 1GB CPU cache would be very expensive, consume a huge die space and have low yields.