Should i buy primocache?

[Space Lynx]

I don't understand why anyone would do this. RAM is dirt cheap. Just buy bigger ram sticks.

Even DDR5 is cheap now, top of the line 2x16gb (32gb total) 6400 cl 32 ram is only $149 shipped and sold by amazon.

 

[80251]

primocache's use isn't only as an SSD L2 read/write cache for HDD's, one use case noted by primocache is using it as an L1 read/write in-memory cache that functions more as a non-volatile RAM disk (because it's contents are stored on shutdown). I don't remember the rationale for this use case though but it's on primocache's website.

 

[chrcoluk]

I don't understand why anyone would do this. RAM is dirt cheap. Just buy bigger ram sticks.

Even DDR5 is cheap now, top of the line 2x16gb (32gb total) 6400 cl 32 ram is only $149 shipped and sold by amazon.

Delayed write is probably as stated by others the most important feature for reducing wear from frequent log/journal writes.

NTFS even with write caching enabled flushes writes after 5 seconds now, so you could have all the ram in the world but it wont do much. This is a means to get round those restrictive algorithms.

I think I am going to try this out on my laptop and a VM and monitor the SSD logged writes to see the impact, it goes without saying I wouldnt use the feature without a UPS or battery backed up storage.

 

[Mussels]

If you do buy it a couple things to note is something like a virus scan can push files you had cached out of cache until accessed again. Similarly, having 7-zip calculate folder hash will put files in that folder into cache. If caching a HDD, using an SSD as L2 cache works well for performance boost, even SATA will give a huge bump in random reads. I'd recommend an SSD you don't care about having a lot of bytes written to if using L2 caching.

I found delayed write reliable, it won't let windows shutdown/reboot while it's flushing cache unless you do a forced shutdown. If you have occasional shutdowns due to power loss or unstable OC then I wouldn't enable delayed write.

That about sums up my experiences too
The thing is that if it caches things scanned by your AV, is that most AV scans only do specific file types like .exe's so this is a non issue unless you do full disk scans daily or something


Was on their forums looking around and found this old post that matched my findings, if you have a UPS and are willing to risk data loss (don't use this on an unstable machine!) you can severely reduce writes to an OS drive SSD

The longer the write delay the more you can save, since many of those writes are temporary and discarded (think temporary images loaded from a web page, they never needed to be saved long term) - but the more risk you have of missing data in a crash

 

[80251]

@Mussels
Does the deferred write (i.e. to L1 cache) end up saving you power (because there are less writes to the SSD)?

 

[lexluthermiester]

@Mussels
Does the deferred write (i.e. to L1 cache) end up saving you power (because there are less writes to the SSD)?

Not as such, but saves on SSD wear to a certain degree.

 

[Mussels]

@Mussels
Does the deferred write (i.e. to L1 cache) end up saving you power (because there are less writes to the SSD)?

I wouldnt assume so, as SSD writes are very cheap
Best case with saving 10% writes you MIGHT save 10% power - but thats assuming the drive has a perfect idle and wakeup in that time period you chose for the delay, so i'd assume less.


Edit: On their forums, someone stated that windows write buffer only lasts two seconds - enough to mitigate the worst case scenarios, without excess RAM usage


I reinstalled it yesterday, 1GB read and 10s delayed write cache on C: and 32GB read cache on my 2TB games NVME - played SC2 and BL3, and C: is mostly edge, discord and steam (but not the games themselves)
One important thing with primocache is that if the built in windows cache has the content, primocache wont register the cache hit


Hits on the game cache: poor so far. But i'd expect that, with changing games and a new cache.

The C: drive is doing as expected: reducing the writes (That trim effect before it writes is magical, MS needs to implement a buffer like that themselves)
Reads only helped 8% of the time (Meh)
Writes? 19.7% reduction. That isn't something to write off (hah) as meaningless and proves a small write cache is often more beneficial than a read cache.
28,961 writes that were trimmed before needing to be written on a 10 second delay - that'd be larger with a longer delay, but with more data loss risk.

The settings used:

I don't recall the specifics of the write mode options, i'll find them out in case they're relevant to anyone here

Simple enough with that information.
Native? Waits 10 seconds, writes. (This obviously still saves a lot of wasted writes even with just 10 seconds)
Intelligent: Aims to keep the cache under 90% if the PC is idle, then follows native rules
Idle-Flush: Holds onto the data until the PC is idle, if it can
Buffer: aims to keep cache between 40% and 80%
Average: writes smoothly over time to prevent sudden disk usage (probably good for RAID arrays and the delay mech disks spinning up)
Slow writes? Use the average mode.

An example of how that works with the "average" mode is with a 120s time, it'll math out the data in the buffer and write it as a speed that takes 120s to complete the write. If more data is added, it speeds up.


My systems rock-stable at present, so i'll try intelligent mode and see if i get better than 20% write reduction still on 10s over the coming days.

 

[Klemc]

I use ImDisk, it's free :

i set "Web browser" "cache folder"" in RAM, it works well (needs command line or cache folder directory config depending if web browser's Chrome/Mozzilla).

I've 16Go RAM on my DeskMini, RAM is allowed dynamically, others options too (TRIM...).

 

[80251]

The problem with RAMdisks is that they're static, whereas primocache, properly configured can dynamically create read/write caches in both RAM (L1) and fast NVME M.2 SSD's (L2).

I use a fast M.2 NVME SSD (sk hynix gold 1 TiB) to provide a read/write L2 cache for my slow, QLC, samsung QVO 870 and a 1GB L1 cache.

I don't really have enough system RAM to experiment much with primocache's L1 write deferred cache or the necessary UPS to make its use less risky. I also don't trust my memory overclocks. As Mussels has noted though it can really reduce the number of writes to SSD's. I'm thinking of expanding my L2 write cache because a surprising amount of data is getting written out to the 4 TiB samsung 870 QVO. The downside to my use of primocache is that my L2 cache is so big it has a large system memory footprint (11.95 GiB) because I have 670GiB of L2 cache for my games/applications SSD. My L2 cache hit rate is excellent though at 84.4%. I'm thinking of reducing the size of my L2 cache R/W: 602.11GB/66.90GB since most of it is going unused and creating an L2 write cache for my OS SSD.

 

[Klemc]

I don't understand anything of all that (L1-2 ! CPU, RAM, SSD ???..., don't answer), except that overclock RAM breaks Primocache

Buy bigger RAM, why not, if DDR4 it's cheap, because it looks you use a lot this app.

 

[Mussels]

After another day-ish of use:

Games drive cache hits are above 50%

Windows drive has slightly reduced the writes going from 10s to 15s

I don't understand anything of all that (L1-2 ! CPU, RAM, SSD ???..., don't answer), except that overclock RAM breaks Primocache

Buy bigger RAM, why not, if DDR4 it's cheap, because it looks you use a lot this app.

If you have an unstable PC, you have more problems than a RAM cache.

 

[lemonadesoda]

Looks like this thread needs updating! https://www.techpowerup.com/forums/...p-your-hdd-and-or-reduce-ssd-trashing.107670/

 

[chrcoluk]

On my system my writes are about 40/60 ratio to reads on the OS SSD, consumer workload is typically considered mostly reads, but Windows and browsers write so much data, its not surprising. Its a little better ratio on my laptop but still very heavy writes.

About 5-10 gig a day on windows logs, registry writes. browser files, much of it is also doubled up writes due to journaling as well.

There was a push to prevent browsers writing media as you watch, Netflix etc. but I think that feature ended up been abandoned, so even watching a movie in Netflix is written to your SSD as you watch it.

 

[Mussels]

just an update to this:

C: drive sitting around 20% write reduction still

D: drive bouncing 40-60% - i guess windows is also caching some of these games, but not all of them.

No, I dont see any performance differences with gaming - but a 20% longer life span of the C: drive, i'll notice. (and almost anyone can afford 1GB or less for a 10 second write cache)

 

[chrcoluk]

If I understand you right you can choose to use something like a hdd as a write cache not just RAM?

 

[lemonadesoda]

If I understand you right you can choose to use something like a hdd as a write cache not just RAM?

Why not use a floppy disk? Then when it wears out, just pop in another!

 

[lexluthermiester]

just an update to this:

C: drive sitting around 20% write reduction still
View attachment 286187


D: drive bouncing 40-60% - i guess windows is also caching some of these games, but not all of them.
View attachment 286188


No, I dont see any performance differences with gaming - but a 20% longer life span of the C: drive, i'll notice. (and almost anyone can afford 1GB or less for a 10 second write cache)

This is as good a reason as any to use this program.

 

[80251]

If I understand you right you can choose to use something like a hdd as a write cache not just RAM?

I'm sure you could theoretically use an HDD as a write cache with primocache but it would be pointlessly slow to do so. The write cache Mussels is talking about is in RAM and reduces the number of writes to his SSD.

I use primocache's L2 write cache (on a much, much faster 1 TB, NMVE, M.2 SSD) to minimize writes to my slow 4 TB 870 QVO QLC SSD and before that to my 2 TB HDD which died.

 

[chrcoluk]

I'm sure you could theoretically use an HDD as a write cache with primocache but it would be pointlessly slow to do so. The write cache Mussels is talking about is in RAM and reduces the number of writes to his SSD.

I use primocache's L2 write cache (on a much, much faster 1 TB, NMVE, M.2 SSD) to minimize writes to my slow 4 TB 870 QVO QLC SSD and before that to my 2 TB HDD which died.

With async writes it wouldnt actually affect interactive performance, the point of my interest in delayed write cache in primocache isnt performance but to reduce wear. I noticed the L2 in the screenshot for writes which tweaked my interest.

 

[80251]

With async writes it wouldnt actually affect interactive performance, the point of my interest in delayed write cache in primocache isnt performance but to reduce wear. I noticed the L2 in the screenshot for writes which tweaked my interest.

If the async writes were going to an HDD it might because eventually the DRAM cache on the HDD is going to be filled up at which point it'll have to write the data out to the platters and if any writes overflow the DRAM buffer (e.g. as in a sustained write) it'll have to begin writing data to the platters immediately. I used to notice stutters when quick saving in some games to my HDD.

 

[chrcoluk]

If the async writes were going to an HDD it might because eventually the DRAM cache on the HDD is going to be filled up at which point it'll have to write the data out to the platters and if any writes overflow the DRAM buffer (e.g. as in a sustained write) it'll have to begin writing data to the platters immediately. I used to notice stutters when quick saving in some games to my HDD.

Well RAM would still be caching data as well. But I cant think of any workload on the OS drive that would would fill the RAM and local HDD cache, its basically logs, and registry writes. I am not talking about using it as a cache for a games drive, steam downloads or large file copies.

When a HDD is used as an OS drive and is slow, its because of the reads not the writes, unless we talking about the pagefile and hibernation. Those would be the exceptions, also I am suggesting it might be useful in "some" cases not in "all" cases, remember the world isnt black and white. You falling into the trap of because its not a useful idea for you then it means its no good for anyone else.

 

[Mussels]

If I understand you right you can choose to use something like a hdd as a write cache not just RAM?

You use things in order of performance, i'll add example values

RAM (8GB cache)
SSD (120GB SSD)
Mech drives/RAID (40TB RAID array)


This way, writes can come in until you run out of RAM at 100% full speed (my 32GB cache would last quite some time), then slowly shrink as the mech drives wrote away as fast as they could. This would also remove any stutter while waiting for the drives to wake up and get writing.

In that above example the SSD could do write caching too, but primarily would be used for read caching - you want the most commonly accessed files on there. They might be .exe files or .dll's, but since mech drives have poor seek times, it's better to fill it with 10,000 small files than 100 larger files.



As for "hows the cache going"

Here's the day four update:
Games cache is at ~75% hit rate: (Windows 11 may be caching the rest)

Despite being write only, the C: drive cache still managed to have an almost 1% hit rate too - I guess it wanted to read something back immediately after writing it

Oh and writes are down to 71.5%, just a 28.5% reduction

It doesn't matter what SSD you have, a 30% longer life span is a 30% longer lifespan.

 

[chrcoluk]

You use things in order of performance, i'll add example values

RAM (8GB cache)
SSD (120GB SSD)
Mech drives/RAID (40TB RAID array)


This way, writes can come in until you run out of RAM at 100% full speed (my 32GB cache would last quite some time), then slowly shrink as the mech drives wrote away as fast as they could. This would also remove any stutter while waiting for the drives to wake up and get writing.

In that above example the SSD could do write caching too, but primarily would be used for read caching - you want the most commonly accessed files on there. They might be .exe files or .dll's, but since mech drives have poor seek times, it's better to fill it with 10,000 small files than 100 larger files.



As for "hows the cache going"

Here's the day four update:
Games cache is at ~75% hit rate: (Windows 11 may be caching the rest)
View attachment 286493

Despite being write only, the C: drive cache still managed to have an almost 1% hit rate too - I guess it wanted to read something back immediately after writing it

View attachment 286494
Oh and writes are down to 71.5%, just a 28.5% reduction

It doesn't matter what SSD you have, a 30% longer life span is a 30% longer lifespan.

gotcha so its like how ZFS works.

 

[Mussels]

gotcha so its like how ZFS works.

Quite possible - i've not kept up on the full details of all the file systems other OS's are using.

Longer term testing results:

I managed to get a 95% hitrate on the games drive cache playing the same games repeatedly, but i'd left it able to do write caching by accident - so a steam update ate the cache. sad.
Adjusted settings, but that reset the stats.
Load times were already fast, but now it's things like intro videos and loading animations holding me up, not the hardware itself.


My C: drive was updated to a 60 second cache and forced write-only, and it's info's still good to use.

How to interpret this:
Cache is write only, 60 second delay. Despite this it's got a 37% *read* rate where it wanted to read and re-use that data before it was even written the first time (temporary files in browser, is my guess)

There are some deferred blocks (~5%) right now as i'm torrenting a linux ISO (Latest Mint), but the key takeaway is that the total writes are at 65% (70%, once that download finishes)
30% extra lifespan on my 970PRO means it will outlive me.

Oh and the windows xbox overlay confirms it doesnt cause issues with directstorage - this should cache for that too in the future, resulting in some stupidly fast load times
(3x NVME drives, two have partitions)

edit: it seems torrents *really* benefit. I assume because smaller pieces get stitched into bigger ones, the longer the delay.

(and then moving the completed file to the other drive while still in the cache, boosted the read-hit rate. Interesting for use with a mech drive that moves a completed file elsewhere.)

 

[80251]

@Mussels
You're getting a lot out of that 1 GiB L1 write cache!
I have a 1 GiB L1 cache too but it's set to R/W and I think it's why I had some incredible problems with Metro:Exodus saves getting corrupted (my memory passed 10 hrs. of memtest86, 4 hrs. of orthos 4 core testing (large FFT's), a windows 10 memory test and 2 hrs. of prime95 but Metro:Exodus still had problems with my memory overclocks because my saves were being corrupted by the shared L1 R/W cache of Primocache. I upped the V-DIMM and all those issues disappeared.