StoreMI versus Optane Memory (cache), Optane 16/32GB NVMe versus Patriot Scorch 256GB NVMe?

[VirtualLarry]

Just wondering what the best bang-for-buck is here. Assume system has a 4TB HDD 5400RPM.

Would I be better off, with an Intel B360 board (and then I would have to go with an i3-8100 as a minimum), and an Optane 16/32GB drive, versus a B360 / i3-8100 and SRT with a 256GB NVMe (Scorch for $43) and HDD,

versus an AMD B450 board and an A200GE (and HDD), with either StoreMI and a 16/32GB Optane NVMe, versus StoreMI with a 256GB NVMe Scorch drive.

I mean, Intel versus AMD: With AMD, I can use a low-cost APU, as long as I go with a B450 board. With Intel, I need to go with a B360 board and above, as well as a ($130) i3 or above, to use Optane Cache, and I believe also to use SRT.

Then there's the difference between Optane cache, and a normal 256GB NVMe, and StoreMI tiering versus caching with Optane or SRT and NVMe, with respect to hit rates.

I don't really have any seat-of-the-pants data for either of these scenarios, so I could use some guidance.

The primary use case would be an everyday browser box (with emphasis on Twitch and YT) for a friend with an older box. Said friend doesn't understand drive / storage management. (This is the case with too many of my friends.) So I wanted to provide HDD-type space, with SSD speeds for boot / browser / apps.

 

[cbn]

For Browsing page out, I am interested in how StoreMI (with 256GB NVMe/4TB HDD) compares to Intel Optane cache (16GB or 32GB/4TB HDD).

P.S. Here is some data from Seagate showing not so much nand is needed for cache in a corporate work machine (over 5 days):

https://www.silicon.co.uk/workspace...f-nand-124069?inf_by=5c75ed3d671db8c4238b49bf

 

[IntelUser2000]

I mean, Intel versus AMD: With AMD, I can use a low-cost APU, as long as I go with a B450 board. With Intel, I need to go with a B360 board and above, as well as a ($130) i3 or above, to use Optane Cache, and I believe also to use SRT.

SRT is a separate technology from Optane caching. TH had a test where they paired 600p(NVMe SSD) with Optane Memory using SRT. If you have a compatible system its better to use the official Optane Memory application.

But SRT seems to be in a phase of being discontinued? So I cannot fully recommend this.

You could also go with PrimoCache. Linus did testing on the AMD platform.

If you cache an NVMe SSD, it'll slow it down somewhat because caching creates some contention. The QD1 read on Optane Memory drops to 2/3rds when cached versus standalone for example.

Personally I want to get my hands on a 32GB module and see how it performs but my 16GB module seems sufficient for browsing/media usage.

 

[cbn]

The QD1 read on Optane Memory drops to 2/3rds when cached versus standalone for example.

1.) With that in mind, (and assuming only basic office tasks are being performed*, see graph in my previous post) I wonder how StoreMI with 32GB Optane + HDD compares to Intel Optane application with 32GB Optane + HDD for paging out during browsing?

2.) StoreMI has higher 4K Q1 Read?

*therefore not being a situation where Optane and HDD are not swapping back and forth so much in StoreMI.

 

[cbn]

https://gadgets.ndtv.com/laptops/reviews/amd-storemi-review-1878682

The above review using Samsung 860 EVO doesn't show much drop off in 4K QD1 Read when standalone SSD is converted to a StoreMI fuzed volume. (StoreMI retains 92% of the 4K QD1 Read)

It would be awesome to see Optane also hold 92% of its 4K QD1 Read.

 

[IntelUser2000]

2.) StoreMI has higher 4K Q1 Read?

Every cache slows down. It's nature of how things work.

You can see from Linus's test that StoreMI has lower performance than both Optane Memory and PrimoCache. So CDM isn't showing things but CDM isn't that great of a benchmark anyway.

Also that Seagate analysis is extremely optimistic. Benchmarks show such Hybrid drives being faster than HDDs but still ways away from SSDs and you can easily overcome it in general usage. It's not just the capacity, but because the cache is NAND. You can't use a cache to make it faster than a standalone drive made entirely out of the cache. 3D XPoint is faster than HDD so you can get it close to a NAND SSD, but again you can't get 3D XPoint cached system to be faster than a 3D XPoint drive.

 

[cbn]

Every cache slows down. It's nature of how things work.

StoreMI is actually auto-tier (not cache). Not sure how much difference in overhead there is, but I was thinking it might work similar to standalone particularly if certain data could be pinned? With this noted, I also wonder if Intel Optane application could be made lower overhead (for certain targeted functions) if certain files could be either pinned or removed from the cache and used as standalone?

You can see from Linus's test that StoreMI has lower performance than both Optane Memory and PrimoCache.

Yes. it definitely did have lower performance but their benchmark method did use a rather large data set. This method caused a lot of swapping data back and forth because the fast tier is so small. NOTE: One of the games in Steam Shot below is Fallout 4 ( a game that with DLC is over 100GB in size).

This, in contrast to the usage Virtual Larry is proposing:

The primary use case would be an everyday browser box (with emphasis on Twitch and YT) for a friend with an older box. Said friend doesn't understand drive / storage management. (This is the case with too many of my friends.) So I wanted to provide HDD-type space, with SSD speeds for boot / browser / apps.

 

[IntelUser2000]

I did some reading of StoreMI and PrimoCache as well. People said the latter works quite a lot better.

Algorithms always have an overhead. Unless you do comprehensive testing in real world applications, you won't know the detail. The advantage of StoreMI in VirtualLarry's case is that AMD platform offers it for free and PrimoCache costs $30.

 

[cbn]

This thread (with the predicament of cheaper processor vs. more expensive processor as a factor in the decision making) got me thinking about how in 2017 Intel had a chip called Pentium G4560. (A chip I really didn't like that much to be honest because I was into refurb hardware at the time).

In any event, this was a chip that most major websites didn't bother covering (that much)....but lots of Youtubers made videos with experimenting and benchmarking from various angles.

Maybe Intel should take a cue from the G4560 phenomenon and bring Optane caching to Pentium? This to encourage more experimental benchmarking and potentially practical usage (like RAM Extender*). (Having an unlocked multiplier** and particularly some desirable form of Intel Upgrade service*** would help tremendously too.)

Right now the only video I have seen for Optane as RAM Extender is the one from LTT:

(And I haven't seen any tech websites cover Optane as RAM Extender)

*I did some experimentation myself (just in these forums) with 58GB 800p and 16GB Optabe using AMD A10-7860K and was impressed with the tech for paging out browsing.

**Fairly likely to happen ( I reckon) as there is no current unlocked 2C/4T (last one was Core i3 7350K) and locked pentium is already at 4 GHz.

***for software unlocking to 4C/4T or maybe even 4C/8T too (since Pentium is based on a 4C Coffee Lake die)

P.S. Just thinking how interesting a Intel Upgrade Service 4C/4T (without AVX) or 4C/8T (without AVX) would be for stability testing. Maybe an easy 5 GHz without much cooling?

 

[IntelUser2000]

Maybe Intel should take a cue from the G4560 phenomenon and bring Optane caching to Pentium?

Yea, not only that they should make it work on Celerons, including Gemini Lake chips. Basically every Intel platform. Bonus points if they can work with AMD and let it work there as well.

I think if they want to use volume to lower cost then they should drop the stupid mentality of trying to upsell their Core processors. If they really believe in the 3D XPoint technology, they should aim long term and aim for DRAM-replacement. Start slow. Caches for SSDs, then on small embedded systems, then on chromebooks. Then bigger, and bigger and bigger.

Unfortunately I'm not sure if they realize this or are willing to do so. Side effect of chasing short-term profits.

 

[cbn]

Found by reading this post at Reddit that a person can change the read i/o promotion policy on StoreMI from fast to slow.....with normal or slow working better with small drives:

The default aggressive read io promotion policy works better with bigger and fast drives. It judges promotions once every 2 seconds and promotes stuff as soon as you touch it with a read iop. Setting it to normal or slow can help if the drive is really small, but I think about 200gb or even 120gb should be good enough for very normal use cases.

 

[cbn]

Some users in this reddit thread reported problems with using NVMe as fast tier for StoreMI:

Another user posted the technical explanation for this specific issue a few months ago - basically, if your computer crashes during certain write operations involving an nvme drive using storemi, the whole thing goes corrupt and it's a mess even trying to get it formatted back to square one. That happened to me exactly during a file copy while my system wasn't totally stable (I had oc'd manually too high). It's specific to the way nvme interacts with storemi, has nothing to do with the drivers. So if you think your computer will never ever crash during a write operation, you're fine. Otherwise, just use the nvme on its own.

However, same user that wrote the explanation above did report StoreMI works great with SATA SSD:

Only update I have to offer is that it works great with a SSD/hDd combo, not so well with nvme

 

[VirtualLarry]

Interesting about StoreMI, NVMe, and unstable hardware. I'm running 2x 660p 1TB NVMe drives in RAID-0 (B450 NVMe RAID), and I've had a few hard-freezes, and so far, nothing's corrupted so much that it wouldn't boot, much less re-format. But I did, when I created the array, specify "write-through", and NOT "write-back", for this very reason.