Intel Optane Memory is SRT Cache...Again

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

nzalog

Respectable
BANNED
Jan 2, 2017
541
0
2,160

Yes, I know the difference between Persistent-Memory and nonpersistent-Memory.

You do know that RAM and Storage are interchangeably used as Memory and Storage by the OS?
Like when you run out of RAM and use a Page file? Or when you start Caching files into RAM to access them faster. Or hell, when you use a ramdisk.

There is no requirement for the running state of your computer to be in a non-persistent memory, we just don't have the tech to make persistent memory as fast as non-persistent... yet.

So again, memory is anything that stores data, temporarily or permanently. So... if you're being "pedantic" be accurate. Thanks. Bye!
 

Eggz

Distinguished
When does this intended use case happen? It's an honest question. Intel is marketing this at older computers with spinning hard drives (HDDs), but Optane requires a brand new motherboard and processor running on a 200-series chipset.

Is there even such a thing as an old computer with a brand new guts in it, and if so, wouldn't it usually have an SSD already? Perhaps there are some new, cheap tablets or laptops with an available M.2 slot for an Optane cache.

Or maybe the advertising is really targetted at system builders looking to increase profits. If you build new low-end computers, you'll have a higher profit margin using cheap parts such as HDDs. Pairing an HDD with an Optane gum stick in a low end machines would also justify a price increase for even higher profit margins. I guess it makes the most sense from that perspective, but I don't know how big of a market there will be for that.
 

Morc01

Prominent
Mar 31, 2017
2
0
510


Sorry, but you DON'T know the difference between memory and storage! Memory is what the CPU can directly access, read/write data and execute code from, be it RAM, ROM or anything else. Storage is what stores your data (be it code, or data that the code uses) and the CPU can't access it directly, be it an HDD, SSD, or anything else that retains its content when power is out. A RAM disk is not storage because storage is not an OS level concept. Persistence has nothing to do with the distinction between the two, though storage's definition implies persistence. That Google definition is wrong, I wonder who wrote it. I had to register to right this misinformation that others seem to back. Manleysteel must be old enough to learn the definitions that become blurred and used interchangeably by average users who don't understand the difference. If term memory would cover both RAM and disk why PCs need both and not a disk only?
 

nzalog

Respectable
BANNED
Jan 2, 2017
541
0
2,160


I get that maybe you took a computer science class once and now you can't think outside of that tiny box. I'm still really confused why either one of you keeps talking about age. If it makes any difference, my first computer was a 386SX with 4MB of RAM. It was running windows 3.11 on top of DOS 6.22... (Hope that qualifies me for the age thing now?)

There is nothing special about RAM other than it happens to be way faster than most other memory types, so using it to hold the running state and to talk directly to a CPU just makes sense. However, you are thinking in a itty bitty box of how you were taught to think in. A computer can be designed to run programs from any memory type it does not have to be anything specific outside of what we design it to use. There is a reason why we move from SDRAM to DDR to DDR2 to DDR3 to DDR4 if memory was magical it would never change.

The reason there are these distinctions between things like SDRAM, DDR, DDR4, SSD, HDD, Tape, and floppy is because of the purpose we intentionally assigned to them based on a few things: How fast the memory is, if it is persistent or not, and how expensive for the size/speed combination. Outside of how people design a computer, there is no technical reason we can't use a floppy drive to talk directly to the CPU and hold the running state, it just would be incredibly slow and only 1.44MB worth of memory.

We designed computers they way we want, and we chose to use a specific type of chip. There is no limitation saying we can't design it to use another type of chip. It's all comes down to the "memory controller" which is on the CPU (these days, wasn't always on the CPU though, I know that because age ;) ). That memory controller can be designed to use any type of media, even an SSD.

Take care.
 
G

Guest

Guest


Absolutely ridiculous. You seem to take a specific Harvard architecture paradigm, and extend it to a ridiculous level. By your definition, only part of L1 cache would be memory, everything else is storage, including RAM. But that's your definition, which no one agrees with.

nzalog is completely accurate. You should listen, you might learn something.
 

Morc01

Prominent
Mar 31, 2017
2
0
510


Fine, you qualify for the age criteria, happy now :)

You talk about something that could be, and most probably will be with Optane DIMMs and co. We talk about the here and now (and the past). A CPU can't talk to a HDD directly, or run programs from it, hence it's called storage. RAM it can talk to, hence it's called memory. Something that a CPU can lose and will lose (it's memory) unless it's stored on the storage or rebuilt from data stored on storage. There was a reason these two were distinctly named. A book is not a memory, yet it can store memories. Your brain is not storage, because it'll lose the memories when the light goes out. Although this analogy is slightly wrong because ROMs will not lose data yet the CPU can run code stored on it and storage will not last forever either, but I guess you understand my view. The industry will not call RAM-like devices as storage, and will not call disk-like devices as memory for a few more years purely because how they are used in a computer.
 

nzalog

Respectable
BANNED
Jan 2, 2017
541
0
2,160
A memory controller can be designed to write data into Jello pudding if you design it to that. CPU's use memory controllers to talk to memory no matter what is on the other end.

Either way I'm gonna call it quits here. I'm not getting anything out of this and it's not worth the effort.
 

bit_user

Titan
Ambassador
I think it'd be simpler and cheaper for QLC SSDs to do this internally, not unlike how MLC SSDs got a speed boost from having a SLC cache/buffer.

For myself, I would just continue to use NVMe MLC (or Optane, if it's cheap enough) SSDs for boot. For bulk data, I'll just use QLC instead of HDDs.

Otherwise, I don't see my future changing much until my CPU has HMC or HMB2. At that point, I expect we'll have 3D XPoint DIMMs.
 

Glock24

Distinguished
Sep 5, 2014
91
13
18,635
The Optane cache is supposed to be used in systems that have HDDs, which would be low end systems. But then Intel artificially restricted the use of Optane cache to only i3, i5 and i7. It would make more sense in systems with Pentium or Celeron processors, which are more likely to have a traditional spinning HDD.
 
There are two instances that I hope get introduced into the discussion at some point.

1. Any performance discussion is worthless w/o including the major productivity bottleneck for PC users ... THEM ! When does a new technology become beneficial ?

a) When it improves productivity ?

b) When it gets better scores in benchmarks ?

One can create a MS Office script which links hundreds of keystrokes and measures the time to complete. Was a time when this was relevant when individual task completion was measured in seconds, but today it's microseconds. And the reality is, the time it takes for the user to recognize screen data, make the appropriate decisions and keystrokes in between each command in the script dwarfs the time it takes to complete the script. The test therefore has no real world value.

Bigger, better, faster always costs more but all too often it does not result in a productivity increase....

- An engine rebuilder whose heaviest engine that they handle is 400 pounds gains what from spending more on a 2,000 pound hoist ?
- If I take the Porsche to work instead of the SUV, will I save anything on my commute time in 35 mph rush hour traffic ?
- A legal firm gains what from switching form HDs to SSDs ? Will the secretaries type even one more brief in a day ?
- An e-tailer has 500 CS reps taking on line orders .... will they make any extra sales in a given day as a result of switching to SSDs ?

I has an engineering aide who asked for an SSD to "improve productivity". I asked him to create a "case" for it by analyzing cost to buy, system down time, productivity improvements and dollars saved from increased productivity ... the case couldn't be made. The biggest factor in his analysis was boot time (he assumed 15 seconds, we measured 5.6) but the reality was, he arrived in the morning, started the PC, took off his jacket, got himself some coffee, chit chatted a bit and sat down at his desk several minutes later.

The reality is in most tasks, the user remains the bottleneck and the system ... benchmarks opening 50 windows in Chrome with various hardware changes are meaningless when you are never going to undertake such a task in the Pcs life.

On the other hand, for many work station tasks and video editing for example ... the user can execute a command and it can take seconds, minutes even hours to complete. Here benchmarks have a direct correlation to productivity.

2. When talking about RAM versus Storage, I remember using AutoCAD in the late 90s when the investment in faster SCSI drives was worthwhile as it did affect productivity (Geez, brings to mind our 1st 1 GB HD was $1,000). This was because, despite whatever amount of RAM was present, AutoCAD forced writes to disk before undertaking many tasks. Today it doesn't do that to the same extent tho it must in order for the time save and undo functions to work. But now for example, even opening a 8 MB file takes exactly the same amount of time where that file is on a SSD, SSHD or HD.

In short, a technology is only an advantage when it actually benefits the user. Benchmarks all too often have little impact on PC performance actually performing every day tasks. The kind of examination that would prove enlightening is taking a typical enthusiast build and seeing how that type of user's productivity would be improved with the technology under evaluation. Using a video editing hobbyist for example and starting with a base system reasonably free of bottlenecks ....

16 GB 2400 => 32 GB 2400
16 GB 2400 => 32 GB 3600

SSD w/ OS and programs + 2 TB HD => 2 TB HD + 32 GB Octane
SSD w/ OS and programs + 2 TB HD => 2 TB SSHD w/ 32 GB NAND

I don't case what a disk benchmark says... tho it is certainly a reviewer's obligation to present the results for those that do. But w/o a means to assess "real world" impacts using apps that I use, a reviewers article and testing is something I won't read past the index if I can't assess the actual impacts.
 

firefoxx04

Distinguished
Jan 23, 2009
1,371
1
19,660
Ive been saying this all day, this is NOT MEANT FOR GAMERs nor is it meant for "enthusiasts"

This kind of product will work well for a ZFS ZIL. ZIL does not need a while lot of space, and in the past getting something super fast in 8GB form would cost OVER $1000. Intel has a very solid product here to fulfil that very features on the cheep, so cheep that it should no longer be overlooked as an option when deploying ZFS in the home or small business.

IT blows my mind that people think intel is marketing a low capacity storage device to gamers or end users. Its just cheep enough to where these people actually think they are the target audience. Not even close, but its so cheep you could buy one if you really wanted to.
 

TDRare

Distinguished
Jun 1, 2016
6
4
18,515
My main concern with the "caching concept" is that with 32gb of cache on my 4tb drive, I'm going to have a significant miss rate, even with the 500gb or so that is frequently used.

And with the "small boot volume" idea, we run into the stubborn problem of user files etc., that Microsoft stubbornly refuses to give us an OS designed solution for. Windows 10 still expects users' files on C and recommends against doing otherwise, and even if we move common folders to other storage, Windows continues to place many user data files on C, causing a continual battle to manage small boot volumes. Many Windows program installers, even when pointed to another hard drive, will install many components to C.

For the Windows user base, we need the Windows OS to natively support advanced setup where we can choose or OS drive, our program drive our data drive, etc. Windows Isolated File Storage system, that many programs use, stores everything deep within C:\Users\"username"\Appdata\etc.

My two cents, YMMV.
 

alextheblue

Distinguished

RAM doesn't replace faster storage (yet). RAM is volatile. You could have 128GB of RAM but when you read/write large chunks of data to non-volatile storage, the storage device is the bottleneck. With that being said, Optane is currently overpriced and undersized.

This has been done, very old idea, never worked very well for a number of reasons. Mostly relating to the costs and non-volatile nature of RAM.

Again, RAM doesn't replace storage. The second a program needs to read or write anything to non-volatile storage, you're storage-bound and throwing gobs of RAM in your system isn't going to cure that. With that being said, yes, Optane is overpriced and for most scenarios a decent SSD is plenty good.
 

alextheblue

Distinguished

Yeah a HDD is great for loading up Solitaire and browsing the web. The second you run intensive games or do other disk I/O heavy tasks, the HDD falls on it's face. After using an SSD system for a while, I can barely tolerate spinning disk systems. They're annoying to spend any length of time on, especially if I'm trying to repair or clean someone's HDD-only system. I bought my niece an inexpensive HP laptop for her birthday, and after making sure it was reliable (RAM testing, mostly), I cloned and then ripped out the WD Blue 1TB and slammed in an Evo. Man what a difference, especially when multitasking. Oh, and I ran Samsung Magician to make sure OS settings were SSD-kosher... said everything was good to go already. Nice to know! In Win7 I definitely needed Magician to adjust settings from the cloned HDD OS install.
 
Status
Not open for further replies.