I must be missing something
Using Optane as cache for spinning rust? Talk about hype exceeding reality. Remember when Xpoint was supposed to be the SSD killer?
Intel has announced Optane memory products in M.2. format to ship in the second 2017 quarter. Optane is Intel’s brand for its 3D Xpoint memory, jointly developed with Micron, and a non-volatile memory that is faster than NAND but slower than DRAM. Intel says (PDF) “a hard disk drive coupled with Intel Optane memory affordably …
"Agreed. Why would i stick 16GB (or 32GB) of anything on an M.2 slot, when I can just put a 500GB SSD on there for 150 'merican pesos?"
Did ANYONE read the article.
It acts as a cache between disk and main memory / processor.
So yes, you could use your 500GB SSD for $150, but how much is that 4TB SSD again?
Big fucking whoop, you can just use DRAM or an SSD for that. For this to be good, it's got to be either cheaper than SSD and available in higher capacities, or vastly faster. This is neither.
What is revolutionary/interesting about a 32GB cache module that performs at the same level as an SSD?
This really frightens me. Is this why Intel hasn't been publishing hard numbers (latency, IOPS, etc.)? Do we know the price of this? It makes me worried that Xpoint may not live up to the hype... Hopefully things will be more clear when we see the datacenter version of Optane.
I had to research this for a job assignment. According Intel's numbers, for a copy from an internal drive to an external drive. the NAND SSD to NAND SSD copy speed was 284MBytes/sec while the OPTANE SSD to OPTANE SSD copy speed was 1.93GBytes/sec.
This was hard drive storage.
NAND SSD to NAND SSD copy speed was 284MBytes/sec while the OPTANE SSD to OPTANE SSD copy speed was 1.93GBytes/sec.
I suspect manufactured numbers there (by Intel), because that would be the slowest SSD I've ever seen. Good PCI-E/M2 SSDs (as they are basically the same thing) have read speeds >2 GB/s and write of 1.5 GB/s.
It would be interesting (and I guess we will see soon enough when they actually start shipping them) to see comparisons with equivalent devices; if they're saying an Optane M2 with PCI-E is faster than SSD, it should be compared against an SSD connected over M2/PCI-E.
Microsoft's Surface Studio is a case in point, uses a combination of 5400rpm spinning rust and an M.2 cache. A replacement SSD for the 5400rpm HDD has been shown to increase the Desktop's performance massively. At least we should be thankful you can replace the HDD, given the way Apple are taking things of late.
Intel must think its 2012 still.
I have to admit, I really wanted to be excited ... but the more Intel does (and does not) reveal about Xpoint, the more I think someone there really missed the boat. And the dock. And the coast.
An M.2 card that small? As just a cache? A cache to ... what? It's not like M.2 slots grow on trees. Is this new Xpoint cache to a RAID5 of plain spinning rust somehow going to be significantly better than the M.2 NAND and a RAID5 of hybrid SSHDs that I already have?
Personally, I'd rather use high density DIMMs and give up a pair of RAM slots for said cache. At least then I'm not losing PCIe lanes that my SLI graphics need. (Or am I?) Is Intel finally going to give us a proper number of PCIe lanes for today's hardware needs?
If Intel has a point to this product, I guess I am clearly not getting it.
We've covered the small capacity. Now let's talk about the large chips.
Look at those beasts. They're huge! If this is what it takes for 32GB capacity, they're going to need 6 or 7 die shrinks before this can fit a marketable amount of capacity on that board. Might as well back RAM chips with capacitors or something.
Best case I can think of is for storing a DB transaction log. Any better ideas?
I did wonder with yesterdays Lenovo laptop article, how a laptop was a sensible place to put this right now (assuming it's price per GB will reduce over time). I'm guessing they're trying to sell to early adopters / CEOs / PHBs?
So, you're taking up a DIMM slot with something 1/5 the size and 1/10 the speed of DRAM, or you're taking up an m2 slot with something 1/30th the size and twice (or more) the cost of an SSD.
I know I shouldn't have got my hopes up for 1TB class DIMMs at only 2-10x slower than DRAM, but one can dream.
Intel's promise with Optane has been that it's NV and doesn't wear like flash (that is, it doesn't require a block erase whose endurance is a few hundred cycles.)
This product is pointlessly small, and certainly no faster than the many NVMe flash products on the market. But if it's write endurance is extremely high, I guess that's a good sign. In the sense that, assuming Intel manages to make it 100x more dense, it would have a write-endurance advantage, if no other, versus flash.
Pretty scummy of them to provide no real info, though. For instance, does it provide standard NVMe, or is it some other one-off interface? Obviously, being M.2 it's just a PCIe device, but perhaps only the Intel chipset recognizes it, and only uses it for caching.
I get all the disappointment, which I share.
However I am starting to see some potential value add their providing in this first-gen tech. Despite lower or comparable IOPS (to SSD), vastly lower storage density, and lower or comparable sequential read/write (to SSD), it does have some characteristics that are easy to gloss over:
* Latencies are much lower -- I can see this potentially being very beneficial for types of workloads that do need very high bandwidth. Think of database logs, perhaps swap files.
* High IOPS @ low queue depths -- This is pretty significant. We're so brainwashed into seeing the latest SSD's with ~300K random IOPS. But if you look at most real-world workloads, queues depths are very low, typically less than 4. If you compare low queues depths of Optane (> 100K), the best class SSD's are only performing in the ~10K or so range.
* Endurance -- Someone else mentioned this, which is also valuable to know, especially for server workloads.
I know I held out hope that gen 1 (while not perfect), would blow away SSD's, but with a price premium of early adoption. The reality is that gen 1 will probably only optimize certain workloads, and while not exclusively be server-oriented workloads, the benefits to the mainstream not be significant. But I fully expect that future generations that break the bus barrier (moving to DRAM interface, or something new), combined with vast cost reductions and thus storage densities closer to par with SSD's, that we'll be in great shape.
Gamers are (some of) the most demanding users. But look at every SSD review in existence, including the highest end server-grade models. Wow those amazing sequential speeds. Now, look at how much they're improving your game performance. Zilch. Nodda. Nothing. Not even load times are better. SSD's look amazing on paper, but in real world workloads, the results vary wildly.