On Friday, January 15, Intel announced the discontinuation of certain of the company’s Optane SSDs for consumers PCs. Naturally this is making Optane users curious about the future of the entire product line. Is this a big move?
In a word: “No.” It’s a relatively small part of overall Optane shipments, and it is probably the least profitable part of the overall Optane business.
First, let’s look at why Intel is making Optane SSDs in the first place. Those with good memories may recall that I wrote a blog post way back in 2016 called Why 3D XPoint SSDs Will Be Slow. This was even before Intel announced the Optane name. The crux of that post is that an SSD isn’t a good place to put 3D XPoint memory because the SSD interface bogs down XPoint’s 1,000-times speed benefit enough that it is nearly all lost. Optane SSDs, which cost about half as much as DRAM, or 5-10 times as much as a NAND flash SSD don’t provide consumers a big enough performance advantage to warrant the additional cost.
But Intel’s strategy was not focused on the SSD. The company’s goal is to couple Optane DIMMS (formally: The Optane DC Persistent Memory) to its server processor sales to provide an important cost/performance advantage over its rivals. The DIMM format uses a very high speed DDR4 bus, and this allows XPoint’s true potential to be realized. I explained the benefit of Optane DIMMs in great depth in my most recent 3D XPoint report.
To make Optane viable in the server market it must be priced below DRAM, and production costs can only be made sufficiently low if the “Optane Media” (3D XPoint memory) can be produced in high volumes. Intel wanted to create demand for 3D XPoint to drive its cost down, and SSDs appeared to be an easy way to do that. In the meantime the company had to suffer losses on this product line. This seems to be working to some extent. See the post on The Memory Guy: Did 3D XPoint Costs Reach Break-Even?
But given that Intel’s whole reason to make Optane SSDs is to boost factory volume, then the net benefit to Intel will be increased by growth in chip volumes and reduced by any support requirements.
The SSDs that Intel is phasing out require a lot of support. Since they are aimed at the broad consumer market – they must go through rigorous tests to assure backwards compatibility with older CPUs and systems. They are also being sold to consumers, rather than OEMs, and the higher level of interaction with consumers increases the company’s support costs.
Meanwhile, these SSDs appear to be selling in relatively low volumes. These low volumes, and the SSDs’ low capacities, as low as 16GB, have a multiplicative effect: Not many chips are driven by this particular market. The company would do better to sell one tenth as many SSDs to the data center, each with more than ten times the capacity, than it would to continue to sell smaller capacity SSDs direct to the consumer.
This move isn’t that surprising. Only a month ago Intel announced a number of new NAND flash and Optane SSDs, many of which were a follow-on generation of existing first-generation Optane products. During this roll-out Intel did not announce any upgrades to the products that are now discontinued. Consumer focus was shifted to the H10 and H20, which are hybrid Optane/NAND SSDs.
In the end Intel isn’t importantly reducing its chip shipment volume, and it is grooming itself to have fewer Optane losses by taking this step.
Hi there.
Just for better understanding to a non storage speciallist like me … maybe we could use a metaphor as like … NAND Flash NVMe SSD could be a ‘calash’ … in its way to the town (CPU) through a very bumpy cobblestone road (PCIe 4.0). On the other hand, RAM DIMMs could be Red Bull F1 racing cars ‘flying’ also to the town over an inmaculate race pavement (DDR5 memory channel). Finally Optane ‘modules’ maybe could be Indy cars … which have the option … in their way to the town, of using the ‘very bumpy cobblestone road’ or ‘inmaculate race pavement’ …
Well … I am not pretty sure if you could consider the previous as a more or less adecuate metaphor … … For sure you could improve it with at least a couple of expert ‘touches’ …
My very ‘wet’ dream : Installing my CentOS linux distribution over an Optane DIMM PMem module … on my laptop … 512GiB would be enough.
If for some minutes we discard the ‘money factor’ … Could too much power disipation stop the show? I am thinking in an Alder Lake Xeon ThinkPad ‘mobile workstation’ …
If I am more sensitive to latency than to the bandwith … would PCIe 5.0 change very much the ‘ecuation’?
Am I just being too much ‘creative’ or ‘ambitious’ ????
Thank you very much … and regards from Spain …
David,
Thanks for the analogies.
Although I have never heard of your 3-vehicle version before, I have heard people say that using Optane Memory over an NVMe interface is like trying to drive a Ferrari down a rutted dirt road – the road is what limits its speed.
I have to admit that I had never heard of a ‘Calash” before I read your comment. For other readers I will explain that it’s a kind of horse carriage.
Micron devised a very different analogy a few years ago. Do a search on this site for “Tomato”!
I’m not sure, but I think that Optane Memory dissipates less heat than DRAM, although write cycles do use a lot of energy. I suspect that the memory wouldn’t be the showstopper, but since Optane DIMMs are only supported by server processors today, then I would guess that the processor’s power and cooling needs might require a bigger battery than a person would want to carry around.
Fun ideas! Thanks for sharing them.
Jim
True … I said “Alder Lake Xeon ThinkPad” … it would probably had been more precise … ‘Sapphire Rapids’ Xeon ThinkPad … This IS a server processor, isn’t it????
I guess one problem could be the modules format themself … DIMM vs SO-DIMM … but I feel optimistic today 🙂 … so maybe third generation Optane modules could help with this … … or … maybe I could find a kind of ‘adaptor’ in an very far Oriental bazaar (yes … too much wishful thinking … and I am not that young … ) …
Asuming your words … I will reckon PCIe 5.0 is more or less fucked up due to the NVMe interface …
Another issue would probably be on the SW side … … letting me using a DIMM PMem module as a ‘drive’ … as my main drive … but you know, linux kernel people are pretty efficient …
Regards.
Ps.- Very interesting the Ryan Baxter’s tomato analogy … … you’ll understand I can’t wait to introduce Optane ‘modules’ within … or maybe Howard Marks already did it in the comments: “Storage class memory is like driving to the supermarket or walking to the corner store 20 min” … so with some luck I’ll be a ‘corner store’ man … 😉