With Intel’s Cascade Lake rollout last month came with a co-introduction of 3D XPoint Memory in a DIMM form factor, the Optane DIMM that had been promised since the first introduction of 3D XPoint Memory in mid-2015. A lot of benchmarks were provided to make the case for using Optane DIMMs (formally known as the Intel Optane DC Persistent Memory), but not much was said about the pricing, except for assertions that significant savings were possible when Optane was used to replace some of the DRAM in a large computing system.
So… How much does it cost? Well certain technical reports in resources like Anandtech probed sales channels to see what they could find, but The Memory Guy learned that the presentations Intel made to the press in advance of the Cascade Lake rollout contained not only prices for the three Optane DIMM densities (128, 256, & 512GB), but also provided the prices of the DRAM DIMMs that they were being compared against. I’ll get to that in a moment, but first let’s wade through the fundamentals of Intel’s Optane pricing strategy to understand why Intel has needs to price it the way that it has.
In Objective Analysis’ report on 3D XPoint Memory, and in several presentations I have made over the past four years, I have explained how Optane prices must follow the general trend of the chart below.
I won’t go into a long explanation but will briefly say that the different levels of the memory/storage hierarchy in this chart only make sense if they fit into their natural places on the line of orbs. A new storage level must be cheaper than the next-faster technology, and faster than the next-cheaper technology. The speed of any of these technologies is determined by physics. Pricing is determined by a number of factors, and all come down to a business decision.
In order to fit into its natural place in this chart 3D XPoint must be less expensive than DRAM. If it were not, Optane’s target customers would simply purchase more DRAM. Intel understands this point, and has explained over the past few years that 3D XPoint Memory would sell for about half the price of DRAM. The company’s Optane SSD pricing follows that approach.
Let me take a small diversionary path here to explain to the more technical readers something that they may not have encountered before. There are two basic ways to price a product: Price to Cost, and Price to Market. Typically both of these are profitable. For example, a no-name-brand shirt at a discount store is likely to be priced a little higher than its manufacturing cost, since consumers will be looking for the best deal. This shirt is priced to cost. A Gucci shirt bought at a distinguished haberdasher is likely to be priced several times as much as the no-name shirt even though it didn’t cost several times as much to produce. This shirt is priced to market — certain consumers will spend a lot more for a big-name brand. That’s nice if you can do it.
Intel’s Optane has the opposite problem. Optane’s price, half that of DRAM, means that it is priced to market — that’s the price that it can sell for. Since it currently costs more than DRAM to produce Intel has to sell it at a loss. That loss was spelled out in an earlier post.
With Intel’s Cascade Lake rollout last month pricing for all densities of Optane DIMM module were revealed, but not many people noticed since it was in very small print on a backup slide. These prices were used to justify certain statements about cost savings on graphs in the body of the presentation.
I had a look at these prices to see if I could understand how Intel sets the price of the Optane DIMM and found some interesting consistencies.
The table below shows the DIMM prices I found in the backup slide, and their corresponding price per gigabyte:
|
Type |
Density |
Price |
$/GB |
|
DRAM |
32GB |
$374.71 |
$11.71 |
|
DRAM |
64GB |
$708.25 |
$11.07 |
|
DRAM |
128GB |
$1,913.21 |
$14.95 |
|
DRAM |
256GB |
$5,952.00 |
$23.25 |
|
Optane |
128GB |
$577.00 |
$4.51 |
|
Optane |
256GB |
$2,125.00 |
$8.30 |
|
Optane |
512GB |
$6,751.00 |
$13.19 |
One of the first things to notice is that both products’ price per gigabyte increases with increasing density. In DRAM’s case this is a necessity since it’s very difficult to pack a lot of DRAM onto a single module and, since the market is limited (probably due to the high price) certain customers are willing to pay a premium for a very high density DIMM. The first argument implies that this is price-to-cost and the second implies price-to-market. It’s hard to tell.
While I haven’t tried to understand how the DRAM DIMM prices have been decided, I can draw a link between them and the Optane DIMM prices.
First, let’s look at the price per gigabyte. Those numbers from the table have been charted in the graph below:
While the relationship between the two 256GB or the two 128GB modules seems confusing, there’s a consistent relationship between next-lower densities. That is, the 512GB Optane DIMM is priced 43% cheaper than the 256GB DRAM DIMM, and the 256GB Optane DIMM is priced 44% lower than the 128GB DRAM DIMM. So even though the price per GB for the DRAM DIMM rises significantly for the 128GB and 256GB densities, Optane is priced to keep pace at about half the price of a DRAM module with half the capacity. Intel is pricing to value.
Here’s another chart that plots the prior column’s data in the table. It compares the module prices by density.
Here we can see that an Optane DIMM is priced slightly higher than the next-lower-density DRAM DIMM, or at roughly half the price of DRAM.
While I can’t tell you whether Intel plans to continue to price its Optane DIMMs this way, it’s pretty clear that the company is taking advantage of the difficulty DRAM makers have building 128GB and 256GB DIMMs to keep its Optane DIMM prices as high as possible. This move will help offset some of the company’s losses on the product.
This is just one of the many ways that Objective Analysis looks at market data to provide our clients with insight on their competitors’ strategies. Companies that would like to benefit from such insight and to succeed while others struggle are welcome to become clients by contacting us to discuss ways we can help you to achieve your goals.
It’ll be very interesting to see how Intel will cope with the DRAM price erosion of roughly 10% per quarter right now.
Michael,
Good to see your name!
Intel is committed to make Optane work. They are losing money on it today, and if DRAM prices continue their slide (which we predict will continue) then they will lose even more money.
As long as they offset that loss with increased processor sales then everything is good! It will probably work!
Jim
So where or what is the yield issue for the 3DxP?
Julian,
Don’t be too hasty to put this all onto a yield issue. Any new memory technology is faced by a world of issues, as we saw with the 3-year delay in bringing 3D NAND into price competition with its 15nm planar counterpart.
There’ a blog post I have been meaning to write for a long time likening 3D NAND’s challenges to a more mundane example of driving a rented car. Watch for it!
Intel and Micron are ramping processes that have never previously been mass-produced. It’s a herculean effort!
Jim
The DRAM price points in the above comparison are not accurate in today’s market. They may be what were presented in the file you are referring to, but it would be best to verify the DRAM price points for a proper comparison. For example, the 32GB at $374 is incorrect. For a May end time frame, the pricing was more in the $130 ~ $140 range or lower.
Rick,
You’re right to point out that DRAM prices have dropped significantly since late last year. The Intel presentation’s numbers show us Intel’s strategy, though, rather than current prices.
Intel may have done this price/performance analysis 6 months ago. It really doesn’t matter. What matters is that Intel revealed how they intend to price Optane modules.
The company has always openly stated that Optane would be priced in proportion to DRAM. If DRAM prices fall then Optane prices can be expected to fall by the same percentage.
Jim
Your analysis helped me understand why Intel hasn’t enabled Optane support for non-Intel systems (i.e., Epyc and Power). I had assumed they didn’t want those alternatives to benefit from the performance advantages of 3DXP. Now I see that they need the margin from Xeon CPU sales to offset the losses on the Optane DIMMs. Eye opening. Thanks, Jim
Nathan, Always a pleasure to hear from you!
Something that you will appreciate is that Intel needed to put a lot of “Secret Sauce” into the processor’s MMU to make it work with Optane – even the DDR4 bus had to be modified!
If Intel wanted for competing processors to support Optane they would need to provide this IP to their competitors. If they did that then they would lose a competitive edge, since Optane provides a price/performance advantage to the system as a whole.
It will be interesting to watch all of this play out over time!
Jim
I used to think the competitive edge was the motivation, but your economic analysis convinced me otherwise.
OTOH, since AMD is challenging Intel on CPU performance, anything Intel can do to offset the CPU performance issue at the system level probably helps Intel’s sales.
OTOH, chip manufacturing costs always benefit from increased volume shipments, so opening up Optane to third parties might make it more cost effective, which would also help Intel’s margins.
OTOH, since Intel has such a dominant share of the server market, the incremental sales generated by supporting non-Intel systems wouldn’t have much impact on costs.
OTOH, …
Nathan, I have lost track of how many hands you have!
If this technology behaves anything like NAND flash then it won’t meet its production cost targets until its wafer starts come within about an order of magnitude of DRAM’s. That won’t happen very soon if Optane only sells into the server market. It’s a concern!
PC purchasers haven’t warmed up to the idea of paying a lot more a faster Optane NVMe SSD. Perhaps Intel will find a compelling reason for PC users to adopt Optane DIMMs. Until then the company will have to be satisfied with losing money on its Optane business.
Jim
Optane also has a “who goes first” marketing problem. No one wants (I wouldn’t) to put an unproven Si part into a critical system. Plus, you get to pay extra for the privilege. Even if the MU/Intel people have done the work to prove the reliability, it is a hard sell. A hard sell for two companies not good at selling.
The new hope is NV-DIMM-P, just emerging from JEDEC. https://www.jedec.org/nvdimm-p
Doug, Thanks for the comment.
I don’t think it’s as hard of a sell as you do.
OEMs have had enough time to get a pretty solid understanding of 3D XPoint reliability in the two years since Optane SSDs became available.
But that doesn’t necessarily tell them how reliable the Optane DIMM is. Google bought Intel’s first-sold Optane DIMM at the very end of last year and presented it, broken, to Intel on a plaque at the Cascade Lake formal introduction in April. I have no doubts that it broke because Google put it though rigorous stress tests. Other hyperscalers most likely have done the same. This has doubtlessly given them an understanding of the product’s limitations. They still seem excited to start using it in production systems.
The NVDIMM-P is a good alternative, but it also has been designed to use new memory technologies, so it will face the same reliability concerns as Optane.
Over the long run I would expect for both to see widespread use.
Best,
Jim