Report: Optane DIMMs Provide Only Modest Performance Improvements
When Intel unveiled Optane in 2015, it declared it had created a fundamentally new type of memory that could revolutionize the industry. Four years later, signs of that revolution have been hard to come by. Optane DC Persistent Memory — Optane DIMMs loaded into memory slots with higher capacities but less speed — are supposed to change this. New reports, however, suggest Optane’s performance benefits are modest and the difficulty of improving that situation could be considerable.
Inventing a new type of RAM and building an entire ecosystem around it is a heavy lift. Intel has acknowledged that encouraging companies to adapt software for its Optane DC PMs will take time. But the willingness of companies to make that investment will depend on the performance uplift they can expect from doing so. There are several headwinds facing Intel and Optane at the moment.
First, Optane performance when used as DRAM is intrinsically variable depending on the size of the data being read. Idle read latency ranges from 180 – 340 ns, compared with 70ns for DRAM. Latency also can depend on power consumption (DIMM power consumption ranges from 12-18W).
Second, while Optane PC DMs were originally intended for commercialization by 2017, they’ve slipped into early 2019, allowing DRAM densities to increase and prices to drop. That’s particularly important at the moment because, as we’ve covered, DRAM prices are in freefall and expected to continue declining through the first half of the year.
The macroeconomics of the DRAM market aren’t Intel’s fault and have nothing to do with Optane’s success as a technical standard, but they still shape the market Optane has to compete in. Steadily improving capacities shrink the capacity advantage Optane can offer, while cratering prices makes it harder for Intel to price competitively. RAM prices will eventually rise, however — the boom-and-bust cycle of the DRAM market has been a regular facet of life for decades. If Intel can keep bringing Optane prices down and capacities up it should be able to overcome this issue.
The third problem, and possibly the biggest, is that Optane’s performance advantages appear to be fairly modest and potentially narrowly-tailored. Customers can run up to 30 percent more VM’s per server when using Microsoft HyperV. SAP Hana users can either restart a database 12.5x faster or save 39 percent on the cost of a large, in-memory database. Intel has said it’s shooting for a 1.2x improvement on performance-per-dollar over DRAM on target workloads.
“The performance doesn’t seem very compelling … and more performance with more memory is not a fair comparison,” Linley Gwennap of the Linley Group told EETimes. “There are some apps that benefit … but not at the 2× to 4× gains you typically expect from a new technology.”
Intel hasn’t commented on its Optane roadmap or the likelihood of further near-term performance improvements. Chips are expected to last for five years of read/write cycles when deployed in servers and retain data for three months when powered off. This is likely a non-issue in the server space, where systems are highly unlikely to be powered down for any length of time. Support for the persistent memory features Intel wants to deploy will take time to build into applications, but if the performance advantages aren’t large enough, manufacturers won’t invest in the first place.
Four years after its unveiling, Optane seems stuck in a place between hype-cycle and proven competitor. The chips work. The storage performance is good. There’s a genuine niche and Optane offers capabilities that NAND, even SLC NAND, can’t match. Its reliability is higher than NAND, it performs better at low queue depths, and it’s at least closer to the NVRAM holy grail that everyone in computing has wanted for decades. But it’s still unclear whether Intel has carved out a small niche for itself in the larger memory market or invented a technology that could one day supplant either NAND, DRAM, or both. How rapidly Optane DC PM is adopted for servers and databases could tell us a great deal about the technology’s long-term chance of breaking out of niche status.
- Intel Launches New Xeons With Up to 56 Cores, 400W TDP
- Intel Adds Optane Support to Pentium, Celeron Processors
- Micron Will Buy Out Intel’s Share of Joint Optane Fab