Crisis? What crisis? —

Intel: Meltdown, Spectre silicon fixes coming 2018; 3D XPoint RAM, not so much

And the roadmap for the transition to 10nm remains as murky as ever.

Intel: Meltdown, Spectre silicon fixes coming 2018; 3D XPoint RAM, not so much
Intel

As part of Intel's fourth quarter financials release, CEO Brian Krzanich promised that chips shipping this year would include true hardware fixes for the Spectre and Meltdown attacks.

The promise to ship chips immune to the attack leaves many questions unanswered. It's not clear if the fixes will be revisions of current generation Kaby Lake, Coffee Lake, and Skylake parts, or if the modifications will be constrained to the Cannon Lake processors that are expected to ship this year. Nor is it clear what form the fix will take: better, higher-performance versions of the microcode and workarounds already being rolled out, or deeper modifications to the processor's speculative execution and branch prediction behavior.

The company's delayed transition to a 10nm manufacturing process also remains murky. At CES, Intel claimed that it had shipped some unspecified chips built on 10nm last year. The first half of this year will see low volume production, ramping to high volume in the second half. But exactly what processors—in what configurations, when, and in what volumes—remains unknown. Both Cannon Lake, built on the 10nm process, and Ice Lake, built on the refined "10nm+" process, are planned, but the company has said little concrete about exact timelines.

Krzanich also said that DIMMs using 3D XPoint memory won't have any revenue impact this year. 3D XPoint is a kind of solid state storage that in principle blends the characteristics of RAM and flash memory. Like RAM, it's addressable and writeable on a byte-at-a-time basis (in contrast to flash, which can only be written and erased in large blocks), but like flash, it's persistent, requiring no power to retain its value. Intel has a few 3D XPoint products on the market already, configured as either PCIe or M.2 storage, but the company has been promising to sell non-volatile DIMMs based on the same technology. Processors could use these as if they were RAM—just RAM that holds its value when the power is off. This is particularly appealing for applications such as databases. It would allow, for example, high-performance in-memory databases that are persistent across reboots without ever having to be written out to disk.

Existing 3D XPoint products have limited longevity—like flash, they degrade after each write cycle—and while it appears that Intel has addressed this to a satisfactory extent for storage, system memory is much more demanding. It's not clear that all the technical problems have been solved.

3D XPoint DIMMs were expected some time this year, but with the revenue guidance, it looks now like they won't arrive until 2019 at the earliest.

Overall, Intel had a strong quarter, with revenue up eight percent to $17.1 billion, and for the full year, up nine percent to $62.8 billion. At least for now, AMD's newly competitive products don't appear to have a great deal of impact; Intel's data center revenue was up 10 percent last quarter, with average selling prices rising eight percent. Desktop processors did decline—five percent volume, two percent selling price—but the company attributed this to the declining PC market in general rather than competition from AMD's Ryzen and Threadripper products or Intel's own shortages of Coffee Lake desktop chips. Going forward, the company says it isn't expecting any financial impact from Spectre or Meltdown, but the attacks do introduce new risks and so the outlook may change.

Channel Ars Technica