Short Supply of Intel’s Gemini Lake Confirmed by PC Maker
by Anton Shilov on November 21, 2018 11:35 AM EST- Posted in
- SoCs
- CPUs
- Intel
- Gemini Lake
- Hardkernel
- Odroid
Intel has publicly admitted that demand for its processors exceeds supply, which is why it is currently prioritizes shipments of its higher-end Core and Xeon products over cheaper chips. Various manufacturers of actual PCs are now confirming the backlog in orders, giving us additional details to the story. As it appears, shortages of Intel’s Gemini Lake SoCs for low-cost PCs are so severe that the earliest opportunity for some of the PC makers to get the chips will be in February or March.
Earlier this week a company named Hardkernel launched its first x86-based Odroid-H2 compact PC platform, a device akin to a Raspberry Pi. The 4.3 inch × 4.3 inch platform is based on Intel’s quad-core Celeron J4105 SoC and can be equipped with up to 32 GB of DDR4-2400 memory, an M.2 SSD, and has an essential set of wired I/O interfaces (DP 1.2, HDMI 2.0, two GbE ports, two USB 3.0, two USB 2.0, 3.5-mm audio jack, and an S/PDIF).
The launch was labeled a success and the first batch of 2,000 units was sold in 24 hours. As it appears, the company did not expect demand to be so high, so it did not prepare more units. Apparently, Hardkernel cannot order more Gemini Lake SoCs from Intel right now. The chip giant reportedly said that it would be able to deliver the chips in February or March, but could only provide a more exact delivery schedule in January.
“We have very small problem now,” the company said “We have sold out all our first production lot in 24 hours. We are so sorry for that. In fact, we conservatively prepared only 2,000pcs of H2 boards because we could not predict this situation. We are contacting Intel for the lead time of the Gemini Lake processor. Due to Intel's serious supplying issue these days, they can supply the Gemini Lake SoC in February or March 2019. […] They also mentioned that they can give us a firm delivery schedule in January.”
Considering the fact that Gemini Lake is an entry-level product, it is not surprising that it sits rather low in Intel’s list of priorities. In fact, a rumor has it that Intel is considering to outsource some of its low-end SoCs to contract makers of semiconductors, though considering all the complexities with re-spinning the chips for someone else’s process technology, this is something that will be a hard thing to do.
One thing that particularly attracts attention is the volume of SoCs that Hardkernel might require. The first batch was only 2,000, so it is unlikely that the second batch would hit 20,000. Meanwhile, Intel does not seem to have spare Gemini Lake SoCs at all, which once again emphasizes the scope of the company’s problems.
Related Reading
- ASUS Comments on Intel Shortages, U.S.-China Trade War
- Intel Investing $1B to Meet 14nm Demand: Prioritizing High-End Core and Xeon
- Intel Announces Q3 FY 2018 Earnings: Record Quarter
- Intel to Split Manufacturing Group into Three Segments
Source: Hardkernel (via FanlessTech)
Facebook
Twitter
RSS
26 Comments
View All Comments
abufrejoval - Wednesday, November 21, 2018 - link
market segmentation, they want to sell more expensive chips like the C3000 to those who want more RAM. Been using 16GB on J1900 and N3700 w/o problems for years. Perhaps Intel MBs actually check and disable >8GB in the BIOS, ASrock sure doesn't mind.yeeeeman - Wednesday, November 21, 2018 - link
Too bad AMD ditched the Temash APUs. They had good performance with low power consumption and with small upgrades and 14nm process they could have been a great alternative to Gemini Lake parts.jamesindevon - Wednesday, November 21, 2018 - link
If Intel was sufficiently desperate, the SoFIA Atom-on-TMSC products should be usable as-is with a custom motherboard for the very low-end.watersb - Wednesday, November 21, 2018 - link
I am finally seeing Pentium Silver N5000 laptops available for sale. Wonder if they will sell out at major US retailers this holiday weekend.Valantar - Wednesday, November 21, 2018 - link
2000 of a die this small - how many wafers would that be? One? Two? More?olafgarten - Wednesday, November 21, 2018 - link
According to my very rough calculation, a perfect yield would get roughly 3000 of these out of a single wafer.Valantar - Thursday, November 22, 2018 - link
So assuming 90% yield, which one would really hope they're at for a 4-year-old process with three major refinements under its belt, they'd be able to get more than enough for a second round of these out of a single wafer. I very much doubt many of these dice bin below the power/clock speed requirements of the J4105 - it's not even the highest bin of Gemini Lake. This speaks quite loudly to just how precarious this shortage is. Sure, Hardkernel is probably WAY down the list of who gets first dibs on wafers, but this is still very much not normal.The shortage seems to be hitting big-name brands hard too, even in the business/enterprise segments - I've been on a waiting list for a new Dell laptop at work for a couple of months now. They normally have 3-day (or less) delivery (I work at a university with something like 5000 staff, so we buy a lot of PCs), but the Latitude I'm waiting on (with an i7-8650u) keeps getting pushed back, only recently getting a confirmed delivery date at all.
IntelUser2000 - Wednesday, November 21, 2018 - link
Geminilake die is not that small.It's around the size of Intel's dual core dies.
voicequal - Thursday, November 22, 2018 - link
Not sure this delay can be pinned entirely on Intel. It sounds like this company grossly underestimated demand, so they've been caught flat-footed with no 2nd batch orders prepped with their suppliers.Santoval - Thursday, November 22, 2018 - link
I don't understand how they were able to pair the Celeron J4105 with 32 GB of DRAM when it supports (according to its ark.intel.com page) only 8 GB of DRAM. The M.2 SSD support is also weird. According to Intel this CPU has merely 6 PCIe lanes, but these are not even 3.0 lanes, they are PCIe version *2.0* lanes. Did they use that 960 EVO in the picture with x4 PCIe 2.0 lanes, thus halving its speed? Why on Earth would anyone do that?Still, this *could* be done. Going from 8 GB max to 32 GB of DRAM should be impossible though.