What SemiAccurate has learned about Intel’s upcoming 10nm ‘release’ paints a contradictory picture for the company. It is the polar opposite of a real launch of a manufacturable product, a PR stunt to keep the stock price from crashing.
Authors Note: This article and analysis would normally be for subscribers only, however we feel a duty to inform the public of the facts in this instance.
In a really nice find, ComputerBase found a Lenovo Ideapad 330 with a 10nm Cannon Lake CPU aboard. This means Intel’s 10nm process is all on track and everything is all right, right? That is the intended message but it both contradicts what SemiAccurate moles have been saying for years now and Intel’s CEO have been saying for weeks, that the 10nm process doesn’t work. But it is coming out, right? Right. So what is actually going on?
When we first heard about this ‘release’ we were immediately skeptical because our sources have not said anything about a breakthrough in the process but they are consistently clear about the intractable problems. So how is Intel doing it? We started digging and found the answers.
The chip in the Ideapad 330 is a Cannon Lake based 10nm CPU from Intel that has been blessed with the marketing name Core i3-8121U. If you look at the main Ark page this part isn’t currently listed but if you search for it specifically you end up on a spec page. There are a lot of interesting bits to be gleaned from this page starting with the name. 8121U connotes a low end i3, the lowest end of the non-crippled all that much range. Think products that go into bottom-feeder special laptops, then step up half of a notch. Volume, not quality.
The specs show it runs at 2.20GHz base with a 3.2GHz turbo frequency all for a TDP of 15W. That 15W is of course stated in the name, the -U suffix is for that TDP, -Y for ~9W parts, and others for higher wattages and/or feature sets. In short while this dual core CPU doesn’t clock that high, it also doesn’t pull much power, about what you would expect. That isn’t bad until you notice a few other bits, first off that the i3-8121U doesn’t have a GPU, it is sold as a 2+0 in the parlance (CPU cores + GPU clusters), something Intel has never done before.
If you look at the two 8th gen parts that bracket the 8121U, you will see that the i3-8109U and i3-8130U are both 14nm chips from the Coffee and Kaby Lake, respectively, families. Both are 2C/4T parts with a GPU although Intel won’t give out any details about what the GPUs are any more. In any case they are at least a 2+1 configuration. What is interesting is that the 8130U runs at 2.2/3.4GHz (Base clock/Turbo clock) but the 8109U is at 3.0/3.6GHz because of a nominal 28W TDP. This should not be a -U CPU but who am I to question Intel’s sane, logical, and not at all random naming schemes?
Can You See The Graphics?:
One thing that is widely known about modern CPUs is that the GPU, where present, can take up a large portion of the TDP. While we won’t get into the nuances of power distribution, cooling, and TDP, lets just assume that half the TDP is for the CPU, the other half for the GPU under normal operation. That means the 10nm 8121U at 2.2/3.2GHz is slower at peak turbo than the 8130U at 2.2/3.4Ghz on an older 14nm process, at the same TDP. Worse yet for the 8121U, the 8130 _HAS_ a GPU. Let us recap, a 14nm CPU with a GPU is faster within a 15W cap than a 10nm CPU without a GPU running at the same 15W.
To make matters more painful, SemiAccurate’s sources say that the 10nm 2+1 die that the 8121U is based on was meant to be a 9W -Y part. Even if this isn’t true, the 10nm 8121U woefully worse than the 14nm parts, somewhere around half the efficiency of the 14nm competition. Chips with two cores at lower speeds and twice the power but without graphics aren’t likely to sell on merit in the open market.
Hang On A Sec:
By now the eagle eyed among you will have noticed we said that the 8121U is a 2+1 die but we are harping on the fact that is doesn’t have a GPU. What’s going on? To be blunt the 8121U does have a GPU on the die, it’s just that Intel can’t actually get it to work. This is entirely consistent with what SemiAccurate has been telling you for years now about the problems with the 10nm process, it was broken then, is broken now, and will be broken for far longer than Intel is admitting to. And they know it. (Note: Go back and look at their promises about 10nm during each of the quarterly analyst calls for the past 3 years, you will see a trend.)
If you think twice the power for all the Wattage is a bad marketing message, think about costs. Crushingly bad yields aside, the 10nm process is vastly more expensive than the 14nm one that the faster competition is built on. Sure you get a smaller die but the 8121U has likely half or more of that die turned off in the form of the GPU. The parts that Intel can effectively sell are about the same area but on a more expensive process than a hypothetical 14nm 2+0 die.
Any guesses what that does to Intel’s ASPs? Add in yield issues and that cost goes up by more than 3x. (Note 2: Yes I know what this intones and am being purposefully optimistic.) And just for giggles, add in the cost of an external GPU, all to go up against a faster, more efficient 10nm part. If you haven’t guessed by now, OEMs aren’t exactly relishing the thought of putting this CPU on the market. It won’t sell, it is inferior to the 14nm parts in literally every way, and costs a large multiple of what they do. There is no win for Intel on 10nm Cannon Lake parts other than as a PR stunt.
Why Now, Why This?:
So why is Lenovo putting this turkey out? Do they have a warehouse full of them that someone else needs the space for? Do they see an up side that isn’t portrayed by the specs, tech, manufacturability, or anything else? Is the device actually real or is it just an error in a database scheduled quarters ago that someone forgot to delete? SemiAccurate once again dug in and found out all the details.
The idea is pretty simple, Intel needs a win to counteract the well deserved pain it is getting from the 10nm meltdown. Since their PR strategy has made them universally hated among the press, there are few sympathetic ears out there other than paid shills. Even the most ardent of sycophants are calling Intel on their spurious claims so for the company, it is put up or shut up time. So Intel is going to make it look like they are putting up while not actually doing so because they can’t. The 10nm process not working is the spanner in the works in case you didn’t get it.
Twist Arm Backwards:
Intel can’t make 10nm parts at economically viable yields. Intel can’t make 10nm parts that have a salable feature set. Intel can’t make 10nm parts that beat their 14nm predecessors. But they can make small numbers of 10nm parts that, when you fuse half the die off, kinda sorta work at twice the power levels of the 14nm parts, just slower. If you take a fraction of the top bin of these parts, you get the killer device known as i3-8121U that literally none of the OEMs want to touch with a 10 foot pole because it will be death on the shelves.
Even if Intel subsidizes these parts to zero or below, the chips wouldn’t sell other than to geeks and reviewers. Why? Because battery life will be abysmal. Even if in real world use the CPU TDP isn’t 2x that of the 14nm parts, it is significantly more, and the external GPU that can never be turned off will eat up a chunk more too. This isn’t going to be a laptop that wins awards and everyone in the supply chain and OEM community knows it. To sell them, Intel needs to twist arms. And that is exactly what SemiAccurate’s sources tell us they are doing.
We are told that this PR stunt is going to be quite bounded for several reasons. First is the cost of making these CPUs, a large multiple of the cost of the 14nm parts. Second is supply, Intel is taking the top bin of the 10nm production lot, screening those, and ending up with the 8121U, two cores and no working GPU. Think about the piles of very expensive sand chunks that didn’t make the cut, a fraction of the top bin is not a huge percentage. Third they won’t sell on merit either to the OEMs or the public so there has to be a lot of subsidy dollars involved, directly or indirectly. SemiAccurate is told that still isn’t enough so Intel is politely applying pressure to grease the OEM wheels.
At this point OEMs are smiling and nodding because they have to. Intel has scraped up about 100K chips that meet the cut to distribute among OEMs. Each OEM has been asked to make one model featuring a 10nm part and to, “Make it look real“. Depending on the number of OEMs that get blessed with these parts, each one should receive between 5-20K chips to sell to the public, then job done. (Note 3: We are told this 100K is a one time deal and will not be followed up by more i3-8121Us or any other 10nm parts until volume production ramps in well over a year)
Officially Intel now has a triumphant launch of 10nm parts across a dozen or so OEMs which has to be real, right? The 10nm parts work, obviously have been in production since late 2017 as Intel said, and the crushing 10nm problems are anything but. Could a dozen OEMs make a dozen laptops if there were really crippling production problems? Intel is going to try and spin the 10nm meltdown as a choice aided by this data point.
A Little Math:
Think back to the past 20 or so chip launches that were actually real, each was preceded by a claim of dozens of OEMs and a wall of laptops sporting the devices. Any guesses what we will see at Computex this year? This whole 10nm ‘launch’ is designed to look real by being designed to look like the past launches even if there is no way it could be. If you look at the numbers, Intel sells about 250M chips a year now, give or take a few tens of millions. Lets call it 667K a day or so, weekends and holidays included.
That means that the 100K 10nm CPUs Intel is forcing OEMs to take account for ~15% of ONE DAY’S production at Intel or 0.0004% of their yearly output. Now think back to Intel’s statement that production has been going on since late 2017 and everything is fine. It took the company six months to make 15% of a single day’s volume with their entire 10nm output. And half of that chip flat out doesn’t work. Still think nothing is wrong with 10nm? Still think it is ‘planned’? Still think they know what the problem is? Still think they have a fix? Still think that production will ramp in 2019 as promised?
Not The End Of This Story:
In the end we have a chip being built on a troubled 10nm process. In six months Intel can make 15% of a day’s production on those production lines. The resultant chips are abjectly broken, they are 2+1 but the +1 doesn’t work which means they are selling a CPU with half the die turned off, an expensive proposition given the cost of the process and the shatteringly low yields. Even with the GPU turned off, the resultant CPU uses twice the power of the 14nm devices to run slower than those with a GPU.
OEMs won’t touch these 10nm parts either voluntarily or with ‘not bribes’ so Intel has to twist arms and force them to make laptops and “Make it ‘look real“. Why? To put out a data point that they can build ‘truth’ and ‘alternative facts’ around when it comes time to talk to analysts. The 10nm Cannon Lake parts aren’t real and never will be viable, financially or technically speaking. Feel free to believe the PR messaging but you can’t say you don’t know what is really going on now.S|A
Latest posts by Charlie Demerjian (see all)
- Another Epyc Rome TCO data point comes out - Nov 12, 2019
- What is the name of Intel’s Cascade Lake +5 server? - Nov 9, 2019
- AMD launches Threadripper 3, 3950X, and Athlon 3000G - Nov 7, 2019
- Intel messaging hits a new low - Nov 5, 2019
- Intel releases the world’s largest FPGA - Nov 5, 2019