Wednesday, July 17th 2019
Intel's CEO Blames 10 nm Delay on being "Too Aggressive"
During Fortune's Brainstorm Tech conference in Aspen, Colorado, Intel's CEO Bob Swan took stage and talked about the company, about where Intel is now and where they are headed in the future and how the company plans to evolve. Particular focus was put on how Intel became "data centric" from "PC centric," and the struggles it encountered.
However, when asked about the demise of Moore's Law, Swan detailed the aggressiveness that they approached the challenge with. Instead of the regular two times improvement in transistor density every two years, Swan said that Intel has always targeted better and greater densities so that it would stay the leader in the business.With 10 nm, Intel targets improved density by as much as 2.7x compared to the last generation of 14 nm transistors. He addressed the five year delay in delivering the 10 nm node being caused by "too aggressive innovation," adding that "... at a time it gets harder and harder, we set more aggressive goal..." and that's the main reason for the late delivery. Additionally he said that this time, Intel will stay at exactly 2x density improvements over two years with the company's 7 nm node, which is supposed to launch in two years and is already in development.
When talking about the future of Intel, Swan has noted that Intel's current market share is 30% of the "silicon market", saying that Intel is trying to diversify its current offerings from mainly CPUs and FPGAs to everything that requires big compute performance, in order to capture rest of the market. He noted that Artificial Intelligence is currently driving big demand for such performance, with autonomous vehicles expected to be a big source of revenue for Intel in the future. Through acquisitions like Mobileye, Intel plans to serve that market and increase the company's value.
You can listen to the talk here.
However, when asked about the demise of Moore's Law, Swan detailed the aggressiveness that they approached the challenge with. Instead of the regular two times improvement in transistor density every two years, Swan said that Intel has always targeted better and greater densities so that it would stay the leader in the business.With 10 nm, Intel targets improved density by as much as 2.7x compared to the last generation of 14 nm transistors. He addressed the five year delay in delivering the 10 nm node being caused by "too aggressive innovation," adding that "... at a time it gets harder and harder, we set more aggressive goal..." and that's the main reason for the late delivery. Additionally he said that this time, Intel will stay at exactly 2x density improvements over two years with the company's 7 nm node, which is supposed to launch in two years and is already in development.
When talking about the future of Intel, Swan has noted that Intel's current market share is 30% of the "silicon market", saying that Intel is trying to diversify its current offerings from mainly CPUs and FPGAs to everything that requires big compute performance, in order to capture rest of the market. He noted that Artificial Intelligence is currently driving big demand for such performance, with autonomous vehicles expected to be a big source of revenue for Intel in the future. Through acquisitions like Mobileye, Intel plans to serve that market and increase the company's value.
You can listen to the talk here.
111 Comments on Intel's CEO Blames 10 nm Delay on being "Too Aggressive"
And let's not forget Intel pushed forward the SSD market as well with their X25-M. I know, different division, but just saying, they weren't just sitting still like some like to think.
And let's not forget, Ryzen only matches Intel's current IPC. If Intel didn't mess up with Ice Lake, Ryzen would have been playing catch up from the very beginning.
It was still shot down in court, so I'm not defending the practice. I'm just saying, it wasn't something as obvious as a pay off.
The more important issue, which most seem to miss, is Intel's lack of a proper backup plan. Intel have put good development effort into Sunny Cove, which features their largest improvement in single thread performance since Sandy Bridge, but a lot of that effort is "wasted" when they don't have a suitable node to produce it on. If they had only done a small effort to prepare the architecture for 14nm as well, they would have been in a much better position right now. Luckily though, it's not like their development efforts have slowed down significantly, and Sapphire Rapids scheduled for 2021 should offer additional single thread improvements over Ice Lake/Sunny Cove, which seems to get a rather short "lifespan" in the market. So we should have some exciting years ahead of us with several improvements from both Intel and AMD, after several years of boring refreshes. Don't forget that it was Intel's repeated "screwups" on 10nm which allowed AMD to catch up. Ice Lake/Sunny Cove was supposed to launch in 2017.
Also, for the record, 6-core Coffee Lake was know before the launch of Zen. Mainstream 6-cores for the initial Skylake(2015) was not realistic due to the state of the 14nm at that time, most people have forgotten that it was very troublesome for some time.
That is no "choice", it's like being mugged at gun-point on the street is a "choice"...
But anyway, I'm glad those days are over.
Finally there is a choice.
Phew !
Wikichip cooberates this, for the engineer inclined. i7-990x Nehalem/Westmere says hello.
IPC man... it can't be identical chips with it changed. if there weren't innovating the cores, there'd be more of them, rest assured.
Haswell and Skylake have been major architectural changes to facilitate massive improvements for AVX, specialized acceleration and to some extent multi-core scaling, but unfortunately the IPC gains have been fairly small.
Ice Lake(Sunny Cove), which is two years overdue, seems to rectify some of this, by improving significantly integer multiplication/division, cache and load/store operations, all of which supposedly offers ~18% IPC gain, while only offering marginal improvements in SIMD over Skylake-X.
how come 10nm compared to 14nm will have 2.7x density improvement but 7nm will have 2x density improvement?
and anything else that he said was completely lies and there is no evidence of it in real world.
Either Mr. swan is a moron or he think that we are bunch of morons!
Assuming things are similar enough for Intel/TSMC 14nm and 10nm/7nm, Zen+ > Zen2 went from 4.8B transistors to 3.9 + 2.09 = 5.99B transistors. Intel does not disclose the transistor count on current CPUs but 8700K is estimated to have a little above 3B transistors and 9900K a little less than 4B which puts Intel's 14nm density roughly on par with TSMC's (and/or architecture density on par with AMD).
- Zen+ is 4.8B in 213mm^2 die (22Mtr/mm^2)
- Zen2 is 3.9B in 74mm^2 CCD (53Mtr/mm^2) and 2.09B in 124mm^2 IOD (17Mtr/mm^2)
There seems to be more and more evidence that IO does not scale well and relatively less dense IO die seems to support that notion. When it comes to cores themselves, 7nm is more than twice as dense as 14nm - 2.4x, give or take. Zen2 has about 40% more transistors in cores compared to Zen+ (and this might be a conservative estimate given changes with IO)
Sunny Cove improvements over Skylake pretty much mirror the changes AMD did with Zen2, increase in transistors needed to implement it should be roughly similar as well. Now, while manufacturing dies does not seem to be a big problem (in terms of yields) heat at 14nm is definitely a problem. 9900K is evidence enough. Now, imagine a 40% larger die with similar power density... What are you talking about?
Lets leave Intel aside for a moment. TSMC does not say direct comparison between 16nm and 7nm on their web page because they have a 10nm process between these. However, TSMC does state 16 > 10 was 2x density (15% faster, 35% less power) and 10 > 7 was 1.6x density (20% faster and 40% less power). This makes out 16 > 7 having 3.2x density improvement. Granted, TSMC's 16nm is a little less dense than Intel's 14nm and Intel's 10nm was initially (that 2.7x claim) intended to be denser than the current version but the scale is correct enough.
The stated density improvements are naturally somewhat real-life but best case scenarios. High performance versions of manufacturing processes that are used for CPUs or GPUs are less dense and less power efficient than mobile-oriented process versions.
Me, I'm not held back by my quad core (didn't even bother with HT), but at the new prices, I'm tempted to upgrade just because.
It's painfully obvious they had the capability to deliver more cores to the mainstream market but chose not to. To me as a consumer, the fact that Intel is better at extracting money means jack shit, all I see is a company that has been sandbagging for almost a decade.