Upcoming Intel discussion.

Wowsers! Two big pieces of news;

https://www.anandtech.com/show/1657...-on-intel-ip-blocks-for-foundry-cores-on-tsmc

Intel will also open up its foundries to many other customers;

https://www.anandtech.com/show/1657...ndry-services-ibm-collaboration-return-of-idf

 

https://semiwiki.com/semiconductor-manufacturers/297534-insiders-view-of-intel-foundry-services/

Pretty good summary of the challenges facing Intel with this foundry play. It's gonna be a tough ask but they've done some pretty amazing stuff in the past so who knows?

 

Interesting. Wouldn't licencing x86 custom designs have implications for AMD?
ie, if try to you licence x64, wouldn't AMD be able to say "yeah, nah" ?

I gotta wonder if Via might mount an interesting challenge. That would be amusing. VIA x86 on an intel process. By all rights, they can do it.

 

would be the same as AMD currently licencing designs via their Semi custom biz

AMD had issues just manufacturing x86 on global foundries initially, but that's all sorted now, and I believe it goes both ways

 

Intel just rebranded all it's latest and all future nodes. Essentially they are going with new names to be more similar to tsmc/samsung densities. So previously intel 10nm was actuallly more dense than tmsc 7nm (~101m transistors / mm2 vs ~95m for tsmc 7nm). Now it's called Intel 7. And Intel 7nm becomes Intel 4 etc. They also picked up Qualcomm as a future customer, on intel 20A (was 5nm). Any bets on if they actually hit Intel 20A in 2024?

https://www.anandtech.com/show/1682...nm-3nm-20a-18a-packaging-foundry-emib-foveros
https://www.anandtech.com/show/1684...tomer-qualcomm-jumps-on-board-for-20a-process

 

TSMC has committed to spending $100B in the next 3 years. Intel has committed $20B. Either there are a whole lotta gub'mint subsidies coming Intel's way or they are magicians, I can't see how they can catch up and surpass TSMC in 4 years given TSMC's relentless progress, and their massive customers. Oh and at the same time Intel needs to learn how to be a foundry and hope everyone forgets about LG and Ericsson. If they can pull off even one of those two massive tasks, they will make a shit ton of money and it will be well deserved.

 

Different production base and starting points.

https://www.tomshardware.com/news/ic-insights-top-15-makers-of-semiconductors

Intel already produces nearly twice the amount of IC compared to TSMC. If you already have the production capabilities and capacity in place you naturally would spend less to ramp up. Intel just announced its already producing more 10nm wafers than 14nm. The EUV machines machines they already have in place are projected to last them into 20A. TSMC is also a pay to play that covers a wide range of services not just the headline grabbing cutting edge stuff, conversion of their older process foundries to newer processes costs more.

The custom foundry service is a completely different ball game. The biggest problem with the service was it didn't use industry standard EDA's but rather Intel custom stuff, this created all sorts of problems for porting and creating designs. The shock with the announcement on this expansion (more like re-launch) is the fact that Intel started using design tools that everyone is already using.

 

I think you were looking at the 2020 numbers, the gap is below 50% in 1Q21 ($18.6B vs $12.9B). That also doesn't account for Intel's memory business, which they've sold, and is about $1B/quarter. I tend to agree with the sentiment but the reality of their last 5 years of denial about the issues with 10nm doesn't paint a pretty picture. Most analysts agree Gelsinger is a good choice so maybe he'll fix up the direction.

I noted that line about 10nm volume from Intel, but curiously they left out any mention of yield TSMC has over half of the installed EUV base globally and claim 65% of EUV production volume. They are ahead of everyone in EUV, unquestionably.

Yeah their words say they've learned their lesson, but they need to back it up with actions. It's not just the technical stuff like EDA tools either, TSMC is absolutely dedicated to the foundry customer, they don't exist without it.

I think at some point Intel will have to spin off the foundry service, somehow I don't think they would allow anyone to actually compete with Intel products on the same process. The Qualcomm announcement is interesting, but it's so far off that there's plenty of time for either side to bail out. A cynical person might say that Intel gets a headline and Qualcomm gets to scare their existing suppliers.

edit: Qualcomm have clarified:

 

Summary article from Ars which neatly encapsulates the node naming convention change and the pivot to customer-facing-foundry business model.
Includes a gallery of shiny graphics which were no doubt also shown in the articles linked above, and some commentary re. Qualcomm's interest.

https://arstechnica.com/gadgets/202...map-lays-out-the-post-nanometer-angstrom-era/

 

Between AMD and Apple M1, looks intel is shook. Will be interesting!

 

To be doing anything else other than looking to widen their market share across every pie they have a finger in would make their management completely incompetent.
So expanding the foundry business to external customers isn't exactly a kinghit out of nowhere (he says, with 20/20 hindsight).

Value judgements about whether their management is, in fact, incompetent and biting off way more than they can chew in competition with TSMC/Samsung/GF will be an exercise left to the reader.

 

https://semiaccurate.com/2022/05/16/intel-shows-off-advanced-silicon-at-vision/

Pics at link, pretty crazy

 

I count 52. Though I assume some of them are just spacers.

Pretty nuts.

5
6
10
10
10
6
5

 

Intel 13th Gen Core ‘Raptor Lake’ and next-gen HEDT ‘Sapphire Rapids’ are both rumored to launch October

https://videocardz.com/newz/intel-1...ire-rapids-are-both-rumored-to-launch-october

It is said that 13th Gen Core Raptor Lake would appear in October with support for existing motherboards but also for the new high-end Z790 platform. The rumor alleges that there will be new 700 motherboard series including H770 and B760 series launching next year. A very similar launch schedule was adopted for 12th Gen Core series, therefore, one could guess that K-series CPUs will be among the first to launch.

14th Gen Core Meteor Lake-S in 2024 to be 7nm and DDR5 only.

Allegedly.

 

RL having support for both DDR4 & DDR5?? a repeat of the skylake/kabylake memory transition years..

 

Then that is the one to buy. All before suck as

 

i hope they ditch the p and e cores stuff and go back to real cores or some madmen find a way to oc those e cores into p core
https://www.tomshardware.com/how-to/intel-alder-lake-cpu-overclock-guide
so you can only is e cores in groups of 4 why?

 

Because they share the same timing circuit - and cache, and whatever else.

I hope they make more like it. They won't do any harm once the scheduler's ironed out - and that's software.

 

I for one am a big fan of P & E cores. Whilst I generally avoid background apps like the plague, with the way hardware manufacturers and software devs/OS's going these days requiring background services, it makes sense to allocate lower powered cores/threads specifically to said tasks. Intel taking a leaf from the ARM playbook.

 

Um... what? Haha. You can OC the E cores too. You can't turn them into P cores though. It's like you can't turn an orange into an apple.