From 2H2021 TSMC will be producing Intel CPUs on 5nm (and beyond)

[Mihle]

18 minutes ago, ONOTech said:

Someone correct me if I'm wrong, but isn't AMD moving to Samsung for their GPUs and CPUs 2H 2021?

If that is true that is news for me.

[StDragon]

3 hours ago, RejZoR said:

Not really. You may look at it that you need more waffer surface area to achieve same results, but you need to throw away far less surface area of it in the end, resulting in better usability of the waffer surface area vs massive monolithic processors. So, when you draw a line, it's more efficient to churn out tons of tiny chiplets and "glue" them together than making way less massive chips, throw away 1/4 or 1/3 of them during the process.

Sure, chiplets are a work-around to a yield problem. But that doesn't diminish the fact more transistors added than lithography can keep up with. As nears as I can tell, monolithic dies are physically getting larger, but the wafers aren't. I mean, I could be wrong about that last part, but I don't think so.

[wanderingfool2]

8 hours ago, minibois said:

Huh, a lot of people sort of ridiculed that hedge fund's recommendations, but Intel actually did it (of course they had this in the works far before it was suggested to the last week).

 

Let's hope they will be able to scale with the demand of their CPU's and of course let's hope the product will perform well!

Think it's important to note though.  The hedge fund people were asking Intel to abandon their chip manufacturing, not just supplement it.  If the rumors that the most high end ones will still be with Intel, there might be a reason for that.

 

As tot he topic, it would be interesting to see whether there is any performance difference with this.

[CarlBar]

1 hour ago, StDragon said:

Sure, chiplets are a work-around to a yield problem. But that doesn't diminish the fact more transistors added than lithography can keep up with. As nears as I can tell, monolithic dies are physically getting larger, but the wafers aren't. I mean, I could be wrong about that last part, but I don't think so.

 

Chiplets shouldn't add much to it, and they actually save a lot. remember the bulk of the silicon volume is in the IO Die and thats on a different higher production volume process, (in part because this stuff doesn't shrink down or get more power efficient on smaller nodes). That plus fewer dies thrown away at the end due to fewer having a killer defect doubtless means more actual CPU's per wafer.

[Jet_ski]

14 hours ago, minibois said:

Huh, a lot of people sort of ridiculed that hedge fund's recommendations, but Intel actually did it (of course they had this in the works far before it was suggested to the last week).

 

Let's hope they will be able to scale with the demand of their CPU's and of course let's hope the product will perform well!

This is a typical thing hedge funds do: they find out about a credible rumor so they act on it and write a letter to generate a lot of publicity for their position. They kinda fabric credibility for themselves. But the fact is Intel was already planning on it and the hedge fund’s letter, I promise, when directly into the trash can.

[Kisai]

22 hours ago, minibois said:

Huh, a lot of people sort of ridiculed that hedge fund's recommendations, but Intel actually did it (of course they had this in the works far before it was suggested to the last week).

 

Let's hope they will be able to scale with the demand of their CPU's and of course let's hope the product will perform well!

The hedge funds recommendation was to divest (sell) their own fabs and chip-related parts of the business, which is basically suicide for Intel's tech.  As we've seen, every time, too much consolidation of tech in too few companies results in stagnation, and should something happen at TSMC or Samsung, or another pandemic, too much of the manufacturing logistics gets disrupted, and prices escalate.

 

These are not cookie factories. You can't go make your own computer chips like you can oreo's. Should TSMC have a bad quarter, or a factory is leveled in an earthquake, that supply is permanently disrupted. You can't just spin up another factory overnight. 

 

We've seen this before:

https://www.electropages.com/blog/2020/12/power-outage-micron-facility-could-cause-global-memory-shortage

 

2020 - Micron power cut, global memory shortage

 

https://www.tomshardware.com/news/storage-shortage-coronavirus-device-makers-asia-HDD-SSD

 

2020 - (Seagate) Manufacturing was stalled for three days due to covid, causing a shortage.

Which was like 2012's shortage caused by flooding.

 

Would the disruptions have been as bad had any of the manufacturing capacity have been in the US or Ireland? It likely would have resulted in less of a shortage and more of a logistics nightmare where instead of ordering things to produce a computer or a phone in china, a delay in shipping parts would result in a stalled manufacturing line.

[Brooksie359]

23 hours ago, TrigrH said:

Unlike basically everyone else, Intel is a direct competitor of TSMC. Its not exactly in TSMC's best interest to feed its competition unless its actually worthwhile. If I was "TSMC" I wouldn't sell fab space to intel unless the price was doubled. 

Intel isn't a direct competitor though. They only manufacture their own stuff so Intel doesn't compete with TSMC for customers as Intel is not in the business of manufacturing other peoples designs. Sure some products that use tsmc for a fab compete with Intel products but that is indirect competition. 

[soldier_ph]

TSMC to the rescue ! 

[Spicedaddy]

Not a bad idea to outsource some production to TSMC, but I hope Intel doesn't abandon fabrication entirely. It's basically 3 players (Intel, TSMC and Samsung), it wouldn't be good for consumers if one goes away.

 

I really hope Intel 7nm comes soon and isn't delayed like their 10nm process.

[porina]

1 hour ago, Sugadaddy said:

It's basically 3 players (Intel, TSMC and Samsung), it wouldn't be good for consumers if one goes away.

There's far more players in the fab space, but I presume you mean ones that are somewhat capable of playing within a generation or two of leading edge? 

 

1 hour ago, Sugadaddy said:

I really hope Intel 7nm comes soon and isn't delayed like their 10nm process.

It's already delayed. Question is how much?

[Vorg]

TSMC can't meet the needs of the customers they already have. Shortages will get worse and prices will skyrocket even more.