News TSMC and Intel rumors stoke Taiwanese fears of losing the 'Silicon Shield

[Admin]

Taiwanese media criticizes potential collab between TSMC and Intel.

TSMC and Intel rumors stoke Taiwanese fears of losing the 'Silicon Shield : Read more

 

[bit_user]

I think there's no realistic chance of TSMC IP being transferred to Intel.

I also doubt TSMC is going to put their most cutting edge nodes in the US, as much for practical reasons (i.e. keeping them located in the same place the main R&D is happeing) as much as anything else. Even after a node is deployed, a lot of tuning and refinement is still happening to get yields up and costs down. So, there will necessarily be some delay in these nodes reaching fabs located outside of Taiwan.

 

[paviko]

I don't get it. Intel 7 ( Core 13th and Core 14th?) served well against AMD TSMC 4. With X3D exception, which is for gamers, Intel is really good. Also 13900 and 14900 where more like overclocked CPU needed only for top mark scores. 13700, 14700 have good efficency. Now we have Core with TSMC 3 and actually only 285 (equivalent of 14900?) shows much better power charecteristic and even is not fastest. The same was with Apple SoC TSMC 5 vs TSMC 3, not much improvement, in benchmarks only because new instructions are used aka AVX.
Another point is Core architecture that is sooo old. I also don't get why no new arch after 15 years.

 

[JamesJones44]

in benchmarks only because new instructions are used aka AVX

Apple silicon has always supported AVX2 via Rosetta 2 translation (aka x86/x64 compiled applications, not aarch64 applications), it does not support AVX-512 on any line. Apple Silicon does support SIMD instructions natively via NEON. As far as I know M4 and M3 lack full SVE/SVE2 support which is somewhat equivalent to AVX-512. Which vectorization instruction set you are referring to as "in benchmarks only because new instructions are used aka AVX"?

 

[Gururu]

In current environment there will be no compromise from U.S. about who should lead the space. If we can pull out of Euro allies so easily, Taiwan (or TSMC) is dead in the water.

 

[bit_user]

I don't get it. Intel 7 ( Core 13th and Core 14th?) served well against AMD TSMC 4.

Zen 4 used TSMC N5 for the CCDs. And you're just focusing on desktop. AMD designs the CCD chiplets for servers, first and foremost. In that realm, there was no contest - Genoa and Siena both trounced their Xeon competitors. It wasn't until Intel launched Sierra Forest (made on Intel 3) that they truly had an answer. However, right after Granite Rapids launched, AMD had Turin and that took us back to a place of AMD leadership. It's only in a minority of specific workloads, like those harnessing AMX, that Granite Rapids can beat Turin.

Another point is Core architecture that is sooo old. I also don't get why no new arch after 15 years.

Intel changes things as they go. I'm sure there's no part of their latest cores that hasn't been redone several times, since the Nehalem cores launched way back in 2008.

Also, I consider the E-cores to be a newer microarchitecture. I think Tremont was the last massive overhaul, but Gracemont and Skymont were pretty major updates.

 

[bit_user]

Apple silicon has always supported AVX2 via Rosetta 2 translation (aka x86/x64 compiled applications, not aarch64 applications),

Only by translating them to 128-bit NEON ops. As you said, Apple still doesn't meaningfully support SVE!

 

[NinoPino]

Taiwan will loose his "shield", only at the moment they will decide to mass product the most advanced node in the Arizona facility.
I hope TSMC management, and Taiwan government, will be able to resist to the U.S. government pressure for transferring the leading edge production to the U.S.A.

 

[Mcnoobler]

Intel in the US... or TSMC in Taiwan/AZ...

As a US citizen, who should I root for... hmmm...

Save Taiwan at all costs. Give them whatever they need!