You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Intel is redesigning the X86 ISA all over. The new architecture APX will have 32 registers and other features.
Remains to be seen if the software vendors like Microsoft will bite or not. But it will be hard for AMD to get a patent cross license for the new ISA this time.
Intel is redesigning the X86 ISA all over. The new architecture X86S will have 32 registers and other features.
Remains to be seen if the software vendors like Microsoft will bite or not. But it will be hard for AMD to get a patent cross license for the new ISA this time.
X86S is dead and the 32GPR Registers is APX AMD will likely copy them anyway.
AMD doesn't put resources into these stuff imo they simply follow Intel in these regards.
Yeah I meant APX. Look, the AMD X86-64 patents have expired. So AMD has nothing to counter offer Intel this time like they did with AVX.
Intel has tried to make X86 a monopoly several times in the past. This would be nothing new for them.
There were many companies with a X86 license in the past and now basically you only have AMD. For example National Semiconductor and NEC used to have 8086 licenses. Intel tried to kill these 3rd party licenses when the i386 came out. They failed with regards to AMD and Cyrix. Then Intel tried to kill the X86 clones again with the Pentium.
With the Pentium, AMD and Cyrix could no longer make exact hardware clones and ended up making their own chip designs.
Don't be surprised if Intel does not license APX and AVX10 to other companies. If Intel try to make X86 a monopoly again.
X86S is dead and the 32GPR Registers is APX AMD will likely copy them anyway.
AMD doesn't put resources into these stuff imo they simply follow Intel in these regards.
In the first place, AMD basically follows Intel's instruction implementation.
In the first place, AMD is a manufacturer of compatible machines, so compatibility under the same conditions is crucial.
N2 GAA and then CFET are independent of the substrate base wafer used. So essentially it doesnt matter if it is silicon or anything else. TSMC is the leader in GAA and will be in CFET.
On separating Intel, this has been done before
1) create separate company. Separate finances completely
2) Foundry is found to have severe financial issues that people do not want to pick up. It cannot IPO as it is a financial disaster.... Losses, Massive cash need, Massive depreciation
3) Foundry is "sold" to a new company with requirement that the new company provide wafer capacity to the product company. The sale price is negative. The new company accepts the depreciation and losses in exchange for payment which the product company lists as "pre-payment" on future wafer purchases
4) Product/Parent company take one time write off on amount paid to eliminate the foundry.
5) Product company goes on to be very successful and profitable.
In the first place, AMD basically follows Intel's instruction implementation.
In the first place, AMD is a manufacturer of compatible machines, so compatibility under the same conditions is crucial.
Can you imagine what TSMC will be like in 10 years?
Will this huge demand for semiconductors continue? Even if demand increases, if this is just a temporary silicon cycle, it will hurt companies that have made large investments.
I am more concerned about margin investments not in 10 years, but now, as the usual semiconductor cycles downturn is now 5-7 years overdue. You could think COVID was the point when the new cycle started, but fabs were working at full steam, powered by panic buying.
The presentations by Google and Nvidia at Hot Chips this were were impressive. The Intel coverage starts next week. Quite a contrast on technical detail.
FWIW, I don't think N3B is really a node advantage over N4P.
Per TSMC:
N4P provides an 11% performance at the same power over N5
and
N3B provides a 10-15% performance at the same power over N5
N3B probably has lower power consumption at lower clocks, but given the market need for 'more performance than last year', the clocks are always pushed as high as they'll go within the overall thermal limit.
Overview TSMC N3 is the 3-nanometer FinFET semiconductor manufacturing node developed by Taiwan Semiconductor Manufacturing Company (TSMC). It is a major milestone in TSMC’s process technology roadmap and represents the first step into the 3nm era, succeeding the N5 and N4 nodes. Unlike Samsung...
