AMD APUs to Become Efficient Faster Than Moore's Law

[Brian_R170]

Has anybody plotted AMD and Intel's efficiency (performance per watt) improvements over the past 10 years or so? Just wondering if the goal is as lofty as it sounds or if it's not really far off from what we've actually been experiencing in the recent past.

 

[f-14]

Moore's law is not just about transistors. Its a philosophical idea that every two years chips will become either twice as powerful, OR twice as cheap to make.

Moore's original comment applied to transistor count but got extended to other areas that happened to fit at that time.

But no matter which Moore variant you look at for desktop CPUs though, things have been almost completely stagnant for the past three years. GPUs have been almost stagnant for quite a while too.

meant to +1 but hit the wrong comment in my excitement.

ya i was just trying to remember the transistor count in the Pentium 133MHZ MMX chip vs. a core i7 4770 so i could break down the numbers.

 

[falchard]

Only way for AMD to beat Intel at this point is to break Moore's law and probably Physics since 80% of consumers will still buy Intel even if its slower than AMD.

 

[randomizer]

last time I checked, Moore's law was about transistor count doubling every 2 years.

The observation was that number of transistors that can be placed on a chip at the minimum cost per transistor doubles every two years, not simply that the number of transistors doubles. Pure transistor count could probably increase at a substantially greater rate, but it would be prohibitively expensive.

 

[InvalidError]

Pure transistor count could probably increase at a substantially greater rate, but it would be prohibitively expensive.

Transistor count per chip is still increasing but instead of increasing on a per-die basis, it is increasing mostly through multi-die packages and die-stacking lately. It makes sense as a way to mitigate defect density rate and optimize individual processes for their specific uses when mixing otherwise incompatible things like DRAM and logic.

Imagine GPUs with 4GB of TSV-attached custom DRAM...

 

[Bloud Mai]

A slot based SOC chip solution with 8-12 X86\64cores and 64/128 compute cores with a built in DDR4/5 controller would do the trick. Computational power in the Terra Flop +' per second range utilizing HSA at 300ish W TPD "more performance per watt than any single chip on the market today" Current I7's hover around 150-175 GigaFlop's per second.

A little healthy competition never hurt anyone especially the consumer "Namely Us".

 

[sykozis]

If AMD closed, it's likely your intel chip might mean next to nothing when they're forced to discontinue x64 instruction sets in their next line of CPU's until they create their own (If they can do so without infringing on AMD's design), as it isn't likely AMD would sell them.

If AMD disappeared overnight with nobody taking over their assets, there would be no one administering the licenses and nobody to stop Intel from continuing to use the x86-64 extensions.

Even if someone did pick up AMD assets, the instruction set cross-licenses are most likely perpetual and non-transferable so whoever purchased AMD would likely be unable to revoke Intel's perpetual license.

If AMD were to go "out of business", the patents would still be valid. Patents don't become invalid just because the holding company fails. It's also very likely that Intel would attempt to buy the patent anyway.

If AMD were to go "out of business", any cross-licensing deals would become null and void as a "cross-licensing" deal requires 2 parties.

 

[InvalidError]

If AMD were to go "out of business", any cross-licensing deals would become null and void as a "cross-licensing" deal requires 2 parties.

A perpetual license is perpetual regardless of what happens to the issuing party.

If a perpetual license could be repealed at the issuing party's convenience, Intel would have shut down AMD a long time ago. Intel tried many times to do so in court and failed.

 

[faye__kane]

==-
Why was this article even posted? AMD didn't actually DO anything. They just said they sure wish they COULD do something 6 years from now.

It wouldn't even make an interesting press release.

faye kane ? girl brain

 

[Duckhunt]

AMD can win the battle. Just steal some of the underpaid key folk in Intel. Then reverse engineer and bang. You have it. AMD needs to learn from Mitt Romney and vulture capitalism.