How the heck does Intel switches to smaller nodes so quick?

[Chronobodi]

Like, the 32nm is just out barely starting with the 980x and the Arranedales, and now there's a 22nm slated for next year already... it's ridiculous...

Isn't it like, once they have a node up and running, they're like, "OK GUYS LET'S DUMP THIS NODE AND KEEP GOING ON!"

like, AMD is still on a mature 45nm node and that works ok for them... i don't know, it just seems way too quick to kill off 32nm this time.

 

[EXT64]

AMD is going to 32nm very soon. However the answer to your question I believe is the Pentium 4. After getting burned during that period of little innovation, Intel has been pushing their engineers hard to keep their tech moving. And Intel has the money and fabs (and brains of course) to keep up this pace. Don't worry, soon the node shrinks will get a lot more difficult (and exciting).

 

[Chronobodi]

so, what happens when we get to the 10nm-1nm node? Then what? Quantum Nodes?

 

[gamerk316]

^^ The limit is probably 16nm or so. After that, we'll probably need to switch to quantum or optical computers...

 

[1965ohio]

Most of the switches are not as quick as they seem. Most of this has been in the planning for a few years. The processes are just getting more successful yields and less mistakes so they are being released faster. With AMD, low yields on 4 cores just means more x3 or x2 CPUs for them to sell. Other than the Core 2 Solo... they just don't disable a core like AMD. They shoot for the most successful productions methods which gives them the highest yields. With Intel, if the yields are poor, they just disable the cache or some other features and label it a Celeron (even Celeron has dual cores). Since the progress on die shrinks has been good for Intel, they are able to release them at a faster pace than the past.

 

[g00fysmiley]

simple answer... intel has alot more money to spend on R&D, amd will be at 32nm here shortly unless there are yield problems, but supposedly that got straightened out

but yes it appears amd will be on 32nm while inel moves to 22nm (like how intel has been on 32nm for a while and amd still on 45nm... but amd still did well in comparison in power consumption and heat on a larger die so it hasn't been a huge deal

 

[EXT64]

Also, with so many fabs, they can do their upgrades in parallel with little risk of hurting production of current products.

 

[MU_Engineer]

Like, the 32nm is just out barely starting with the 980x and the Arranedales, and now there's a 22nm slated for next year already... it's ridiculous...

Isn't it like, once they have a node up and running, they're like, "OK GUYS LET'S DUMP THIS NODE AND KEEP GOING ON!"

like, AMD is still on a mature 45nm node and that works ok for them... i don't know, it just seems way too quick to kill off 32nm this time.

The reason Intel can move from node to node so quickly is that they have about 20 times the market cap of AMD's fab, Global Foundries, and can afford to do so. Considering that building a 22 nm fab would cost about as much as Global Foundries is worth, one can see why they would stay on process nodes longer. They want to amortize their fab equipment a bit more and make more money before spending billions on new equipment. Plus, there's a lot more to a CPU's performance than just the transistor size. Yes, having smaller transistors allows you to put more transistors on a given die size and likely clock it a bit higher as well. But the overall design of the CPU and the specific process you are using plays a big role as well. Look at how AMD's K8 microarchitecture on their excellent 90 nm silicon-on-insulator process absolutely embarrassed Intel's NetBurst P4s and Pentium Ds on 65 nm.

 

[jimmysmitty]

MU has it right. Plus Intel has been at this for a long time. And even during the Pentium 4 era, Intel still was the leader in process fabrication. In fact the Pentium D on Prescott (90nm) sucked but Ceader Mill (65nm) was actually much better. Most CPUs were 65w TDP parts then. And everyone else was still behind, AMD was behind by about 2-3 years in process tech then. Now its 1-2 years they are behind.

but Intel has the money (the largest part of their earnings goes back into R&D and FAB production), resources and as I said before, have been doing it a long time.

its like with HighK/Metal Gate technology. People think it just appeared at Intel but in reality they started it about 5 years before it even hit the market because they knew that there would be a soon hit limit with just plain silicon.. Just like Intel started to design Sandy bridges Quick Sync 5 years ago.

 

[MU_Engineer]

quick answer

MOOREs LAW

"Moore's Law" is a general observation Gordon Moore made about how the number of transistors in CPUs increased over time, not a rule that Intel must slavishly adhere to. Moore himself even said in 2005 that his observed trend will eventually prove false as the limits of transistor miniaturization are reached, due to the fact that you get down to the size of a few atoms' thickness in transistor gates and the transistors no longer work properly.