4 ms·
Sadly, it’s looking like Intel won’t do 5nm until 2023 at which point TSMC is projecting 3nm. They won’t be able to catch up. On the other hand Intel is a huge
by akmarinov 6y ago
Sadly, it’s looking like Intel won’t do 5nm until 2023 at which point TSMC is projecting 3nm. They won’t be able to catch up.
On the other hand Intel is a huge national security interest for the US, so expect the government to help them out where possible.
- pastrami_panda 6y ago> On the other hand Intel is a huge national security interest for the US This. I'm actually surprised they have grown so uncompetitive considering the critical products their CPUs are presumably used in. But Intel has been a money making machine for many years, I don't think most people realize the amount of resources they've accumulated when they compare them to AMD et al. I just hope they don't make more lousy investments and actually focus on sorting out their core business problems.
- beezle 6y agoFirst (and I'm no expert here) I don't believe that (x)nm is the be all end all of cpu performance. Architecture still matters. Second, perhaps the AMD/Apple fan boys have not noticed but Intel has many times the volume of AMD and whatever ARM volume Apple will have (Apple is typically 10% of lap/desktop sales, up this year from Covid). The world requires Intel to meet its computing needs and no magic wand will give the other guys that capacity anytime soon. It is way to early to count Intel out.
- tomjen3 6y agoIf you can get it there is no reason not to buy an AMD product in a laptop, or to design one in, unless you absolutely need thunderbolt support. And laptops sells way more than desktops do. This is a winner takes all market and there is no reason to go with the loser.
- fomine3 6y agoIntel Ice/Tiger Lake Platform has still advantages for idle or light use (not heavy computing) power consumption on mobile thanks to their continuous improvement.
- dragontamer 6y ago> First (and I'm no expert here) I don't believe that (x)nm is the be all end all of cpu performance. Architecture still matters. The same architecture on a smaller process will use less power and use less silicon (usually making it cheaper: but yields make that equation complicated). A 100mm^2 design using 50W will only be 50mm^2 using only 25W on a smaller node. Since 300mm wafers are always 70,000 mm^2 in area, you now produce twice as many designs in 50mm^2 instead of 100mm^2. ------- Now what made Intel get stuck on 14nm+++ is that Intel got really good at increasing 14nm++++'s clock rate. Dennard scaling is dead: a larger process may offer better GHz these days. But if you can solve the yield issues, the march towards tinier and tinier transistors is a winner. Less power used, less chip-area used, and therefore cheaper and more-efficient chips.