3 ms·

Feels like Microsoft during the Balmer years. Instead of Linux and Mac ascendant, you have AMD and ARM... it doesn’t seem impossible to think a Satya Nadella co

Feels like Microsoft during the Balmer years. Instead of Linux and Mac ascendant, you have AMD and ARM... it doesn’t seem impossible to think a Satya Nadella could turn things around.

SoSoRoCoCo 6y ago
Good observation. The only thing that kept MS going was the massive inertia of Win & Office. It bought them time to pivot to the cloud. And now we have things like VSCode, an ElectronJS app, which is the first microsoft product I've adored in two decades. And TypeScript, which has made the world a better place.
mhh__ 6y ago
Realistically, Intel only have to get one process node right to either be right back in the game or buy themselves years more time to play with. They are currently somewhere between 3-5 times behind TSMC lithographically, but nowhere near that far behind in raw performance, which suggests that they could at very least knock AMD off their perch in single-threaded work. Obviously process nodes are much harder than pretty much anything in the cloud, but this is a serious test of (pure and simple) whether their leadership has the (gender-neutral) balls to not back down.
- SoSoRoCoCo 6y ago
  > 3-5 times behind TSMC lithographically, I'm not entirely sure that is true. TSMC plays fast and loose with their definition of node. Any fab folks want to speak up? What I mean by that: https://www.pcgamesn.com/amd/tsmc-7nm-5nm-and-3nm-are-just-numbers https://www.pcgamesn.com/amd/tsmc-7nm-5nm-and-3nm-are-just-n... "And also goes some way to explaining why, despite TSMC offering a nominally 7nm process, the general consensus has been that Intel’s 10nm design is pretty much analogous. But what’s 3nm between fabs? At that level, probably quite a lot. But if the 7nm node is more of a branding exercise than genuinely denoting the physical properties of that production process then you can understand why there’s supposedly not a lot in it."
  - mhh__ 6y ago
    https://en.wikipedia.org/wiki/Transistor_count#MOSFET_nodes https://en.wikipedia.org/wiki/Transistor_count#MOSFET_nodes Look it up yourself. Intel being on 14nm vs (say) 5nm is actually quite flattering to Intel.
    - SoSoRoCoCo 6y ago
      That only lists the number, not what the # actually means in terms of actual lithography or more importantly, transitor performance. It used to be minimum feature size, or just L of the gate, but with finfet it can be an overloaded term. Scaling in x of .75 and y of .7 lead to 10% performance improvement per node at Intel. TSMC hasn't been that clear. And that doesn't even account for increase in metal layers, pitch of layers (or if they use poly lower layers to get even faster gains), or average track density due to above/below electromigration minimums. EDIT: All of this stuff is usually stated at ISSCC every time a new process is announced, so it isn't NDA. I haven't followed this in years which is why I was asking for a process person to step in.
      - mhh__ 6y ago
        And what information is there that isn't under NDA?
        brennanpeterson 6y ago
        Everything in M1 and everything on Intel 14nm. Go read wikichip. It is there. IEDM just happened, tons of detail there. 3nm, sure, good luck.
    - brennanpeterson 6y ago
      Intel 14nm is 42/70/52 (fin/cgp/mmp) with a 7.5 track height. HD stamBitcell is 0.05um2 Very restrictive design rules Tsmc 5nm is 27/54/36 with a 6T and 7.T option. More open design rules. HD SRAM cell is about 0.025um2. This is almost exactly what you expect for 1 node of litho scaling. The more difficult part is that the designs and related process optimizations for intel are different.
- kllrnohj 6y ago
  > They are currently somewhere between 3-5 times behind TSMC lithographically No they aren't. The nanometer names are only kinda comparable within a single company, but they are not comparable beyond that. It's like saying that Chrome is 6x further ahead than Safari because Chrome is on 87 and Safari is only on 14. Same thing with fabs here, it's just a version number that counts down and has been for a long while. In this case specifically, Intel's 10nm is more or less the same density as TSMC's 7nm. TSMC's 5nm is then a significant 1.8x the density of their 7nm process, so right now Intel's fabs are ~identical to what AMD is using (7nm), but Apple's use of TSMC's 5nm then puts them a generation ahead. TSMC is leading, but by a single generation not by 3-5x or whatever. Or to put it in year terms TSMC is about 2 years ahead of Intel right now, Intel just now getting "volume" 10nm up & running while TSMC was at a comparable node in volume 2 years ago. As far as "performance" goes, though, so far TSMC, even their 5nm process, cannot quite match Intel. Remember that as far as the fab goes, performance is just what clock speeds can be hit, and Intel's clocks are the highest - they have the fastest transistors. 14nm+++++ is a meme/joke, but the actual transistor speed of it is still downright impressive. How fast a CPU actually runs depends on the architecture, though, which is influenced by density but in terms of pure single-core performance that's really almost entirely just architecture.
  - mhh__ 6y ago
    Intel still can't make 10nm in any decent volume. TSMC are shipping chips at something like 170MTr/mm^2, the bulk of Intel's production (even their new chips) are still closer to 40-50MTr/mm^2
    - kllrnohj 6y ago
      Intel can't make large 10nm dies, or at least isn't, but there's quite a few 10nm mobile parts and they are on their second shipping generation of those parts. Which are shipping in volume, or at least volume-enough for the likes of Dell to put the 10nm CPUs in most of their lineup including the XPS series. Which is about where TSMC's 7nm was 2 years ago, too. And TSMC's 170MTr/mm^2 is still restricted to a single vendor and hasn't ramped up beyond that yet. You can't simultaneously claim Intel is restricted to what their bulk volume lineup is and then use TSMC's non-volume lineup. The bulk of TSMC is still 7nm which is 100MTr/mm^2.

Feels like Microsoft during the Balmer years. Instead of Linux and Mac ascendant, you have AMD and ARM... it doesn’t seem impossible to think a Satya Nadella could turn things around.

SoSoRoCoCo 6y ago

Good observation. The only thing that kept MS going was the massive inertia of Win & Office. It bought them time to pivot to the cloud. And now we have things like VSCode, an ElectronJS app, which is the first microsoft product I've adored in two decades. And TypeScript, which has made the world a better place.

mhh__ 6y ago

Realistically, Intel only have to get one process node right to either be right back in the game or buy themselves years more time to play with. They are currently somewhere between 3-5 times behind TSMC lithographically, but nowhere near that far behind in raw performance, which suggests that they could at very least knock AMD off their perch in single-threaded work. Obviously process nodes are much harder than pretty much anything in the cloud, but this is a serious test of (pure and simple) whether their leadership has the (gender-neutral) balls to not back down.

SoSoRoCoCo 6y ago
> 3-5 times behind TSMC lithographically, I'm not entirely sure that is true. TSMC plays fast and loose with their definition of node. Any fab folks want to speak up? What I mean by that: https://www.pcgamesn.com/amd/tsmc-7nm-5nm-and-3nm-are-just-numbers https://www.pcgamesn.com/amd/tsmc-7nm-5nm-and-3nm-are-just-n... "And also goes some way to explaining why, despite TSMC offering a nominally 7nm process, the general consensus has been that Intel’s 10nm design is pretty much analogous. But what’s 3nm between fabs? At that level, probably quite a lot. But if the 7nm node is more of a branding exercise than genuinely denoting the physical properties of that production process then you can understand why there’s supposedly not a lot in it."
- mhh__ 6y ago
  https://en.wikipedia.org/wiki/Transistor_count#MOSFET_nodes https://en.wikipedia.org/wiki/Transistor_count#MOSFET_nodes Look it up yourself. Intel being on 14nm vs (say) 5nm is actually quite flattering to Intel.
  - SoSoRoCoCo 6y ago
    That only lists the number, not what the # actually means in terms of actual lithography or more importantly, transitor performance. It used to be minimum feature size, or just L of the gate, but with finfet it can be an overloaded term. Scaling in x of .75 and y of .7 lead to 10% performance improvement per node at Intel. TSMC hasn't been that clear. And that doesn't even account for increase in metal layers, pitch of layers (or if they use poly lower layers to get even faster gains), or average track density due to above/below electromigration minimums. EDIT: All of this stuff is usually stated at ISSCC every time a new process is announced, so it isn't NDA. I haven't followed this in years which is why I was asking for a process person to step in.
    - mhh__ 6y ago
      And what information is there that isn't under NDA?
      - brennanpeterson 6y ago
        Everything in M1 and everything on Intel 14nm. Go read wikichip. It is there. IEDM just happened, tons of detail there. 3nm, sure, good luck.
  - brennanpeterson 6y ago
    Intel 14nm is 42/70/52 (fin/cgp/mmp) with a 7.5 track height. HD stamBitcell is 0.05um2 Very restrictive design rules Tsmc 5nm is 27/54/36 with a 6T and 7.T option. More open design rules. HD SRAM cell is about 0.025um2. This is almost exactly what you expect for 1 node of litho scaling. The more difficult part is that the designs and related process optimizations for intel are different.
kllrnohj 6y ago
> They are currently somewhere between 3-5 times behind TSMC lithographically No they aren't. The nanometer names are only kinda comparable within a single company, but they are not comparable beyond that. It's like saying that Chrome is 6x further ahead than Safari because Chrome is on 87 and Safari is only on 14. Same thing with fabs here, it's just a version number that counts down and has been for a long while. In this case specifically, Intel's 10nm is more or less the same density as TSMC's 7nm. TSMC's 5nm is then a significant 1.8x the density of their 7nm process, so right now Intel's fabs are ~identical to what AMD is using (7nm), but Apple's use of TSMC's 5nm then puts them a generation ahead. TSMC is leading, but by a single generation not by 3-5x or whatever. Or to put it in year terms TSMC is about 2 years ahead of Intel right now, Intel just now getting "volume" 10nm up & running while TSMC was at a comparable node in volume 2 years ago. As far as "performance" goes, though, so far TSMC, even their 5nm process, cannot quite match Intel. Remember that as far as the fab goes, performance is just what clock speeds can be hit, and Intel's clocks are the highest - they have the fastest transistors. 14nm+++++ is a meme/joke, but the actual transistor speed of it is still downright impressive. How fast a CPU actually runs depends on the architecture, though, which is influenced by density but in terms of pure single-core performance that's really almost entirely just architecture.
- mhh__ 6y ago
  Intel still can't make 10nm in any decent volume. TSMC are shipping chips at something like 170MTr/mm^2, the bulk of Intel's production (even their new chips) are still closer to 40-50MTr/mm^2
  - kllrnohj 6y ago
    Intel can't make large 10nm dies, or at least isn't, but there's quite a few 10nm mobile parts and they are on their second shipping generation of those parts. Which are shipping in volume, or at least volume-enough for the likes of Dell to put the 10nm CPUs in most of their lineup including the XPS series. Which is about where TSMC's 7nm was 2 years ago, too. And TSMC's 170MTr/mm^2 is still restricted to a single vendor and hasn't ramped up beyond that yet. You can't simultaneously claim Intel is restricted to what their bulk volume lineup is and then use TSMC's non-volume lineup. The bulk of TSMC is still 7nm which is 100MTr/mm^2.