4 ms·
Somewhat.. but a major factor is that TSMC is alone on the leading edge and doesn't have enough capacity for everyone.
by mlyle 6y ago
Somewhat.. but a major factor is that TSMC is alone on the leading edge and doesn't have enough capacity for everyone.
- redisman 6y agoSamsung should be good enough for anything but the super cutting edge (TSMC "5nm"). The Nvidia 3000 series GPUs for example are definitely the most powerful GPUs ever built and Samsung fabbed those.
- mlyle 6y ago> The Nvidia 3000 series GPUs for example are definitely the most powerful GPUs ever built and Samsung fabbed those. Yup, and has been unable to meet nV's demand-- such that this has been accused of being a "paper launch".
- fomine3 6y agoSamsung 8nm is improved 10nm. It's not equivalent to TSMC 7nm.
- 0xfaded 6y agoGiven that hopper is supposedly going to be on 5nm, and seeing the M1 performance, I actually cancelled my early 3090 order. I'm going to go with something cheaper and leave my big purchase until next year.
- bsder 6y agoNot really. As the article points out, this is about chips coming out on older (8" being "older" is funny) fab technologies. There are three major prblems: 1) Nobody holds inventory. This is the primary root of all problems. Because everybody is on just-in-time inventory, nobody can absorb even the slightest shocks. Maximally efficient is minimally robust. 2) Semiconductor industry consolidation The semi companies have been on a competition buying spree over the last couple of years. There's no need worry about a competitor holding inventory and winning a contract from you if you've bought them all out. 3) The "minimum volume" of a chip run through a fab is enormous. An 8 inch wafer is about 32,000 mm^2. If your chip is 2mm per side (pretty big for "popcorn" type chips), a single 8 inch wafer produces 8,000 of them. If it's 1mm per side, you've got 32,000 of them. You probably won't produce less than 10 wafers at a time. So, that means your inventory goes from dead 0 to roughly a quarter million in a single whack with an 18 week lead time (and I'm just counting wafer starts--I'm not even talking about testing and packaging resources getting choked). And most projects never reach a quarter million in volume--and the ones that do generally blow WAY past it in a hurry. Good luck managing that inventory flow.
- entropicdrifter 6y ago>Maximally efficient is minimally robust. I always like to tell my friends and family the same thing by saying, "Efficiency is not the same as efficacy"