Why is this? Is TSMC not set up to make 10nm, anymore? Too much effort on TSMC part to be worth it?
TSMC's current high powered node is 7nm. They got 5nm mass production for mobile chips up and running (Apple mobile chips are 5nm already).
What I meant with regard to Intel is this: Intel were going to produce Xe GPUs on their own 7nm process. Intel's 7nm plans have recently suffered a major setback of approximately 12 months (maybe even more) because of their manufacturing issues.
If Intel want to avoid delaying Xe, thus risking the design and architecture to become outdated in the meantime, they would need to outsource the manufacturing. According to rumors they are in fact in talks with TSMC to let TSMC produce the Xe chips.
Here is an
interesting article on Intel's ongoing dilemma.
How important do you think 7nm is, anyway? Even on 10nm, Intels current lineup are still roughly equivalent to AMDs 7nm chips. As far as gaming goes, anyway. They take the lead in CPU intensive "work" apps but I'd say that's mostly due to greater core count. But, still, Intel does well with 10 cores and if you could take that performance and imagine its 16, 32, etc, they'd be ahead.
Just to clarify: Intel's current Comet Lake CPUs are not 10nm parts. They are still built on a 14nm node (14nm++…). While it is fairly amazing how much life Intel has squeezed out of their 14nm process, it is really showing its age now. The i9-10900K draws massive amounts of power and produces quite a lot of heat. It can boost up to 5.3GHz only for about a minute which is borderline cheating to get good benchmark results.
Zen 3 is technologically way superior at this point. More cores that stay much cooler while drawing reasonable amounts of power.
Intel's next generation of desktop CPUs is Rocket Lake which is expected to arrive at the end of the 1st quarter in 2021. Those CPUs will still be 14nm. They will have a maximum of eight "real" cores (down from ten). They will also feature eight small cores (called a big.LITTLE design because of the eight "real" cores and eight small helper cores).
Rocket Lake is a backport of a CPU that Intel originally intended to manufacture on their 10nm node but since 10nm desktop CPUs are still delayed (since 2016 according to old roadmaps!) it will be a 14nm part. It will be interesting to see a big.LITTLE CPU in the desktop space as that design is usually only found in mobile devices. Whether it will be enough to take back the crown from Zen 3 in gaming remains to be seen. In applications it certainly won't because AMD's 16/12 core CPUs are way ahead.
Intel's first 10nm desktop CPU is codenamed Alder Lake and it is finally supposed to come out in late 2021. I find it hard to believe that Intel will be able to pull off a launch in late Q1 (Rocket Lake) and another one in the same year in Q4 but we shall see. I would not rule out further delays. And 7nm is far future for Intel at this point. I would not even be surprised if that slips into 2024.
Maybe Intel will just skip 7nm and go straight to 5nm, which I believe is what AMD is planning for their next generation?
AMD's Zen 2 (last year) and this year's Zen 3 are built on a 7nm TSMC node. AMD will make the switch to 5nm (Zen 4) towards the end of next year or early 2022.
I very much doubt that Intel is going to skip anything. The 10nm process has a five year long history of delays at this point and 7nm which was supposed to become their savior node in 2022 has been delayed by at least 12 months.
TSMC on the other hand is on a roll. They have successfully executed every single node step in the last few years. The last known major problems occurred on the 20nm node back in 2015.
So, short of the PLA invading Taiwan or a major earthquake, I don't see anything stopping TSMC in their tracks while Intel has major execution problems.
Honestly, I'm not an Intel hater or anything but I really don't see how they are supposed to catch up with TSMC again anytime soon. Your suggestion to maybe take some shortcuts is a recipe for disaster in something as complex as leading edge node development.
Intel may be forced to abandon leading edge node development altogether just as Global Foundries has done leaving the market to TSMC and (to a lesser extent) Samsung. If Intel want to stay relevant they might very well have to become a TSMC customer, too, like the rest of the chip-making world.
What do you mean by this? What sort of concessions?
That is an excellent question especially when considering the political and economical climate between the US and China at the moment. China have the power to veto the nVidia/ARM deal. Convincing China to waive their veto is not going to be easy. If ARM becomes a company under US jurisdiction then Huawei could no longer license ARM chips because of US trade restrictions. This is why Huawei has asked Chinese regulators to veto the nVidia/ARM deal. It has become fairly quiet since nVidia and ARM have made their agreement public. The deal might go through in the end but at what cost?
Samsung is making the gaming cards now? Didn't know that. What do you mean by inferior?
The RTX 3080, RTX 3090 and RTX 3070 are manufactured on a custom 8nm process at Samsung which is a derivative of their 10nm process (kind of like the TSMC 12nm node of the RTX 20xx series was really just an advanced 16nm node).
The node is inferior in so far as the power draw is off the charts. The RTX 3080/3090 can reach nearly 400W of power draw. It also seems that yields aren't too great as the RTX 3080/3090 sold out almost instantly on launch day (mid/end September) and more supply has only been trickling into the market since then.