时光旅行永远是一种风险。When you go to the past, there’s the usual dangers like making sure you don’t talk to your past self, not stepping on any bugs so you don’t trigger the butterfly effect, etc… and when you go to the future, there is the danger that your expectations could be a million miles off of what actually happens.
We’ll do our best to base our speculations on available evidence, in the hopes that they won’t be a百万miles off of reality (maybe only a thousand miles). So, let’s take a look at some future manufacturing tech, and see who might come out on top in the next era of the ongoing battle between red and blue.
不过，这不是整个图片。它确实有帮助他们 - 大概有点，大概是 - 他们是唯一拥有一贯可用股票的品牌，AMD仍在努力跟上对处理器的需求（稍后再详细介绍）。
Regardless of whether these couple small points in Intel’s favor are outweighed by AMD’s multi-generation dominance, there is light at the end of the tunnel for Intel:
流星湖looks close to its final design stage, with英特尔的新首席执行官帕特·盖辛格（Pat Gelsinger）宣布他们将在今年晚些时候将其钉住（至少就IP而言）。
This is a tough one to tell for sure, especially in light of the ‘back-engineered 10nm design to 14nm’ for Intel’s 11th gen, which has been met with a solid ‘meh’ from the tech field.
Some were surprised that this development is in partnership with TSMC. The manufacturer should be well known to AMD and Apple fans, but Intel has traditionally made its own CPUs.
- At least internally, Intel are not afraid to cut their losses on a design. Part of the problem with 10nm was Intel trying (and failing) for so many years to get it to work. This demonstrates to a lot of Intel’s partners that they now prefer to just get a good product out the door.
- TSMC has a proven track record. Honestly, at this point, when you’re dealing with such small margins in the design manufacturing process, it makes sense to go to a company which is already successfully shipping small architecture designs.
一般来说,第一代新process will see modest improvements. For Intel, the big targets for it now are probably cooling and power delivery. However, there could be some reasonable IPC improvements along with this (but nothing too crazy).
Where the 7nm process comes into its own, though, will be in the improvements Intel are touting with its ‘Foveros chip-stacking’ design. This should in theory allow them to jump up in core count to once again go like for like with AMD (if not higher), all while being far cooler and more power-efficient when than Intel’s beyond-aged 14nm process.
更重要的是，一旦释放流星湖，目的是让英特尔回到their “Tick-Tock” approachto shrinking and then optimizing the company’s chip technologies about every two years.
My Magic 8 Ball Prediction
The, Ahem, “Intel i9-14900K”
Cores/Threads:12 core, 24 thread
Boost (all core):5GHz
There is an awful lot of speculation here, but some details—like the cost—are things I’m more comfortable guessing, because Intel will have to be at least a touch aggressive on pricing (based on the rapid price cuts on their 11th generation).
I hope that the jump into 7nm will allow Intel to push their clocks higher (both single and all-core), which, along with IPC improvements and the jump back to higher core counts (thanks to the smaller die), will make them competitive again.
Let’s be honest, Intel could use something along those lines right now! But even in a couple years, we probably wouldn’t complain.
However, I’m saying the Magic 8 Ball reports ‘Don’t count on it’ because I’m not entirely convinced Intel can reach these idealized expectations. In fairness to them, AMD struggled with their first-gen Zen CPUs. It was only really their 2000- and 3000-series Ryzen options where they started to gain steam, and 5000-series chips where things took to the skies. That’s a lot of iteration before being top dog. Which Intel really needs to do to get things right, and get back to the top of the pile.
AMD 5nm (Zen 4 / Zen 5 “Raphael”)
So, with AMD already using TSMC too, as you might imagine the pandemic has been a spanner in the works for their long-term plans.
On paper, Zen 3 is a success. It has been a wildly popular launch, and is a great way to effectively sign off on this generation of CPUs running on the AM4 socket. It’s a return to form for team red, and a good situation for builders to be in, as competition gives us far more performance for our money.
The AM4 socket has not been without issues for AMD. They had to quickly backtrack onchipset compatibility with B450 vs B550。现在消费者已经内化的消息rom AMD that everything after Zen 3will需要更换插座。
据报道，ZEN 4热那亚工程样本的执行速度比现有的ZEN 3 CPU的速度快29％，并且以相同的时钟速度和核心计数。
So, although Genoa is the EPYC (server) design, the core architecture is more-or-less what will be in Zen 4 desktop solutions too. This should rightfully scare Intel, as they’ve only just got themselves more or less back on-par with single-core performance in their 11th-gen CPUs. Zen 3 (specifically the 5950X) still rules the roost here. So, if the improvement is even vaguely close to 29%… watch out, Intel.
As with the previous generation’s improvements, we can expect some reasonable clock advances (let’s see 5GHz+ on the box, please) while also seeing some modest improvements to cooling and power delivery.
So, When is it Coming?
Original plans had EPYC chips “Genoa” and the mainstream CPUs “Raphael” making an appearance in late 2022. However, as mentioned above, TSMC (like every other manufacturer) has struggled hugely to meet demand through 2020 and 2021, and there’s no easy estimates available on how long before things are normalized.
To add a bit of spice to the mix, though, my personal prediction is that this will slip into 2023, right on time to compete with Intel’s 7nm offering.
My Magic 8 Ball Prediction
“It is decidedly so”
因此，这就是我认为会发生的事情。The flagship desktop system CPU by AMD may look as follows:
哎呀，“ AMD Ryzen 9 6950x”
Cores/Threads:16 core, 32 thread
Boost (all core):4.95GHz
I hope that the continued development into 5nm won’t result in more cores, but rather more optimization of the CPU itself. (At this point, AMD’s high-tier chips have far more cores than most programs know what do with, so developers need some time to play catch-up.) Improving the IPC and boost clock performance would be a huge bonus, although I suspect I may possibly be over-estimating the scale of that jump in boost clocks.
That is why, although fundamentally the core should become a little more efficient when going down to 5nm, the TDP stays the same as our current generation. That push for higher boosts is going to need some power to support it for sure. It’s more about how far AMD can push that balance that by 2023.
Basically, a lot of these guesses come down my own suspicion that AMD will remain on top of the pile into 2023. They have a proven design now, and they’re improving on it with each generation of Ryzen. 5nm was always the “end” of the journey as far as they’ve told us so far. Yet, with TSMC testing their 3nm process, it might not be too long into the future before we see Ryzen chips on that…
You know, it was only four years ago that I was写关于Ryzen产品的发布for the first time.
It is due to this recent track record that I’m giving my prediction on the winner of the short-term fight to AMD. Looking forward from now to 2023, they are likely to remain in a far better space than Intel as they are showing no signs of slowing down (at least for now).
“Ask again later”
Am I simply being an AMD fanboy? Are Intel hiding something major to change the game that I’ve overlooked?