Slashdot Mirror
Intel's 10nm 'Cannon Lake' Processors Won't Arrive Until Late 2019 (digitaltrends.com)
At the company's second quarter 2018 financial results conference call, Intel chief engineering officer Venkata Renduchintala said the "Cannon Lake" 10mn processors won't appear in products until the 2019 holiday season. "The systems on shelves that we expect in holiday 2019 will be client systems, with data center products to follow shortly after," Renduchintala said. Interim CEO Robert Swan went on to tout the company's "very good lineup" of 14nm products. Digital Trends reports: "Recall that 10nm strives for a very aggressive density improvement target beyond 14nm, almost 2.7x scaling," Renduchintala said during the call. "And really, the challenges that we're facing on 10nm is delivering on all the revolutionary modules that ultimately deliver on that program." Although he acknowledged that pushing back 10nm presents a "risk and a degree of delay" in the company's road map, Intel is quite pleased with the "resiliency" of its 14nm roadmap. He said the company delivered an excess of 70 percent performance improvement over the last few years. Meanwhile, Intel's 10nm process should be in an ideal state to mass produce chips towards the end of 2019.
Intel's Cannon Lake chip is essentially a shrink of its seventh-generation "Kaby Lake" processor design. Given the previous launch window, the resulting chips presumably fell under the company's eighth-generation banner despite the older design. But with mass production pushed back to late 2019, the 10nm chips will fall under Intel's ninth-generation umbrella along with CPUs based on its upcoming "Ice Lake" design. Intel claims that its 10nm chips will provide 25 percent increased performance over their 14nm counterparts. Even more, they will supposedly consume 50 percent less power than their 14nm counterparts.
Intel's Cannon Lake chip is essentially a shrink of its seventh-generation "Kaby Lake" processor design. Given the previous launch window, the resulting chips presumably fell under the company's eighth-generation banner despite the older design. But with mass production pushed back to late 2019, the 10nm chips will fall under Intel's ninth-generation umbrella along with CPUs based on its upcoming "Ice Lake" design. Intel claims that its 10nm chips will provide 25 percent increased performance over their 14nm counterparts. Even more, they will supposedly consume 50 percent less power than their 14nm counterparts.
> merely shrinking the existing architecture then that means they still haven't fixed the fundamental issue behind the Meltdown vulnerability.
That fundamental issues won't be changed in the next ten years, if ever. They can either keep playing whack-a-mole with different hardware and microcode side-effects, or you can add a very simple (and slow) separate CPU for security-sensitive operations.
Current CPUs are very complex, with out-of-order execution, speculative execution based on branch prediction, multiple concurrent threads of execution, various different types of caches, etc. All of this complexity is there for a good reason - it makes a huge improvement in performance. For that reason, it's not going away, we're not going back the 8086. All the complexity also means operations will effect caches and predictive microcode and other things, so CPU operations will have side effects. Side effects mean you get Spectre and Meltdown style vulnerabilities.
A very simple CPU which doesn't have any modern optimizations (complexity), with a single core running one thread at a time, could be much more secure in this regard. It would also be much slower, so it wouldn't be good as the main general-purpose CPU. It would need to be used to offload things like handling private keys that are particularly sensitive.
You have to post the hard truths as AC. The sad thing is though, there is no one to do it right. As long as racism like "affirmative action" is legal, we won't even have a successful example to point at, almost every company will cuck and virtue signal in varying degrees.
After Intel's laughable Netburst initiative (shilled by Slashdot at the time as 'genius'), Intel gave up the 'very long pipeline' race to 10GHz, and went back to the Pentium 3 architecture, that they crossed with AMD's advances used in the excellent AMD x64 chips of the time. Legal cos of cross-patent agreements between the two.
Pentium 3 + AMD tech = 'CORE', the horrid name Intel has used to describe all its architectures since Netburst (at first core 1/2, and now 'core'. Despite the confusing name, all 'core' Intel chips have one common feature- ZERO hardware protection of interthread memory access.
On a multi-threaded chip, you are supposed to use lock and key. A thread has a 'key' (thread id), and this key must be used to unlock a 'chest' containing any RAM access.
Lock and key takes a LOT of transistors. A lot of energy. And significant time delay added to RAM access. By secretly dropping this CS requirement, Intel gained a massive power and speed advantage over AMD.
Today, thanks to a genius CPU architect, AMD's zen has lock and key, and less than 10% disadvantage in IPC for software compiled to be optimum on Intel's core architecture (most commercial software). If software were optimised for zen (which can issue multiple complex instructions while Intel is optimised for 1 complex and 3 simple instructions) zen would have a greater than 10% advantage over Intel.
AMD's last downside is a 700Mhz gap with Intel (when both are clocked to sane max). Most chips sold do not show this gap, of course, since very few Intel chips are ultra high-end. Intel offers far more cores (and hyper-threading per core) than Intel per dollar.
Early 2019, AMD's Zen 2 (confusingly the new AMD zen parts from this year are zen+) will pass Intel on IPC, and almost catch up on max clocks. All this remember with zen having 'lock and key' and no Intel part til 2021 at the earliest fixing meltdown and spectre.
When IBM slected Intel to provide the dreadful 16-bit processor for IBM's home PC, every other chip company had better 16-bit designs, and some vastly better (Motorola). IBM selected Intel precisely because its chip was so awful (and thus didn't compete with IBM proprietary hardware). However Intel eventually used the mega profits from being the heart of the now generic PC design to create the excellent 486/Pentium 1, just before Intel illegally stole RISC tech from all major players to design the Pentium Pro/2 (for which Intel later paid billions in fines).
Since that date Intel's 'lead' has been a pure consequence of Intel outspending the competition by thousands to one in R+D (and even then AMD has had the lead over Intel on at least 3 periods).
Intel's final advantage was a 'process' lead- but as this article points out, that lead is GONE- TSMC, Samsung and GF are now ahead of Intel. Unless you game at 120 Hz, there is literally no reason to buy Intel today. Intel was always a lousy company. Now its social engineering policies have sunk the entire company.
PS can't use 'less than' and 'greater than' symbols in my text? WTF slashdot.