Slashdot Mirror
Intel Unveils Roadmaps For Core Architecture and Atom Architecture (anandtech.com)
Intel on Wednesday surprised a number of people when it shared not one roadmap on CPUs, but two. AnandTech: For the high performance Core architecture, Intel lists three new codenames over the next three years. To be very clear here, these are the codenames for the individual core microarchitecture, not the chip, which is an important departure from how Intel has previously done things. Sunny Cove, built on 10nm, will come to market in 2019 and offer increased single-threaded performance, new instructions, and 'improved scalability'.
Willow Cove looks like it will be a 2020 core design, most likely also on 10nm. Intel lists the highlights here as a cache redesign (which might mean L1/L2 adjustments), new transistor optimizations (manufacturing based), and additional security features, likely referring to further enhancements from new classes of side-channel attacks. Golden Cove rounds out the trio, and is firmly in that 2021 segment in the graph. Process node here is a question mark, but we're likely to see it on 10nm and or 7nm. Golden Cove is where Intel adds another slice of the serious pie onto its plate, with an increase in single threaded performance, a focus on AI performance, and potential networking and AI additions to the core design. Security features also look like they get a boost.
The lower-powered Atom microarchitecture roadmap is on a slower cadence than the Core microarchitecture, which is not surprising given its history. The upcoming microarchitecture for 2019 is called Tremont, which focuses on single threaded performance increases, battery life increases, and network server performance. Based on some of the designs later in this article, we think that this will be a 10nm design. Following Tremont will be Gracemont, which Intel lists as a 2021 product. Beyond this will be a future 'mont' core (and not month as listed in the image).
Willow Cove looks like it will be a 2020 core design, most likely also on 10nm. Intel lists the highlights here as a cache redesign (which might mean L1/L2 adjustments), new transistor optimizations (manufacturing based), and additional security features, likely referring to further enhancements from new classes of side-channel attacks. Golden Cove rounds out the trio, and is firmly in that 2021 segment in the graph. Process node here is a question mark, but we're likely to see it on 10nm and or 7nm. Golden Cove is where Intel adds another slice of the serious pie onto its plate, with an increase in single threaded performance, a focus on AI performance, and potential networking and AI additions to the core design. Security features also look like they get a boost.
The lower-powered Atom microarchitecture roadmap is on a slower cadence than the Core microarchitecture, which is not surprising given its history. The upcoming microarchitecture for 2019 is called Tremont, which focuses on single threaded performance increases, battery life increases, and network server performance. Based on some of the designs later in this article, we think that this will be a 10nm design. Following Tremont will be Gracemont, which Intel lists as a 2021 product. Beyond this will be a future 'mont' core (and not month as listed in the image).
More pci-e lanes?
Long live RISC!
A Slashdot article that links to the very previous Slashdot article??
Increases in parallel processing should allow simultaneous bi-directional Beowulf cluster submissions unless your a Luddite.
Apps
There. Now the nerds don't have to fret.
That would be pretty impressive if Intel can get commercial chips at 10nm by 2019.
I wonder if Meltdown is fixed.
The biggest problem I see that most software is build around single threads. Back in the day I took Parallel processing course as an elective for my undergrad. It was a small class, (As its time conflicted with the graphics class, which was much more popular) however it really opened my mind on the actual power of Parallel processing. However most application that are running just use one thread, and you can see only one CPU doing all the work, so while the process may take a long time, it doesn't get split up.
I am happy to see more focus on single core speed because unfortunately that is still needed.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
The real underdog in the CPU race.
Get with the times. Alien tech is over 70 years old now. And ILLEGAL ALIEN TECH to boot. I rest my case.
From the summary: "Golden Cove is where Intel adds another slice of the serious pie onto its plate"
Better be carefull with that pie! The hot vapor will probably burn your mouth!
So, until Golden Cove shows up, it will be tweaks and adjustments to what is currently out there, which hasn't been much of an improvement in over three years now. The people at AMD will probably be celebrating, because that means they have 2019 and 2020 where Intel won't have a significant design improvement to compete with their Zen2 based products that will be out in the next quarter(exact timeframe should be announced at CES).
Cache improvements....yea, it will help, but won't be a significant redesign of the CPU design.
Intel owes us money. They lied for decades and betrayed us all for the NSA.
Fuck the NSA.
Fuck Intel.
Most algorithms are mostly parallelizable. And in the real world, you usually have many data sets, like client requests, pixels, game actors, threads, etc, that are all processed in parallel.
The only reason that parallel programming is not as prevalent as it should be, is that C/C++-like languages and coding styles are really bad at that sort of thing, yet dominate. And people still are "new" to doing serious parallel programming due to that. It's still very much treated stepmotherly.
Look at thinks like Haskell, where the compiler always knows what evaluation is required for what, and hence there is nothing in the language preventing automatic parallelization of everything possible. But the compiler is still written in C++, making it way harder. You can still use thread sparks, to get it mostly done.
Coders need to step up, and need to take parallel programming seriously! 1024 threads should be doable blindfolded, with the keyboard behind their backs.
No matter. They screwed over their customers the last few years, and hence I will not even remotely consider buying from them now.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Threads are not always the only answer to parallelism. With tools like OpenCL that takes advantages of GPU cores we can take more of a SIMD (Single Instruction Multi-Data) approach to coding.
In many ways this makes it easier, because there is less of a timing issue and conflicts that can go on. Because you are sending one instruction at a time, however it in parallel are doing it with multiable data sets.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
Most applications spend most of their time in standard libraries, or should. (Some silly developers implement sort themselves, repeatedly).
If popular libraries used parallel programming, applications would get most of the benefit for free. Sorting is a good example because it's expensive, but like many expensive operations it's already implemented in the library.
A huge percentage of code is now run behind a web server, as a web service / microservice / whatever, or some other type of server. The developers write code that Apache or IIS or MS-SQL calls, with the server handling requests. If the API for these types of modules assumed that each request is independent, disallowing modules / scripts reaching outside of the concurrency-safe context without a special global_ call, that could go a long way.
Again I'm not saying that IIS or SQL Server would make it *impossible* to effect global state, but doing so would be a special call, with the normal API being thread safe.
*VOMIT*. Can we draw and quarter every marketing drone in the tech sector yet?
What a steaming piece of shit. No thanks Intel.
Assume I meant SIMD, threads, and everything like that. Parallel programming, as I said.
Servers: could be multithreaded, but in practice many loads are not.
At least for web servers, PHP runs in a multiprocess model, and the PostgreSQL or MariaDB server runs in a separate process.
Browsers: JavaScript.
Script is multithreaded by operating in multiple documents, and even script within one document is multithreaded through Web Workers.
There is no announcement at all on Intel fixing the Hyper-thread security issues.
Instead of fixing the Hyper-thread security issues, and relaunch a more robust version of it, it looks like Intel has taken the easy way out, with their apparent abandoning of Hyper-thread.
RIP, Hyper-thread !!
The biggest problem I see that most software is build around single threads.
That's not a big problem, because most single-thread software doesn't consume much CPU. Most software which does need a lot of CPU is parallelizable, and in fact that has already been done. CODECs, compression algorithms, compiles, renders, photoshop filters... all are already multithreaded.
For the average user, literally the only thing they are doing which would benefit from more single thread performance is gaming. Ultimately gaming still comes down to single threads for single jobs, like rendering. That's why you get better minimum frame rates with Intel than with AMD, the superior single thread performance. That assumes, of course, that you haven't been owned and had Satan summoned all over your hard drive because you were running Intel.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"