More processor news today, as AMD has released a fair bit of information from their plans including Zen 2 and Zen+. Firstly, they've announced a price-cut for existing Ryzen processors, although that does depend on what store you go with. I've checked a few UK stores today and their prices have not yet been adjusted.
AMD has confirmed that the Zen CPU core (known as Ryzen) is moving to 12nm with samples now shipping. They are planning to launch these Zen+ 12nm CPUs in April of this year and they confirmed they will be clocked higher and come with "Precision Boost 2". Since this is only a refresh, there shouldn't be massive differences other than AMD tweaking the design to get the most out of it. AMD confirmed Zen+ will continue to use the AM4 socket, but they will also launch the X470 chipset optimized for Zen+.
Moving onto Zen 2, this 7nm design is now "complete", although they aren't giving out a date for that yet. Most likely launching next year, it's quite surprising that they've finalised it so quickly, it seems AMD are firmly back on top-form with their processors now so it's going to be a fun two years for AMD fans.
A little further out, but Zen 3 using 7nm+ is currently planned to be release in 2020, although very little information is available on that.
They will be releasing the Ryzen 5 2400G (4 cores, 8 threads) and Ryzen 3 2200G (4 cores, 4 threads) APUs which will both come with Radeon RX Vega graphics. These should be launching February 12th this year along with mobile processors that will also come with Vega graphics.
Also, AMD will be releasing Radeon Vega Mobile GPU through this year, which will be primarily aimed at ultra-thin notebooks.
See more details here.
Fun fact: Some of the press slides I was sent, had a picture of the Ataribox in it.
Article taken from GamingOnLinux.com.
Anyway, I might replace my i3570K with some Ryzen+ this year.
Quoting: EikeI wonder if immunity against Spectre and Meltdown can be built into chips and when the first CPUs having it will be announced.
Anyway, I might replace my i3570K with some Ryzen+ this year.
Ryzen is already said to be immune against Meltdown, and while it is hypothetically vulnerable to one of the two Spectre variants, it hasn't been demonstrated in practice.
And yes of course you could make a chip completely immune. Just get rid of speculative execution, as that's what these attacks exploit. That will come at a performance cost however.
Last edited by Brisse on 8 January 2018 at 12:06 pm UTC
Quoting: BrisseAnd yes of course you could make a chip completely immune. Just get rid of speculative execution, as that's what these attacks exploit. That will come at a performance cost however.
Of course, with 90's architecture, you can... ;)
I can't seem to find many laptops shipping with Ryzen APUs. HP Envy was the first and now Acer Swift 3. Don't find these appetizing.
Quoting: BrisseAnd yes of course you could make a chip completely immune. Just get rid of speculative execution, as that's what these attacks exploit. That will come at a performance cost however.I don't think the speculative execution has to be entirely removed to make CPUs immune. I am not an expert but from what I gathered the issue is that after code has been speculatively executed the result is stored in the cache even if the CPU guessed wrong and ran the wrong code, not to mention code that it didn't even have privileges to run. If these results were cleared from the cache when the CPU takes a wrong turn it would at least protect data that isn't supposed to be in the cache right now. And obviously the speculative execution must follow security protocol and not rush ahead and think of the privileges later (looking at you, Intel).
Overall I think this is a point where CPU engineers need to start thinking in terms of "smart" and not purely in the terms of "fast". And if it can't be made smart then make it simple so that you know what the CPU is doing and what it's capable of doing.
Quoting: minj7nm+ or 7nm-? :)7nm+, as stated. It's the accepted term to use.
Quoting: SamsaiI am not an expert but from what I gathered the issue is that after code has been speculatively executed the result is stored in the cache even if the CPU guessed wrong and ran the wrong code, not to mention code that it didn't even have privileges to run.
AFAICT, it's not stored in the cache, but by accessing some memory, this is loaded into the cache, and by checking afterwards how fast you can access the same meory (memory you're allowed to read), you can tunnel data through from the illegitimate code to legitimate code. (Nice technique that is...) Even if you would flush all cache touched by speculatively executed code, you probably could do it the other way around: Prefetch into cache, make speculative execution, and if the data vanishes from cache, you've got information from the dark side.
Last edited by Eike on 8 January 2018 at 12:11 pm UTC
the day when these NDA's that Intel forced on everyone are to
expire, AMD announces a new CPU architecture that is immune.
Last edited by sr_ls_boy on 8 January 2018 at 12:24 pm UTC
Quoting: EikeAFAICT, it's not stored in the cache, but by accessing some memory, this is loaded into the cache, and by checking afterwards how fast you can access the same meory (memory you're allowed to read), you can tunnel data through from the illegitimate code to legitimate code. (Nice technique that is...) Even if you would flush all cache touched by speculatively executed code, you probably could do it the other way around: Prefetch into cache, make speculative execution, and if the data vanishes from cache, you've got information from the dark side.Obviously cache was a mistake.
Patreon
PayPal
See more from me