Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

AMD talks about next year's CPUs

QuizzicalQuizzical Member LegendaryPosts: 22,135


They have a roadmap of the mobile CPUs they expect to launch in 2015.  There are two:  Carrizo and Carrizo-L.  In spite of the very similar names, they're different CPU architectures entirely.

Carrizo is the successor to Kaveri.  It will have 2-4 Excavator cores, the successor to Kaveri's Steamroller cores.  They're still derivatives of Bulldozer, and I wouldn't expect much in the way of performance increases in desktops.  They're also on the same process node as Kaveri.

There are two advances that AMD is hyping in Carrizo.  One is vastly better energy efficiency in "typical-use"--roughly double that of Kaveri, which by AMD's own claims, was nearly double that of Richland.  On the surface, that may sound impressive, but look at how they claim that Kaveri offers about ten times the energy efficiency of 2009's Tigris.  Energy efficiency under heavy loads is better, yes, but not ten times better.  Not even close.

But you know what might genuinely be ten times better?  Idle power consumption.  Most computers are nearly idle most of the time.  If that's what AMD means, then Carrizo could be to Kaveri as Haswell is to Ivy Bridge:  a little better all around, but not a lot--with the exception of vastly improved idle power consumption.

AMD is also offering Carrizo-L, the successor to Beema, with Puma cores.  Those are AMD's relatively low power, low performance cores that have become popular in cheap laptops.  But also interesting is that AMD is pointedly not offering a successor to Mullins, the lower clocked, lower power version of exactly the same chip as Beema.  And to really throw a wrench in the situation, Carrizo-L uses the same socket as Carrizo.

It would be astonishing if Carrizo relies on only a single memory channel, so the socket probably accommodates two, as all relatively high-performance laptop chips have for many years.  And that could easily mean that Carrizo-L also has two memory channels.  A single channel was already a meaningful restriction for Beema, and doubling the memory bandwidth could help Carrizo-L a lot.  But the doubled bandwidth would be useless and even counterproductive for Mullins, where small fractions of a watt are critically important.

AMD also makes no mention of DDR4, so the chips likely don't use it.  Which is just as well if you check prices.  8 GB of 1600 MHz DDR3 can readily be had for $70 or less.  Make it 8 GB of DDR4 and new kit prices on New Egg start at $110 and go up from there.

AMD is promising GCN graphics on Carrizo-L, but didn't say how many compute units.  It might be two, to match Beema, though if they're increasing memory bandwidth, it could plausibly have three or four.  In contrast, they promise "next generation GCN graphics" in Carrizo.  Which is odd, considering that GCN stands for Graphics Core Next, so if you unroll the acronym, they're promising "Next Generation Graphics Core Next Graphics Compute Units".

This is supposedly going to be a new graphics architecture entirely.  AMD is claiming to offer lossless texture compression to reduce memory bandwidth needs.  That sounds an awful lot like what Nvidia did with Maxwell to allow a chip with only a 256-bit GDDR5 memory bus to handily beat the previous generation chips with 384- and 512-bit busses.  That could be a huge deal in higher end integrated graphics chips, which are very much starved for memory bandwidth.  If it works right, which it might not.

Conspicuous for its absence is the complete lack of 20-nm chips.  It wasn't expected for Excavator cores to move to 20 nm, as there wasn't going to be a high performance process node for them any time soon.  But AMD had been promising a Beema/Mullins successor on 20 nm next year.  That now looks delayed or canceled--and it could easily be the latter, if AMD is planning on jumping directly to 14 or 16 nm in 2016.  I'm expecting for Carrizo-L to be nearly the same CPU as in Beema, and probably nearly the same GPU, too, unless AMD decided to add more GCN compute units.

As for timing, AMD expects both Carrizo and Carrizo-L to release in the first half of 2015.  That's a pretty broad time window, of course.  That these are the only new mobile CPUs AMD plans to launch in 2015 means that their upcoming Zen architecture which will not be a Bulldozer derivative and might finally be competitive with Intel again won't arrive until 2016 at the earliest.  Realistically, I've been expecting that to be a 2016 chip ever since Kaveri officially slipped into 2014.

This also says a lot about AMD's upcoming desktop CPUs, as no one makes CPU chips that are primarily targeted at desktops anymore.  Carrizo, like Kaveri and Richland before it, is primarily a laptop chip, even if it will also be offered in desktops.  I'd expect Carrizo to be slightly better as a desktop CPU than Kaveri, for the simple reason that if it were going to be worse, AMD wouldn't bother.  The integrated graphics in it may or may not be a major advance.

AMD could also launch a desktop version of chips that were primarily intended for servers.  See, for example, their current FX line.  Those will tend to not have a mobile version in the first place.  There have been rumors that AMD would launch a server chip with 16 Kaveri cores; such a chip makes a lot of sense in a server, but not so much in a desktop, as the cores would necessarily be clocked lower.  Whether such a chip would have a desktop version at all is unclear; even if AMD were planning on launching such a desktop chip in 2015, it wouldn't have been announced just now.

In other news, AMD announced that Samsung will launch some FreeSync-compatible monitors in March 2015.  Other monitor vendors will presumably do likewise, but aren't ready to announce it yet.  So the FreeSync monitors are coming; AMD has said early 2015 as the target date on that for quite some time.


  • Heihachi07Heihachi07 Member Posts: 1

    So back in the day I used to do radio/tv and vcr repairs and back then we used simple weller stations with temp control. Worked great and never had much trouble tinning the tips or having tips go to crap on us.


    We are the leading the world in providing best learn danish dumps how to learn dutch language prep solutions. Our incredible offers for hawaii questions.

Sign In or Register to comment.