AMD Talks Up The Carrizo APU At ISSCC

[wizard69]

If there's anything I've learned over the years, it's that people don't like being forcibly transitioned, you need to bait them with shiny things

I work on a factory floor where there are many devices that still use RS232 for communications and frankly other legacy communications ports. So it isn't like I don't understand the frustration here for many. That however doesn't mean that I go around insisting that manufactures put RS232 ports in their laptops or Centronics ports for that matter.

In the case of a chip designed for laptops or portables, I really see no reason at all to have SATA support built into a modern SoC. Any modern laptop worth having these days isn't using rotating media and PCI Express is perfectly fine for solid state. In the end all the SATA ports do is waste die space and chew up pin count. In effect if you are building SATA into a mobile chip these days you have your priorities screwed up, it makes as much sense as building in the hardware for RS232 or Centronics ports. To put it in other words you have a chip built for the past not the future.

I actually feel sad for AMD as they seem to be dwelling on the past even has they break new ground with HSA and similar tech. Carrizo is apparently targeting the mobile world of a decade ago, it certainly isn't looking at what the bulk of the laptop and mobile market will be from this point on.

[wizard69]

I'm a big AMD fanboi, but this is ridiculous. A 28nm chip coming in a few months? Intel is at 14 nm. Half. So naturally, AMD focuses on laptop chips, which won't find their way into any laptops except the very cheapest, and even then the power consumption is going to be worse than Intel.

I don't get the 28 nm move either though I do have a theory. That theory says that AMD and Global decided to skip 22nm for a 14 nm process. So the next iteration here may very well be a 14 nm chip. Wishful thinking? Possibly but if 22 nm didn't offer AMD much it might be worth the risk.

As for power consumption there is more to the story than just the feature size. We won't know anything really until AMD releases. I don't hold out hope though as I would have expected more information if they had made dramatic progress. The other thing is this may not be the same 28 nm process of the last couple of years.

AMD is going in this weird direction where all new chips are completely useless. The Opteron X2150 was a bad joke, so was the A1100. Did they even sell a single one? Because servers REALLY need a gpu in them.

GPUs can be very useful in some types of server applications. As for the A1100 it would actually be interesting to see if that was designed in anywhere. I didn't like the focus of that design, I would have rather seen the A1100 come out as a general purpose low power processor.

Okay, now we've screwed up the server market, what do we do? Let's go all out in the mobile sector, where there's no way we can even approach Intel's power efficiency.

Well again we don't know what the power profile is here. However AMD can do pretty good at times, especially when the GPU is a significant contributor to power usage. It is just too early to declare Carrizo a complete loss.

People really looking for low power are going towards ARM anyway. You can build a nice HTPC with an O-Droid or even rraspberry pi2. For 35-40$.

Yep! This is why I never understood the A1100. AMD could have launched a nice family of ARM based chips that could have covered everything from embedded usage to tablets.

[Ansla]

Any modern laptop worth having these days isn't using rotating media and PCI Express is perfectly fine for solid state.

Just looked at a local online store laptop offer: 148 models with SSD and 624 models with HDD/hybrid. Solid state may be the future, but that doesn't mean you can ignore the present.

[bridgman]

Carrizo is apparently targeting the mobile world of a decade ago, it certainly isn't looking at what the bulk of the laptop and mobile market will be from this point on.

OK, I have to ask. What are you basing this on ?

[asdfblah]

Who, ever, upgrades the CPU in their laptop? Laptop CPUs are expensive anyways... Also, for cheap SoCs: why would you want to replace your CPU, if it's so integrated already? Why not simply buy another small motherboard that contains: CPU/GPU/NB and SB (APU/SoC), SATA + USB + network + display ports and sockets for RAM modules...
I just don't get why people would buy cheap, upgradable machines, instead of simply buying a completely new system. It's not like office workers, mobile devices, embedded devices, ..., really need socketed CPUs/APUs/SoCs, as the ARM world has shown.
Once again, AMD should sell fully integrated, cheap boards, by themselves, to defeat Intel and ARM competitors in their own game. AMD also has these advantages:
- they have really good integrated GPUs with open source drivers,
- there are LOTS of x86 (windows/linux/iOS/android/...) software,
- and, Android devs could easily port their apps to x86, if their apps don't already work in x86.
If they do this already, it's because they are short-sighted.
But that's my opinion, I barely have any real-life knowledge about most of this stuff.

[duby229]

Who, ever, upgrades the CPU in their laptop? Laptop CPUs are expensive anyways... Also, for cheap SoCs: why would you want to replace your CPU, if it's so integrated already? Why not simply buy another small motherboard that contains: CPU/GPU/NB and SB (APU/SoC), SATA + USB + network + display ports and sockets for RAM modules...
I just don't get why people would buy cheap, upgradable machines, instead of simply buying a completely new system. It's not like office workers, mobile devices, embedded devices, ..., really need socketed CPUs/APUs/SoCs, as the ARM world has shown.
Once again, AMD should sell fully integrated, cheap boards, by themselves, to defeat Intel and ARM competitors in their own game. AMD also has these advantages:
- they have really good integrated GPUs with open source drivers,
- there are LOTS of x86 (windows/linux/iOS/android/...) software,
- and, Android devs could easily port their apps to x86, if their apps don't already work in x86.
If they do this already, it's because they are short-sighted.
But that's my opinion, I barely have any real-life knowledge about most of this stuff.

I'd say you're pretty close to being right on the spot.

10 years ago I was able to sell 500$ CPU upgrades. But, now l rarely sell a CPU all by itself, and they are usually like 50$. The only time it really happens is when I'm replacing a chip I suspect is going bad. But, even then I try to sell a new computer. I do still sell a lot of RAM upgrades and HDD/SSD upgrades, (SSD upgrade is easy to sell)

I make most of my money doing PC cleanups, but upgrades are an extra revenue thats easy to tap for the most part. Making parts unreplaceable, or unupgradeable would really hurt me in some ways.

[curaga]

Actually I'd like to see AMD push us past the old ways of thinking and ignore the socketed chip market for its APUs. Expecting a socket is living in the past in my estimation. Hey I'd also like to see AMD drop legacy port support, for example SATA to minimize pins on these chips.

I have to ask, what about all the users who need those? "Just buy the competitor's product"? (also, SATA, legacy. Chuckle.)

...forcibly transition people off legacy hardware...

I'm speechless.

[dungeon]

Wizard likely also wants old software to not run on that new platform and to not support old people or new peoples with old ideas only new ideas allowed and prohibit any old ones... I wish i am younger to think the same

Wondering if Carrizo will support Windows XP

[haplo602]

I really don't see much of a future for sockets for these types of processors. When you go looking for a socketed chip you are basically looking backwards in time a decade or more for what made sense back then. With these highly integrated chips, and the types of platforms they support, you are better off not having any sockets at all.

A PCI Bridge to what? You need to remember most of the system is already on the silicon, what would you be hooking up to PCI Express?

The trick to managing pin counts is to delete as much as possible legacy port wise. That means dumping the old USB ports, SATA and whatever else is sitting on the chip that is legacy. This leaves you plenty of room for PCI Express lanes. In this regard I'm imagining a chip with a bunch of USB 3 ports and a bunch of PCI Express ports to support all user I/O needs.

One thing this would do for us is to forcibly transition people off legacy hardware to things like PCI Express based storage. A rational collection of modern I/O ports is forward looking, instead of living in the past.

IDK ... sound card ? DVB-T/S tunner (HTPC) ? Raid controller (small NAS) ? FireWire card ?

You know there are things where USB is still not the answer ....

[rdnetto]

I'm a big AMD fanboi, but this is ridiculous. A 28nm chip coming in a few months? Intel is at 14 nm. Half. So naturally, AMD focuses on laptop chips, which won't find their way into any laptops except the very cheapest, and even then the power consumption is going to be worse than Intel.

AMD is going in this weird direction where all new chips are completely useless. The Opteron X2150 was a bad joke, so was the A1100. Did they even sell a single one? Because servers REALLY need a gpu in them.

Okay, now we've screwed up the server market, what do we do? Let's go all out in the mobile sector, where there's no way we can even approach Intel's power efficiency.

People really looking for low power are going towards ARM anyway. You can build a nice HTPC with an O-Droid or even rraspberry pi2. For 35-40$.

AMD simply can't compete on process when they sold off their fabs and Intel didn't.

They've also screwed their desktop segment as well - the FX-8350 is 3 years old now, and might as well be their flagship (the FX-8370 is just a nicer binning). I've got a Phenom II x6, but I'm looking at getting a Xeon E3 for my next upgrade, unless AMD releases something better in the next half year.