AMD's Carrizo to deliver large performance-per-watt leap

[HEXUS]

The largest leap ever for AMD's APUs according to Dr Lisa Su.

Read more.

[3dcandy]

This and 14nm might make AMD a better bet later this year then...

[watercooled]

Wasn't SHP originally the name for the cancelled FDSOI node? Plenty of websites still seem to be claiming Kaveri is on SHP so maybe it has a different meaning now, but AFAIK it's all but certain it's bulk CMOS.

However I also thought Carrizo was using a different process to Kaveri, the Kaveri node being a high-power one largely developed for AMD whereas the one used for Carrizo is more mobile-oriented.

[bae85]

Carrizo (or equivalent) in AM3+ form would be lovely AMD.

[ZaO]

It's nice to see Amd making these announcements on new stuff lately But, they still haven't made the one that many of us are waiting to hear. I mean, they haven't been specific enough anyway.. I have a feeling we'll be hearing something within the next few months!

[shaithis]

I have a feeling we'll be hearing something within the next few months!

Yup, that Intel have pulled further ahead in the desktop segment probably

[ZaO]

Yup, that Intel have pulled further ahead in the desktop segment probably

haha :/ I don't care if Intel are still ahead. That'll probably remain the case for a while. But, Amd is really overdue in that area now..

[shaithis]

haha :/ I don't care if Intel are still ahead. That'll probably remain the case for a while. But, Amd is really overdue in that area now..

That's what worries me, it's been so long since they really tried to compete that I am at the point of believing they aren't going to. Wish they would just make a statement and let us know either way.

[ZaO]

That's what worries me, it's been so long since they really tried to compete that I am at the point of believing they aren't going to. Wish they would just make a statement and let us know either way.

I feel that. It has worried me a little. But, these new announcements have given me some hope. If we don't at least get some kind of announcement by late Spring, then I might just lose hope tbh..

[MLSCrow]

That's what worries me, it's been so long since they really tried to compete that I am at the point of believing they aren't going to. Wish they would just make a statement and let us know either way.

Carrizo is simply AMD showing the world what it's CMT design was intended to be, which is something different, yet competitive. This is what Bulldozer should have been out of the gate, which, would have blown away the world at the time. However, don't be fooled, this isn't the part that AMD is suggesting will go head-to-head with Intel. Carrizo is meant to be a really attractive part for Notebook vendors (one cheap part, that can hold it's own on the CPU side, not that it'll beat Intel, but it'll do what people need it to do, and a GPU that's good enough to game on, eliminating the need to spend more money on a discrete mobile GPU).

Carrizo is meant to hold us over, until they deliver the knock-out blow that no one saw coming, especially Intel. Although, with all of the hype starting and surrounding Zen, I bet everyone will kinda see it coming. Jim Keller, arguably the worlds greatest chip designer, and the very person who defeated Intel in the past, causing them to frantically dump money into R&D until they came up with something better, has been behind the design of Zen from the beginning, and it's already been leaked that they are going to offer a 95w part, which, at 14nm, and a new design, which is supposed to be incredibly efficient (especially after everything they've learned from ARM and their big.LITTLE design) suggests that they are looking to show up with an enthusiast part (possibly the next FX?) that is going to compete, if not dominate the high-end. It's supposed to take the best of everything Jim has learned about high-clock speeds and high efficiency, from all of AMD's designs, ARM's designs, and even Intel's designs (it's going to have it's own version of Simultaneous MultiThreading!!!), and if he's willing to attempt to compete at the top-end vs Intel, he must have something incredible, because that is an extremely bold move that one must be very confident in, in order to make.

So yeah, my prediction is that we're going to see AMD take the performance crown for the first time in over a decade. Finally.

[MrRockliffe]

I'd like to see this, for competition reasons, but Intel are just outing new generations that are slightly quicker and slightly more efficient, rather than releasing something mind blowing. I imagine they're just waiting for AMD to release something, and then just drop a bomb on them.

Intel have so much money behind them, and with the investment in tablets and phones recently, I can't imagine they've not got something big to release.

AMD were the king of realistically priced processors with their most recent generation of FX chips. It'd be nice to see it happen again.

[MrJim]

It's Intel's lead in process technology & their huge R&D budget which are keeping them so far ahead of AMD. The fact that they're releasing two architectures of enthusiast CPUs within 3 months of each other this year (Broadwell K & Skylake) is testament to that!

[CAT-THE-FIFTH]

AMD does not need to beat Intel but be competitive under £150. For example if the A10 7850K at its original £120 to £130 price point had between Core i3 and Core i5 level CPU performance in both ST and MT benchmarks,I suspect it would have sold much better.

The other way to do it is concentrate on making their dies smaller even if they cannot beat Intel on pure CPU performance,so they can make higher margins than currently.

Edit!!

If you look at Carrizo its around the same size as Kaveri but does away for the need for a southbridge meaning its cheaper overall to implement.

[watercooled]

I'd like to see this, for competition reasons, but Intel are just outing new generations that are slightly quicker and slightly more efficient, rather than releasing something mind blowing. I imagine they're just waiting for AMD to release something, and then just drop a bomb on them.

Intel have so much money behind them, and with the investment in tablets and phones recently, I can't imagine they've not got something big to release.

It's not quite as simple as that. That theory comes up from time to time in CPU/GPU/etc discussions but it ignores so much of the realities of microprocessor development. For instance, as can be see with Zen/K12, AMD have had them in development for quite a long time, probably years before the announcement and while the announcement probably marked some amount of finalisation of the design, it still takes years for it to be in a product on the shelves.

The development pipeline really is that long for processors so instantly responding to a competitor's products with modifications is not feasible - you have to plan many years ahead, predicting where your competitors will be. That's also why blaming AMD for Intel's incremental upgrades is short-sighted, it would be a catastrophic mistake for Intel to have their 5/10 years roadmaps assuming lacklustre competitor performance, as there would be simply no way of reacting if that didn't turn out the be the case. Things like pricing and feature fusing can change, not the processors themselves though. Haswell was always going to be Haswell, Broadwell was always going to be Broadwell.

For Intel to have some fast-response core, it would need to be finished already, and it would be senseless to not release it after the billions sunk into designing it. Not to mention it wouldn't keep improving itself over time so it would need another uArch team to keep working on it. That's a lot of money and effort to spend for no logical reason.

If it was easy to adjust processors as that, I think AMD releases post-K10 may have been different. Bulldozer was what it was and made it to market despite it probably not being what the architects had anticipated in simulations etc earlier on. Or perhaps it was, but they'd expected something more like incremental improvements to Core2 rather than the Nehalem rework. Nothing could be done even up to a few years before release.

The ARM and mobile graphics ecosystem gives us some insight into development cycles. Look how long it takes between core IP release and shipping products, and bear in mind the cores are 'complete' at that point, and AFAIK have reference designs available at fabs for customers to implement into products. It's no lego; there's still a massive amount of work, and years, left before that finished core is available to buy. Same for graphics announcements e.g. Mali, PowerVR, etc.