AMD Carrizo Pre-release thread

[frozentundra123456]

Well, lets see what kind of design wins AMD gets and what the real world performance is. They are definitely right to focus on mobile. I never could figure out why Kaveri came out first on desktop and was so late to mobile. Really, mobile is where APUs make sense.

If Intel 14nm had been on time and performed up to expectations, I would have had little hope for Carizzo, but with the delays and spotty at best performance of Broadwell, AMD could find a nice niche to fill with this chip.

 

[geoxile]

If AMD can achieve this kind of efficiency on 28nm with a fundamentally bad arch, their 14nm cores could be pretty impressive. Let's hope Keller delivers

 

[Enigmoid]

When i write the rest of the uncore this mean that the memory controler is included in this uncore.

I wonder where you are digging to find such arguments while it s obvious
that it has nothing to do with an hypothetic neutralisation, simply DT parts use at best 55nm processes while Beema s PCH is obviously 28nm, even the cut down version used with mobile Kaveris is using 55nm.

That s a blank statement, could you please point us where the slides say so.?..

You said the 45-50% efficiency improvement.

And Beema's PCH is neutered compared to Kaveri.

Fewer PCIe lanes. Less USB or SATA ports. Supports fewer connected displays.

 

[Kuiva maa]

Well, lets see what kind of design wins AMD gets and what the real world performance is. They are definitely right to focus on mobile. I never could figure out why Kaveri came out first on desktop and was so late to mobile. Really, mobile is where APUs make sense.

If Intel 14nm had been on time and performed up to expectations, I would have had little hope for Carizzo, but with the delays and spotty at best performance of Broadwell, AMD could find a nice niche to fill with this chip.

The usual suspects (Lenovo, HP)will definitely offer enough products, there will probably be a few from Acer, Asus (they supposedly have an agreement to collaborate with AMD,they may throw a few carrizo SKUs) and Toshiba and maybe (but that is questionable) Dell will chime in. All in all Carrizo is an APU and I believe it will not succeed ,regardless of performance. Its target group only knows intel or generally buys what is found at the stores ,which again is mostly intel.

 

[Abwx]

You said the 45-50% efficiency improvement.

Yes i said so but you didnt told us where thoses numbers are contradicted in thoses slides.

And Beema's PCH is neutered compared to Kaveri.
Fewer PCIe lanes. Less USB or SATA ports. Supports fewer connected displays.

And there will be fewer ports, display outputs and so on Carrizo compared to Kaveri, not counting that they had enough time to improve this PCH further, indeed ports that are not used can be powered off so i really dont see where would be the difficulty.

 

[Shehriazad]

Yes i said so but you didnt told us where thoses numbers are contradicted in thoses slides.



And there will be fewer ports, display outputs and so on Carrizo compared to Kaveri, not counting that they had enough time to improve this PCH further, indeed ports that are not used can be powered off so i really dont see where would be the difficulty.

Edit: I derped, didn't I? That Kaveri 2.0 slide is like a year old, isn't it? Sheesh...sure looked like it would be something new. xD

 

[Phynaz]

I will tell you this,

If Carrizo at 15W TDP reach 15W TDP Haswell CPU performance but with faster than Broadwell iGPU performance i will call it the best product of 2015.

Also, i would really really like 35-45 or even 55W TDP Desktop FM2+ SKUs.

If AMD's latest and greatest can match what will be a two generation old CPU it will be the best?

That's a pretty low bar to set.

 

[maarten12100]

If AMD's latest and greatest can match what will be a two generation old CPU it will be the best?

That's a pretty low bar to set.

Higher efficiency than Intel's Haswell ULV processors while offering better gpu and multi threaded cpu perf would make it a pretty good product.
Going by those slides AMD won't have any problem hitting that 2x the efficiency of Kaveri.
And if they really brought the idle SoC power down to 1,5W this could provide serious competition.

 

[StereoPixel]

achieves 23% area reduction? if it is true Carrizo has 188mm2 die size

 

[metalliax]

achieves 23% area reduction? if it is true Carrizo has 188mm2 die size

The die size is roughly the same (240-245mm2) however the smaller excavator cores allow for additional transistors elsewhere (PCH, GCN 1.2, UVD/VCE, etc).

 

[StereoPixel]

The die size is roughly the same (240-245mm2) however the smaller excavator cores allow for additional transistors elsewhere (PCH, GCN 1.2, UVD/VCE, etc).

Oh, yes

 

[Enigmoid]

Like with Kaveri (and I was justified) I do not believe the slides.

 

[AtenRa]

Like with Kaveri (and I was justified) I do not believe the slides.

Reason/s ??

 

[Abwx]

Edit: I derped, didn't I? That Kaveri 2.0 slide is like a year old, isn't it? Sheesh...sure looked like it would be something new. xD

It was something new, the numbers on this slide were delivered despite what some are saying out of deseperation, Kaveri is a more advanced product than Haswell i3s overall, put a PC based i3 and another with a Kaveri in any household, the latter will provide better experience and usability, Intel CPU s are offices CPUs and that s all, they do not have the necessary qualities for normal consumers unless thoses only do offices applications, and still, there will be no practical difference with a Kaveri for thoses tasks.

 

[DrMrLordX]

I see that you have edited your post to remove this sentence....

No I haven't. It's still there, and it's still true. Kaveri has superior voltage scaling to the Excavator HDL + AVFS processor described in the slides past a per-module power consumption of about 24-25W.

Can't happen. AM1 doesn't have the pins for dual channel memory (part of the reason it's such a small socket).

A pity then. That's probably a cost-cutting measure.

The die size is roughly the same (240-245mm2) however the smaller excavator cores allow for additional transistors elsewhere (PCH, GCN 1.2, UVD/VCE, etc).

Makes sense. The slides sort of made it out like the die size would be smaller. Oh well! At least they'll save on the overall platform vs. FM2+.

 

[krumme]

The usual suspects (Lenovo, HP)will definitely offer enough products, there will probably be a few from Acer, Asus (they supposedly have an agreement to collaborate with AMD,they may throw a few carrizo SKUs) and Toshiba and maybe (but that is questionable) Dell will chime in. All in all Carrizo is an APU and I believe it will not succeed ,regardless of performance. Its target group only knows intel or generally buys what is found at the stores ,which again is mostly intel.

Even if its efficient as claimed - and we dont know - its Intel deciding if its going to have a market. Both by i3 pricing in the 15 to 25w interval but also by the deals oem get from intel eg offering an entire portfolio perhaps even servers with total "rebate" (many shapes and forms but basically the same economic effect) only if the 15 to 15w products is also sold. Thats the brutal reality. And most likely imo Intel will just kill this little bug if it gets irritating. So lets see - but 15w tdp is where the ulv get a nice profit so its also expensive to kill the apu.

I dont think most consumers for lowmid and low range knows or cares what brand apu it is just that it works for all task. Is light and slim and have good battery life. And foremost cheap.

Still I would not use to much energy to discuss an old failed arch on 28nm - its to long a stretch to make a serious challenge.

 

[krumme]

The benefits of hdl looks great - and for once some solid tech that can make money without some fancy preconditions. Or what ? - and how transferable is hdl to 14nm and future nodes?

 

[Abwx]

No I haven't. It's still there, and it's still true. Kaveri has superior voltage scaling to the Excavator HDL + AVFS processor described in the slides past a per-module power consumption of about 24-25W.

I understand what you mean now, thing is that you wrote that voltage scaling was not as good in a separate paragraph, as such it could has been understood as being a statement that would hold for the whole TDP range.

In your post editing you sticked that statement with the paragaph where you were talking of the TDP over 60-65W, in that case that make sense, indeed the curves are so clear that i was wondering what did lead you to such a conclusion.

That said a socketed 65W, or even 45W, Carrizo would make sense given the APU features, it s not like some people wouldnt prefer a Carrizo over a Kaveri despite the former limited TDP range, not everybody is an overclocker or a numbers cruncher.

AMD should seriously think about releasing either an AM1 version or an updated dual channel such plateform, given the MBs inherent lowish prices they could ask a nice premium for such an APU and still be very competitive for the APU + MB combo.

Even if its efficient as claimed - and we dont know - its Intel deciding if its going to have a market. Both by i3 pricing in the 15 to 25w interval but also by the deals oem get from intel eg offering an entire portfolio perhaps even servers with total "rebate" (many shapes and forms but basically the same economic effect) only if the 15 to 15w products is also sold. Thats the brutal reality. And most likely imo Intel will just kill this little bug if it gets irritating. So lets see - but 15w tdp is where the ulv get a nice profit so its also expensive to kill the apu.
.

You may well be right, Core M is not that revolutionary and it s likely that Intel has quite big inventory given that they stockpiled this production for months.

Although the probability for such products to be contra revenued is low they could still sell them at cost price to make room for their next release in six months or so.

AMD has no other choice than target the segments where Intel makes the bread and butter that allow them to litteraly bribe the OEMs, that s the reason why Zen will be devoted firstly to the server market, in that sense Intel are stupid since they are forcing AMD to shoot for their most profitable segments by negating them a presence in the low costs APUs market, indeed that s not the only decision from Intel that will backfire in the mid term.

 

[krumme]

I understand what you mean now, thing is that you wrote that voltage scaling was not as good in a separate paragraph, as such it could has been understood as being a statement that would hold for the whole TDP range.

In your post editing you sticked that statement with the paragaph where you were talking of the TDP over 60-65W, in that case that make sense, indeed the curves are so clear that i was wondering what did lead you to such a conclusion.

That said a socketed 65W, or even 45W, Carrizo would make sense given the APU features, it s not like some people wouldnt prefer a Carrizo over a Kaveri despite the former limited TDP range, not everybody is an overclocker or a numbers cruncher.

AMD should seriously think about releasing either an AM1 version or an updated dual channel such plateform, given the MBs inherent lowish prices they could ask a nice premium for such an APU and still be very competitive for the APU + MB combo.

Isnt amd in the process of shaving off product and small oem? The above seems like niche in a niche to me. It leads nowhere and when overhead cost is accounted for there is no profit or even a loss.

 

[Abwx]

Isnt amd in the process of shaving off product and small oem? The above seems like niche in a niche to me. It leads nowhere and when overhead cost is accounted for there is no profit or even a loss.

The implementation cost is about zero, all that is needed is to replace the ball grid with pins and that s done, otherwise there is not a single addition to what is mandated by the BGA plateform.

 

[frozentundra123456]

Wow, had to happen. Product isnt even out yet and excuses are starting to pile up, just in case it doesnt do well.

 

[krumme]

The implementation cost is about zero, all that is needed is to replace the ball grid with pins and that s done, otherwise there is not a single addition to what is mandated by the BGA plateform.

Its not done by that - then you just started. Its eg another product to stoc, market, sell, document, support, track and kill.
Its complexity that also takes management time.
Everytime you add a product like that it taxes the organization. Most is more hidden cost.
It needs to be accounted for.

 

[Abwx]

Wow, had to happen. Product isnt even out yet and excuses are starting to pile up, just in case it doesnt do well.

Where are the excuses..?

Or you think that contra revenues have no influence with 46 millions chips?

You know well that Intel can sell their core M at 20$ if they want, so if they do so it would be AMD s fault that a 50$ Carrizo wouldnt sell according to you.

What would you do as an OEM in such case.?.

Its not done by that - then you just started. Its eg another product to stoc, market, sell, document, support, track and kill.
Its complexity that also takes management time.
Everytime you add a product like that it taxes the organization. Most is more hidden cost.
It needs to be accounted for.

Technical support is the same as for BGA, inventory cost has no reason to be more expensive than for other parts and the higher ASP would largely compensate for a cost increased by a few %.

 

[mrmt]

If AMD's latest and greatest can match what will be a two generation old CPU it will be the best?

That's a pretty low bar to set.

A bunch of cherry picked remarks and an underperforming product. I guess we'll soon re-edit that quote of yours about AMD overpromising and underperforming.

 

[monstercameron]

Didn't take long for the nay Sayers to annoyingly butt in, although it isn't unexpected.