AMD expects further APU, CPU declines, cuts chip orders to GF

[ShintaiDK]

Zen and K12 was targetted at dense servers.

Hopefully they can turn the designs more towards semicustom. That will be their only chance. As I said before, going after servers is the nemesis of AMD. SeaMicro was bought 3 years ago. 334M$ lost plus whatever they lost during those 3 years. All in the constant pursue of something out of their league.

 

[mrmt]

I think it's very cute how some people just assume Zen will be released on time and that it will offer superior performance to Intel CPU's.

Especially when every material AMD released to their investors say it will be a low power core.

 

[AtenRa]

Especially when every material AMD released to their investors say it will be a low power core.

Like Broadwell XEON-D ??? you know, the Broadwell High-Performance big core that is used for Dense low power servers

 

[Fjodor2001]

http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MjYxMjY0fENoaWxkSUQ9LTF8VHlwZT0z&t=1

Check slide 9:

Ok, so it was BS then as expected. Nowhere on that slide does it say Zen is only targeted at the Microserver market.

 

[mrmt]

Ok, so it was BS then as expected. Nowhere on that slide does it say Zen is only targeted at the Microserver market.

No, it doesn't. AMD probably knows their investors aren't dumb enough to think that their company would be designing a desktop power hog after reading IR material saying that they are building a product tailored for power constrained environments.

 

[Fjodor2001]

No, it doesn't. AMD probably knows their investors aren't dumb enough to think that their company would be designing a desktop power hog after reading IR material saying that they are building a product tailored for power constrained environments.

So you're confirming your statement was BS and that you have no source backing up your claims. As expected then.

 

[mrmt]

So you're confirming your statement was BS and that you have no source backing up your claims. As expected then.

No, I'm telling you to read the slide.

 

[Fjodor2001]

No, I'm telling you to read the slide.

I did, and arrived are the conclusion previously mentioned. So far you've not been able to prove your claims.

 

[mrmt]

I did, and arrived are the conclusion previously mentioned. So far you've not been able to prove your claims.

I think you are being delusional, expecting something that not only isn't written in AMD slides, it goes against what they bothered to write. In any case, in three weeks AMD should release some new info regarding their product roadmap, I will post something on this thread then.

 

[Fjodor2001]

I think you are being delusional, expecting something that not only isn't written in AMD slides, it goes against what they bothered to write. In any case, in three weeks AMD should release some new info regarding their product roadmap, I will post something on this thread then.

So far you've not been able to produce any concrete evidence and source backing up your claims that Zen is designed only with the Microserver market in mind.

Do you have any? If so link to that and clarify. Otherwise please give up the BS.

 

[mrmt]

So far you've not been able to produce any concrete evidence and source backing up your claims that Zen is designed only with the Microserver market in mind.

Do you have any? If so link to that and clarify. Otherwise please give up the BS.

Why am I supposed to provide evidence against something AMD never stated as the scope of their product? It shouldn't be you providing evidence that AMD is actually working on a high end desktop chip?

 

[cbn]

With Zen + updated GCN + HBM on a 16nm FinFET process, AMD should be able to create all-in-one chips that provide better all-around performance than Intel's competing products. The CPU performance (at least per-core) will not be quite as top-notch, but the GPU performance will be better by a mile - Intel just has no decent GPU tech, even Iris Pro can't come close to a $99 discrete card. For most users, who almost never max out their CPU, it will be a much better choice. The problem with existing AMD APUs is that the CPU portion is too weak and inefficient because it's Bulldozer-based, and the GPU portion is starved of memory bandwidth because it relies on slow DDR3. Zen should fix the former problem and adding HBM should fix the latter. In addition to stand-alone sales, a product like this has the potential to get next-generation console design wins, and maybe even purchases from Apple as well.

I just hope AMD's product binning improves next round.

Sempron 2650 @ 1.45 Ghz? Was there really a point to doing that?

And do all the dual core APUs (based on construction cores) really need to have so much of the iGPU disabled? (Eg, A4-4000, A4-5300, A4-6300 only had 1/3 of the stream processors enabled. A4-7300 improved things, but it is still at only 50% active stream processors and based on the Richland when Kaveri is the current generation)

 

[DrMrLordX]

They're dying a deadly death.

It is bettar thaen DYING a DEATH from JEFF K!!! becauz he HIT YOU with a BAG OF HELL!!!!

Aaaaand back on topic.

Slides aside, didn't we go through this "lulz Zen is 4 low powar" thing already? AMD announced a 200-300W APU based on Zen at the PC Cluster Consortium conference this year. The first source to mention that data on public forums (a poster at Semiaccurate; turns out he was right about the APU) also dropped another bomb, mentioning 32C/64T Zen AND K12 chips. The bit-tech article linked above doesn't mention that, but still . . . how is a ~250W APU aimed at dense server?

AMD has made all kinds of moves lately that contradict some of their slides, even recent ones. Remember the slides and announcements referencing FM2+ Carrizo?

 

[erunion]

Why am I supposed to provide evidence against something AMD never stated as the scope of their product? It shouldn't be you providing evidence that AMD is actually working on a high end desktop chip?

He does that in every thread. He thinks the burden of proof lies on everyone but himself.

 

[cbn]

32C/64T Zen AND K12 chips.

That *could be* pretty awesome (emphasis on could).

If true, then I wonder how well they will do for heavy video editing (in addition to the obvious server application)?

Of course, I also wonder they will price it and how much tax for enabling 2P on certain chips?

 

[cbn]

On this rumored 32/64T chip, I do wonder if it will use the same socket as the rumored 16T/32T APU?

If so, how low end will the processors go?

In an ideal situation a person could buy into a more powerful platform with a relatively inexpensive processor and then upgrade over time. Hopefully if any of these rumors are true AMD has made the necessary provisions for that to happen.

 

[mrmt]

On this rumored 32/64T chip, I do wonder if it will use the same socket as the rumored 16T/32T APU?

If so, how low end will the processors go?

In an ideal situation a person could buy into a more powerful platform with a relatively inexpensive processor and then upgrade over time. Hopefully if any of these rumors are true AMD has made the necessary provisions for that to happen.

Which software can make use of 32T/64T on the consumer market?

 

[podspi]

32C/64T sounds like dense server to me.

Will be interesting to see the clockspeed range on these new chips.

As for the super-high TDP chips, if they are performance competitive and cheaper they'll do well enough in the enthusiast market - they'll still need to compete competently in lower TDP levels if they want any OEM uptake.

Seriously though, low-power cores make the most sense. Have you guys actually bothered to look at what people are buying? Even laptops mostly have CPUs/APUs in the 5W~35W range. That's the sweet spot (unless gunning for mobile.)

 

[myocardia]

Which software can make use of 32T/64T on the consumer market?

Handbrake, MakeMKV, Sony Vegas, and most other video editing software.

 

[cbn]

Handbrake, MakeMKV, Sony Vegas, and most other video editing software.

Yep, and while GPUs can help they can reach the point of diminishing returns fairly quickly in Premiere Pro CS6 and Premiere Pro CC:

https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CS6-GPU-Acceleration-162/ (article dated September 2012)

https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CC-Professional-GPU-Acceleration-502/ (article dated September 2013)

 

[cbn]

32C/64T sounds like dense server to me.

Will be interesting to see the clockspeed range on these new chips.

The rumor we had on Summit Ridge (8C/16T Zen) was that it was 95 watts.

So this is why I am wondering if this rumored 32C/64T Zen would use the same socket as the 300 watt APUs.

And if so, will AMD clock the CPU up to make full use of the 300 watt capability?

 

[Fjodor2001]

He does that in every thread. He thinks the burden of proof lies on everyone but himself.

It was mrmt that made the claim that "Zen is only targeted at Mircoservers". Don't you think the burden of evidence to back that claim up is upon him then?!

 

[DrMrLordX]

32C/64T sounds like dense server to me.

Will be interesting to see the clockspeed range on these new chips.

As for the super-high TDP chips, if they are performance competitive and cheaper they'll do well enough in the enthusiast market - they'll still need to compete competently in lower TDP levels if they want any OEM uptake.

Seriously though, low-power cores make the most sense. Have you guys actually bothered to look at what people are buying? Even laptops mostly have CPUs/APUs in the 5W~35W range. That's the sweet spot (unless gunning for mobile.)

32C/64T could be a lot of things, dependent on clockspeed, as you indicated. At high clocks, you're looking at a head-on competitor to something like POWER8 or some of those 18C Xeon monsters. At low clocks, maybe you've got something that is for dense servers, though I still get the impression that most dense-server clusters are going to be based on 4-8C chips at low clockspeeds in low-power format (look at Broadwell-D). A 32C/64T chip is basically four times (or more) the CPU than what we're seeing in most dense server appliances today.

As for the 35W-and-under category, AMD is targeting that right now with Carrizo. How Zen factors into that equation is anyone's guess.

 

[erunion]

It was mrmt that made the claim that "Zen is only targeted at Mircoservers". Don't you think the burden of evidence to back that claim up is upon him then?!

He did. You refused to accept his evidence as sufficient, without bothering to add any of your own.

You've been given sufficient evidence to show that AMD has said Zen is targeted at microservers(its right on the slide). You've asserted that Zen may be targeted at more things than what AMD has chosen to list. While that's certainly a possibility, it begs the question: "Why would AMD list some target markets, but not others"?
Thus the burden of proof lies on you to show why.

 

[Lonyo]

Since demand for PS4 and XB1 system-on-chips is expected to increase this year, this means that the share of such semi-custom APUs in AMD’s purchases from GlobalFoundries will rise as well. However, it also means that the share of APUs, CPUs and GPUs for personal computers will decline in AMD’s purchases from its main partner. Even in the best-case scenario, if everything goes well for AMD this year, the company does not expect sales of its microprocessors and accelerated processing units for PCs to exceed sales of such chips in 2014.

Over time, the intention for consoles is for the cost of components to decrease. AMD is looking to die shrink the APU (well, Sony and MS are looking for them to do that), to reduce costs.
Reduce costs for Sony/MS means AMD would also expect reduced costs.

If you make 10 million APUs for consoles on 28nm and they cost $50 each, that's $500m in orders.
If you now die shrink and make 12 million APUs on 20nm and they cost $35 each, that's $420m in orders, a decrease in the dollar amount, despite an increase in the number of chips being made.

(The numbers above are all made up but are for illustrative purposes).

Basically the logic of "Console APU numbers increasing means increasing dollar orders at GF" isn't accurate. Not to say it can't be true, but it doesn't HAVE to be true, especially given the expected die shrink from 28nm to 20nm some time in 2015.