Richland ES now out in the wild.

[ShintaiDK]

Also 2014 fits better with DDR4, AMD is not in a position to roll a server/workstation DDR3 version and then a DDR4 version shortly thereafter.

DDR4 will be very low volume in 2014. Even Intel first bets on DDR4 for mainstream with Skylake. Haswell-E/EP/EX gonna push DDR4 from the serverspace first.

 

[ShintaiDK]

Well, obviously it's half baked if there are no drivers or bios updates. But, I've always thought that AMD needed to keep moving some x86 hardware if they have any hope of sustaining enough cash flow to develop ARM SoCs. For job postings and AMD's own info they are looking to produce pretty much a standard ARM cpu for 2014 and at least a semi-custom SoC by ~2017. If that the case they need to be moving some x86 product till 2017, assuming they get some serious design wins.

Of course by 2017 an ARMv8 semi-custom SoC will be a very late arrival. So I have no idea WTH AMD is thinking, they are making a huge bet on the 64b ARM server market being large enough and yet new enough for them to get a significant marketshare. At this point, a VIA like fate would be a good thing - the market cap on VIA is ~9.7B!

Yeah, unfortunately you are making more sense than my own theory, which doesn't mean good things for AMD



I was trying to be desperately hopeful but it just doesn't add up as mrmt pointed out, and others, that there is zero chance 22nm samples would be out and about right now...so much more likely this guy who thinks he has a Kaveri probably just has a Kabini. (which would still be cool, but not as cool as the prospects of him having a steamroller ES)

Well assuming the guy just aint flat out lying. Then drivers and BIOS updates could be anything from a single stepping to something bigger.

 

[mrmt]

Even if it is 20% faster than Trinity, AMD would be better off not bragging about it so that they can keep shipping Trinity. When it get's closer to release, they can start bragging an dropping the prices on Trinity.

You are thinking as a consumer, not as an OEM, which are the great target for Trinity and Kaveri.

An OEM must have ready drivers, support, and the supply chain to prepare a notebook or desktop, this is something that you don't do in less than 12 months, sometimes almost twice this time frame if you need specific features, so if AMD had was supposed to launch Kaveri in a 12-15 months time frame they would be telling OEMs things about performance, thermal specs and what not.

AMD isn't strong enough to dictate how much others will pay for its chips. Their price is more dependent on how well the given SKU fares against the corresponding Intel chip, so again, if anything, AMD had something in mind they would be leaking slides and roadshowing Kaveri/Steamroller to world + dog.

The fact that they are mum means that either they have nothing on the pipeline beyond what is already known or whatever they have isn't a big deal.

 

[Ajay]

Also 2014 fits better with DDR4, AMD is not in a position to roll a server/workstation DDR3 version and then a DDR4 version shortly thereafter.

Right, they don't have the cash flow to support CPU development on both 28nm and 20nm, so it only makes sense to wait on 20nm. Then AMD can hope for some positive press due to a larger performance increase than Intel has been getting lately. Of course, they'd still be slower than Intel in ST, but AMD could be more competitive in MT. I'm sure that a 20nm CPU part would have at 5-6 modules or possibly 24 cores w/MCM for servers. This depend on AMD having the cash to do BEOL on a 20nm part a year from now. With all they are putting into ARM, there may not be enough cash for this (freaking idiots!) unless the economy in the US and EUR improve by then.

 

[wlee15]

The next APU will be a Piledriver refresh, unlikely that someone outside from AMD get a Steamroller based APU this year. I guess he didn't realize AMD delayed Kaveri into 2014, so he believes he get one while in reality it's just a Piledriver based APU.

The only real information that Kaveri is delay is that leaked slide that came out a few weeks ago. Certainly no more reliable than on of the main x264 developers receiving an ES chip.

But, I've always thought that AMD needed to keep moving some x86 hardware if they have any hope of sustaining enough cash flow to develop ARM SoCs. For job postings and AMD's own info they are looking to produce pretty much a standard ARM cpu for 2014 and at least a semi-custom SoC by ~2017.

The ARM cpus that AMD is producing is destined for their SeaMicro servers which basically means all it needs are CPU, memory controller, and the SeaMicro/Freedom Fabric controller. It's basically an anti-SOC.

 

[Ajay]

You are thinking as a consumer, not as an OEM, which are the great target for Trinity and Kaveri.

AMD isn't strong enough to dictate how much others will pay for its chips. Their price is more dependent on how well the given SKU fares against the corresponding Intel chip, so again, if anything, AMD had something in mind they would be leaking slides and roadshowing Kaveri/Steamroller to world + dog.

The fact that they are mum means that either they have nothing on the pipeline beyond what is already known or whatever they have isn't a big deal.

Good point. Though, they could running roadshows on Kaveri without any leaks yet. Bottom line, there's no confirmation yet, so this one leak adds up to zilch.

The ARM cpus that AMD is producing is destined for their SeaMicro servers which basically means all it needs are CPU, memory controller, and the SeaMicro/Freedom Fabric controller. It's basically an anti-SOC.

Yes, but they are also looking for an Technical Fellow for wireless, it's highly unlikely that that position is for servers. Looks like AMD has wider ambitions and that would make sense so that they can get their ARM production volume up to get better margins (they can't sell an ARM CPU for the same price as an x86 Opteron).

 

[mikk]

Trinity is PD. There have been rumors that Richland is also PD + CGN.

Richland is the next APU which is based on Piledriver cores and VLIW4 graphics.

It's also really a strange thing that someone under NDA can share the infos for the public.

 

[Idontcare]

The ARM cpus that AMD is producing is destined for their SeaMicro servers which basically means all it needs are CPU, memory controller, and the SeaMicro/Freedom Fabric controller. It's basically an anti-SOC.

Huh, so basically AMD has everything that Calxeda needs/could use (memory controller and the SeaMicro/Freedom Fabric controller), and Calxeda has exactly what AMD needs/could use (64bit custom flavor ARM microarchitecture).

Wonder if they will do a co-develop of some sorts. If they don't then basically they are destined to be direct competitors as each are gunning for exactly the same customer.

 

[mrmt]

Huh, so basically AMD has everything that Calxeda needs/could use (memory controller and the SeaMicro/Freedom Fabric controller)

AMD has everything that Calxeda needs except for one very important thing: money.

AMD and Calxeda are very small players when you compare them with Qualcomm, Intel and Nvidia. They need to burn a lot of money now in order to stay ahead of the curve or at least on equal footing with their larger competitors.

It is easier to pour money on Calxeda as they have a very lean cost structure, the same cannot be said about AMD, which is so screwed that right now would have negative valuation if we were to use their future cash flows only.

So even if you were to set aside those differences and go for something like a joint-venture, effectively segregating any risks contained in AMD balance sheet, who would like to partner with a company which the very survival is in question?

 

[mrmt]

Good point. Though, they could running roadshows on Kaveri without any leaks yet. Bottom line, there's no confirmation yet, so this one leak adds up to zilch

If this rumor is true I don't think there is any chance to be Kaveri, it has to be Steamroller for servers/workstations. AMD is stated to release Richland around Q213, it wouldn't make sense to release Richland in Q213 to pull it off in Q413 for Kaveri launch. OTOH a Steamroller ES would suit a rational roadmap fine, as there is a space of four quarters between Vishera launch and a hypothetical
Q413 Steamroller launch.

But for that you would have to assume a 28nm Steamroller, which would wouldn't be any big deal, which would explain their silence.

 

[Ajay]

If this rumor is true I don't think there is any chance to be Kaveri, it has to be Steamroller for servers/workstations. AMD is stated to release Richland around Q213, it wouldn't make sense to release Richland in Q213 to pull it off in Q413 for Kaveri launch. OTOH a Steamroller ES would suit a rational roadmap fine, as there is a space of four quarters between Vishera launch and a hypothetical
Q413 Steamroller launch.

But for that you would have to assume a 28nm Steamroller, which would wouldn't be any big deal, which would explain their silence.

Good points. I don't understand the rational about Richland, especially if, as one poster mentioned, it is PD + VLIW4. Not a big deal, unless AMD reduces the TDP significantly for the notebook market.

Looking around the web, it is far more likely to be a Kabini ES or Richland ES as you point out. Some folks on the Semi-Accurate forums think the XBOX 720 will have 4 custom Kabini's instead of Kaveri, curious choice if true, but that's not entirely unusual for MS.

With Abu Dhabi having just been released, I still think it makes sense to wait for 20nm for SR. AMD could add modules, keep decent clocks and drop TDP, which is what they need to do, IMO. So if AMD has the $$s for an FEOL shrink and BEOL on 20nm it would be worth having a 6-9 month gap in their product lineup. This would work out well, if and only if GF delivers 20nm SHP on time.

 

[mrmt]

Good points. I don't understand the rational about Richland, especially if, as one poster mentioned, it is PD + VLIW4. Not a big deal, unless AMD reduces the TDP significantly for the notebook market

Regardless of the reasons we think plausible, the fact remains that AMD is delayed on their roadmap and Intel didn't, so AMD needs a quick, low-risk measure to not fall too deep in the value ladder once Haswell arrives.

Low risk and quick means taking something you already have working and tweaking it a bit, which coupled with the benefits of a more mature node should yield better frequency/power consumption and a bit of IPC. Not something game-changing but better than stick with Trinity for more one year.

 

[SickBeast]

Even if Steamroller came out tomorrow, AMD would still be in trouble. Things are really bad for them right now. Even if it's 20% faster, they still need another 60% to catch up to Intel in single threaded performance-per-watt.

 

[Phynaz]

Certainly no more reliable than on of the main x264 developers receiving an ES chip.

Ask yourself this: Why would AMD send a relatively minor open source developer who is not under NDA an unreleased CPU?

Answer: They wouldn't.

 

[wlee15]

Ask yourself this: Why would AMD send a relatively minor open source developer who is not under NDA an unreleased CPU?

Because x264 isn't a minor open source project. It powers youtube and Hulu among other and among other.

 

[Zor Prime]

x264 is a BIG deal.

Even that Dev said he doesn't believe AMD's left hand knows what the right hand is doing. i.e., he was baffled he received the CPU.

 

[Idontcare]

Because x264 isn't a minor open source project. It powers youtube and Hulu among other and among other.

Well that settles it then. This guy definitely has a steamroller ES, it is the only plausible explanation.

Intel has likely already sent him a Haswell ES as well, just for good measure, because it is x264 open source development after all. That's the big leagues and Intel/AMD know it.

There just can't possibly be a more reasonable explanation for the purported claims, I wish some of these naysayers would just come to their senses and recognize.

Sorry JC, your days of being the premier sanctioned leak for ES results are over, there's a new game in town and it wears an x264 developer badge

 

[SickBeast]

Why would he make it up though?

I could see AMD starting to panic and rush out Steamroller. It's plausible. What we really need to see is a completely new design, though. If you look at performance-per-watt, Bulldozer is an order of magnitude slower than competing Intel chips. Adding 20% here and 20% there is not enough. They need a completely new design.

 

[ShintaiDK]

Why would he make it up though?

Do you really need examples? Really?

 

[Idontcare]

Why would he make it up though?

I don't think he is making anything up, I just think he is mistaken as to what exactly he was sent.

I also don't believe JFAMD was making anything up either, I think he too was just honestly mistaken about what he was being told by the engineers.

People make mistakes, but assuming they don't make mistakes is just us making yet another mistake.

We need to stop that cycle, for the sake of the children, will no one think of the children!?

 

[Phynaz]

Because x264 isn't a minor open source project. It powers youtube and Hulu among other and among other.

You're confusing x264 (the open source software project) with h.264 (the video compression standard).

Here's what I consider to be some major open source projects:
Linux
Tomcat
Java
Webkit
Apache
Python

 

[phenomkid7]

3 pages of bullshit....

... no steamroller in 2013. Get your head out ya ass.

 

[wlee15]

You're confusing x264 (the open source software project) with h.264 (the video compression standard).

Nope the h.264/AVC videos on youtube and Hulu are encoded using x264.

Well that settles it then. This guy definitely has a steamroller ES, it is the only plausible explanation.

Intel has likely already sent him a Haswell ES as well, just for good measure, because it is x264 open source development after all. That's the big leagues and Intel/AMD know it.

There just can't possibly be a more reasonable explanation for the purported claims, I wish some of these naysayers would just come to their senses and recognize.

Sorry JC, your days of being the premier sanctioned leak for ES results are over, there's a new game in town and it wears an x264 developer badge

Let's face it everybody here legitimately believed that Kaveri was delayed to 2014(myself included) based on a leaked slide that appears to be authentic and the usual "sources" "confirming" it.(Let's face it AMD's past execution hasn't been any help either) Now that we have evidence that seems to contradict it. I'm not assuming anything at this moment, but I'm willing to follow this thread along.

 

[inf64]

@wlee15

Thanks for this thread,it's very interesting. Also you seem to be the only guy online who has noticed that irc chat conversation, so kudos to you .

 

[CTho9305]

Huh, so basically AMD has everything that Calxeda needs/could use (memory controller and the SeaMicro/Freedom Fabric controller), and Calxeda has exactly what AMD needs/could use (64bit custom flavor ARM microarchitecture).

Wonder if they will do a co-develop of some sorts. If they don't then basically they are destined to be direct competitors as each are gunning for exactly the same customer.

Did you watch the ARM/AMD talk? You might find it interesting if there's a recording or transcript (was that the one with Rory and the live mic?).