AMD Q414 results

[videogames101]

The R&D gap.

In R&D you don't always get what you paid for, but you sure don't going anything you haven't paid for.

Sure, but do you really think that's a very viable approach? I think jacking your funding against Intel in the same markets you've been competing in for years just isn't the way to win in the long run. Anyway maybe you're correct and that's AMD's best chance, but regardless I might imagine a more interesting question. How does AMD design a product to increase profits without having to significantly increase R&D? I can only presume that Keller knows he's not getting more cash, so this is the question he's asking himself and for that reason seemingly a more relevant question.

 

[AtenRa]

The only area AMD is lagging behind is the Process Node Manufacturing, in mArchitectural design both are very close with intel having a small lead in CPU design and AMD a bigger lead in GPU design.

And in order to understand this, a 22nm FF Kaveri would be a much better APU than Haswell both in Mobile/Desktop and perhaps in Server(iGPU-less SKUs).

 

[III-V]

And in order to understand this, a 22nm FF Kaveri would be a much better APU than Haswell both in Mobile/Desktop and perhaps in Server(iGPU-less SKUs).

This is patently false. Returns have diminished tremendously. This has worked much to AMD's favor -- if every development dollar, or every month's worth of man hours spent equaled a 1% performance increase, AMD would have ceased to exist long ago. Performance gains are not linear, though, and it's taken Intel billions to move a measly 10% per generation. AMD is still making decent progress because there was so much still left on the table (although they've eaten quite a good portion of that up).

 

[AtenRa]

This is patently false. Returns have diminished tremendously. This has worked much to AMD's favor -- if every development dollar, or every month's worth of man hours spent equaled a 1% performance increase, AMD would have ceased to exist long ago. Performance gains are not linear, though, and it's taken Intel billions to move a measly 10% per generation. AMD is still making decent progress because there was so much still left on the table (although they've eaten quite a good portion of that up).

What im saying is just porting Kaveri as it is today to Intels 22nm FF and it would be far better product than Haswell, especially in mobile. The design is fine and in some areas better than Intel(MT performance and GPU), they are lagging in node process.

 

[ShintaiDK]

The design isnt fine. Its why AMD is failing.

And we already had this node discussion before. 32nm sandy bridge runs in circles around 32nm Pilerdivers. And no, I dont care about GPU performance for you to cherry pick on.

 

[mrmt]

How does AMD design a product to increase profits without having to significantly increase R&D? I can only presume that Keller knows he's not getting more cash, so this is the question he's asking himself and for that reason seemingly a more relevant question.

Bobcat did yield much more profits than the Derp family and it was developed with a much smaller budget. It doesn't sell for a high price but it doesn't cost too much to make and Intel didn't have anything to compete against it until 22nm Sandy Bridge.

Profits =! high performance, if AMD is to survive it should not try to beat Intel - or any other orders of magnitude bigger company - on its own turf, it must think smart, get ahead of the trends or go to markets not interesting, or viable to these bigger companies. I think AMD is going towards the latter choice, with the the console chips being such an example: No other bleeding edge company would swallow the kind of margins AMD did.

K12/Zen should not aim to compete against Cherry Trail or SkyLake, this battle is lost and K12/Zen will be DOA if it try to compete on the same market, K12/Zen should enable AMD to build businesses outside Intel or the big ARM companies's area of influence, a niche to AMD to call theirs.

 

[Fjodor2001]

K12/Zen should not aim to compete against Cherry Trail or SkyLake, this battle is lost and K12/Zen will be DOA if it try to compete on the same market, K12/Zen should enable AMD to build businesses outside Intel or the big ARM companies's area of influence, a niche to AMD to call theirs.

So what market segment do you suggest Zen and K12 should target?

Anyway, here's my suggestion:

I think the 4 core version of AMD Zen should aim at ~10-30% lower CPU performance than Intel's top end desktop chips, but with better iGPU and lower price. That is good enough for most people, so it'll sell boat loads of chips.

Then they should also sell an 8 core version without iGPU too. It should outperform Intel's 4 core CPUs in MT loads, and be much cheaper than Broadwell/Skylake-E. Thereby AMD will grab the the high end desktop segment, including gamers.

Just my 2 cents if I was product planner at AMD...

 

[Qwertilot]

Well one obvious option is to use HBM to really go after low end (+) discrete GPUs with the iGPU. Should have a decent market as NV won't be able to follow them there of course and Intel might not be hugely interested.

It'd be rather more surprising if they could outgun Intel with a CPU for high end gaming.

 

[elemein]

So what market segment do you suggest Zen and K12 should target?

Anyway, here's my suggestion:

I think the 4 core version of AMD Zen should aim at ~10-30% lower CPU performance than Intel's top end desktop chips, but with better iGPU and lower price. That is good enough for most people, so it'll sell boat loads of chips.

Then they should also sell an 8 core version without iGPU too. It should outperform Intel's 4 core CPUs in MT loads, and be much cheaper than Broadwell/Skylake-E. Thereby AMD will grab the the high end desktop segment, including gamers.

Just my 2 cents if I was product planner at AMD...

Yeah your plans for AMD are way too ambitious.

They have a far lower R&D budget, are under shifty management and are will be making a completely new design. It will not be able to get even near the performance of top Intel SKUs-- especially not with a better iGP in addition.

I like the optimism. Im not saying Zen will be useless and I'm not bashing on AMD. Simply being realistic. Im expecting Zen to be between the cat cores and the BD family cores. It'll be an interesting design, but not something for high end consumer computing.

 

[Abwx]

Im expecting Zen to be between the cat cores and the BD family cores..

Most irrelevant expectations ever...

To write things like this imply that you certainly dont know the difference between a Kabini and a Steamroller core.

 

[Fjodor2001]

They have a far lower R&D budget, are under shifty management and are will be making a completely new design. It will not be able to get even near the performance of top Intel SKUs

Then why would AMD bother making a completely new uArch to begin with?

If there should be any point to it, it has to be significantly better than the current one. And to be significantly better, my guess is it cannot be more than around 10-30% lower performance than Intel SKUs. Otherwise there would be no point in doing it, except perhaps to improve perf/watt.

Or what's your expectations? That Zen will be only 30-50% as fast as the Intel SKUs?

 

[elemein]

Most irrelevant expectations ever...

To write things like this imply that you certainly dont know the difference between a Kabini and a Steamroller core.

Okay, so instead of just saying "you dont know", why not explain?

We can't have speculations on an unreleased product?

I'm definitely not in the minority when it comes to thinking Zen will be a cat successor. Many people believe that.

Also, I do know the difference between a Jag and a SR " core". Though I'd rather just compare module to a pair of cores since that'd be a much better comparison.

Either way, trust me. I know the differences between the archs. I just have different speculations and expectations from you. Isnt that okay?

I'd love to learn more and even be convinced over to your opinion of the future product with the right explanation, but a "you dont know what youre talking about" won't do it.

 

[elemein]

Then why would AMD bother making a completely new uArch to begin with?

If there should be any point to it, it has to be significantly better than the current one. And to be significantly better, my guess is it cannot be more than around 10-30% lower performance than Intel SKUs. Otherwise there would be no point in doing it, except perhaps to improve perf/watt.

Or what's your expectations? That Zen will be only 30-50% as fast as the Intel SKUs?

I'm part of the group of people that think that Zen won't be aimed at high end desktop computing.

Sorry for double post, on mobile.

 

[Ramses]

What exactly is "high end desktop computing" these days?
A level of performance or actual tasks?

 

[elemein]

What exactly is "high end desktop computing" these days?

Replying cause I was the last one who mentioned the phrase ignore if not aimed at me.

To me, my definition would be along the lines of something like an i7-4790K or 5960X and some big flagship-ish GPU.

 

[Fjodor2001]

I'm part of the group of people that think that Zen won't be aimed at high end desktop computing.

Sorry for double post, on mobile.

But they already have the cat cores to cover the low end and low power segment, both on desktop and mobile/laptop. That design is generally acclaimed for being good for it's segment, and the uArch is quite new.

If they are aiming at higher performance mobile/laptop CPUs than what the cat cores provide, then those new CPU cores can be used on the desktop too. I.e. same as Intel does with it's Core uArch cores. So if AMD can use the new Zen cores on Desktop too, we're back to my previous estimate/guess.

 

[elemein]

But they already have the cat cores to cover the low end and low power segment, both on desktop and mobile/laptop. That design is generally acclaimed for being good for it's segment, and the uArch is quite new.

If they are aiming at higher performance mobile/laptop CPUs than what the cat cores provide, then those new CPU cores can be used on the desktop too. I.e. same as Intel does with it's Core uArch cores. So if AMD can use the new Zen cores on Desktop too, we're back to my previous estimate/guess.

It seems we dont disagree quite as much as you think

Here is my expectation and why;

I believe Zen will be a "medium" core, between Jag and the BD family. I believe it'll scale quite well, but it'll be more aimed at the "wider" audience of consumer computing; ie. Not high end desktop, not tabs, and not any niches. It'll be a chip for most standard notebooks 11.6"-15.6" and have a desktop presence akin to the one the G3258, i3 line, 860K and 6300 have now; great CPUs for those on a budget (ie. Most folk).

I do not see them going for ultra mobile (tabs mainly) because well... Look at Temash and Mullins (wait does anyone even remember those things?)

I do not see them going at Intel's flagship because the R&D budget, lithographic technology available to AMD, the time, the tools and market share available to AMD doesn't really allow them the ability to get at Intel's flagship performance without running high on TDP/missing an iGP/really some weakness. Something has to give.

I definitely think Zen will be an interesting, ambitious and unique design. I just don't see it being a "ultra low power" or "ultra performer" chip. I think the approach is more "jack of all trades, master of none" to snag more affordable consumer market share from Intel in the most common form factors (cheap desktop or standard laptop.) I also think Zen will be aimed to be AMD's only x64 chip, tossing out the BD and Cat family.

Those are my expectations and reasoning. If you disagree, thats fine If you like, provide your expectations and reasoning and where I may have gone wrong.

 

[Abwx]

Okay, so instead of just saying "you dont know", why not explain?

We can't have speculations on an unreleased product?

I'm definitely not in the minority when it comes to thinking Zen will be a cat successor. Many people believe that.

Also, I do know the difference between a Jag and a SR " core". Though I'd rather just compare module to a pair of cores since that'd be a much better comparison.

Either way, trust me. I know the differences between the archs. I just have different speculations and expectations from you. Isnt that okay?

I'd love to learn more and even be convinced over to your opinion of the future product with the right explanation, but a "you dont know what youre talking about" won't do it.

Next Bulldozer iteration, Excavator, will be released in 3-4 months while Zen will be released 15 months later, according to you Zen could be between a Jaguar and an Excavator, sorry but this doesnt make sense, the perfs delta between thoses uarch is not enough to be covered by a third core, and also why would AMD need all thoses years to release a core that would be of Steamroller level at most according to you.??.

 

[elemein]

Next Bulldozer iteration, Excavator, will be released in 3-4 months while Zen will be released 15 months later, according to you Zen could be between a Jaguar and an Excavator, sorry but this doesnt make sense, the perfs delta between thoses uarch is not enough to be covered by a third core, and also why would AMD need all thoses years to release a core that would be of Steamroller level at most according to you.??.

You're missing what I'm saying.

Im saying Zen won't perform as well as Excavator since it's not aimed at even getting close to Intel's flagship.

Though it won't scale as low as Jaguar did when it was designed to take the Temash and Mullin SKUs into account.

When I say "jack of all trades, master of none", I really do mean master of none. It won't perform as well as Steamroller, nor scale as low as Jag, it'll perform and scale somewhere in between.

Again, that's just my educated guess. The reality could be completely different.

 

[Fjodor2001]

I do not see them going at Intel's flagship because the R&D budget, lithographic technology available to AMD, the time, the tools and market share available to AMD doesn't really allow them the ability to get at Intel's flagship performance without running high on TDP/missing an iGP/really some weakness. Something has to give.

Zen will be on 14 nm (Samsung/GF most likely). Intel will also be on 14 nm at the time Zen is released. Sure, those 14 nm process techs will have different characteristics, but still Samsung/GF 14 nm will not be a bad process tech at the time Zen is released.

Zen is also a completely new uArch. Intel's Core uArch is based on a design originating from 2006.

I definitely think Zen will be an interesting, ambitious and unique design. I just don't see it being a "ultra low power" or "ultra performer" chip. I think the approach is more "jack of all trades, master of none" to snag more affordable consumer market share from Intel in the most common form factors (cheap desktop or standard laptop.)

There will be an 8 core Zen CPU according to this. Do you think it makes sense to make an 8 core Zen based CPU, if it's aiming for the low/mid-end laptop and desktop segments?

 

[frozentundra123456]

Next Bulldozer iteration, Excavator, will be released in 3-4 months while Zen will be released 15 months later, according to you Zen could be between a Jaguar and an Excavator, sorry but this doesnt make sense, the perfs delta between thoses uarch is not enough to be covered by a third core, and also why would AMD need all thoses years to release a core that would be of Steamroller level at most according to you.??.

Why did they need all those years to release Bulldozer? Time elapsed does not guarantee a good product.

 

[elemein]

Zen will be on 14 nm (Samsung/GF most likely). Intel will also be on 14 nm at the time Zen is released. Sure, those 14 nm process techs will have different characteristics, but still Samsung/GF 14 nm will not be a bad process tech at the time Zen is released.

Zen is also a completely new uArch. Intel's Core uArch is based on a design originating from 2006.


There will be an 8 core Zen CPU according to this. Do you think it makes sense to make an 8 core Zen based CPU, if it's aiming for the low/mid-end laptop and desktop segments?

IIRC, the real basis for Intel's arch is the PIII. If you compare a P3 block diagram to a Haswell diagram, you see some spooky similarities. It's even older! Though architecture age doesn't really have anything to do with what I'm saying. Do correct me if it's somehow relevant though.

Also, I'm not saying Samsung/GF 14nm will be "bad" or anything like that, simply saying that due to the diversity of process Intel can offer (having 14nm SoC, HBM, etc. etc.), they'll have the upper hand. I don't believe AMD has this ability with their strategy to use open foundries.

I agree, the news that Zen will have an 8-core SKU does bring up a few things. Though I just chalk it up to "Well Haswell scales from 4.5W to 140W through varying different flavours, can't see why Zen won't scale from some ~10W to 95W with the flagship being an eight core with the aim to take most of the mid-low tier market."

Notice that the link describes Zen going up to 8 cores and going up to 95W, not that all SKUs will be those specs.

Also, another reason I forgot is that I don't believe they'll have the plans to make a pin-compatible ARM sister and have it be super-high performing on a consumer socket when it's supposed to have comparable performance to it's x86 sister.

Though, if the reality turns out to be different and Zen is aimed at the i7-6770K (or will it be 7770K by then? Scheduling has been messy lately)'s neck, then I'll more than be happy to admit I was completely wrong with my expectations...

Then I'll be excited cause... you know... it's been awhile since Intel has had high-end competition.

 

[Idontcare]

Zen is also a completely new uArch. Intel's Core uArch is based on a design originating from 2006.

To me, Zen sounds a lot like a 14nm shrink for Jaguar. And you know Sony and MS are both going to be interested in a 14nm version of the 28nm chip they are already using in their consoles.

 

[Ramses]

Replying cause I was the last one who mentioned the phrase ignore if not aimed at me.

To me, my definition would be along the lines of something like an i7-4790K or 5960X and some big flagship-ish GPU.

That makes sense. Everyone else using that definition more or less?

 

[Fjodor2001]

To me, Zen sounds a lot like a 14nm shrink for Jaguar. And you know Sony and MS are both going to be interested in a 14nm version of the 28nm chip they are already using in their consoles.

Just wondering, how come you think Zen sounds like a 14 nm shrink of Jaguar?

AMD says Zen is a completely new x86 uArch, so I don't see how it can be just a 14 nm Jaguar shrink then?

Also, see this. 95 W TDP for 8 Jaguar cores on 14 nm and no iGPU? Naaaah.... really?

Even a die shrunk PS4 APU would be too much at 95 W TDP on 14 nm.