bitsandchipsAMD Reveals the Monsterous ‘Exascale Heterogeneous Processor’

[podspi]

When a new AMD CPU is on the way, AMD users think it's going to leap frog multiple generations of Intel CPU's that have dominated AMD and suddenly be the performance king. When that doesn't happen, you wait for "software optimizations" for the next few years that never happen.

How do people here think projects are engineered? Assuming Zen is not a complete failure, it will leapfrog multiple generations of Intel CPU architectures. If it doesn't it wouldn't be even marginally competitive, and they could maximize shareholder value by folding up .


Because every tech argument deserves an auto analogy:

Have you ever heard of Tesla? What were the state of their Sedans in the 70s? 80s? 90s? Nonexistent. And yet somehow they managed to release an automobile with all the latest stuff (winning lots of safety awards even) and arguably even moved forward the state of the art.

How could they POSSIBLY do such a thing, leapfrogging all previous generations of automobiles? Did they steal technology? Was it dark blood magic? No, it was just the fact that engineering does not happen in a vacuum. The phrase "it's easy once you know how its done" comes to mind.

This goes double if the company in question has a broad cross-licensing agreement with its competitors, which AMD does. This idea that they can only move incrementally from Bulldozer is at best ignorance on how things are actually designed and built, and at worst FUD.

All of that being said, that doesn't mean it will take the performance lead, or even be particularly competitive - because there are lots of other issues AMD is facing, such as: chronically terrible management, lack of financial resources, and employee brain drain.

Personally, I hope it blows the doors off anything Intel has - I'd spend a nice chunk of change happily if it does. I don't think it will, but I do hope it does, and I will not be cheering if it doesn't. Luckily we have the ARMy to keep Intel in check, at least for client/consumer level stuff.

 

[Sweepr]

Lol...

47W, officialy, to get 10% better 3DMark 11 score than a 35W Carrizo, that s what you call untouchable..

Did you see the laughable graphics performance increase AMD expects at 3DMark 11 from their 2016 desktop APU (~15% better than Carrizo)?

http://www.kitguru.net/wp-content/uploads/2015/06/amd_bristol_ridge_expectations.png

If that leak is right Bristol Ridge will barely beat 1-year old by then Broadwell GT3e. And no, it won't touch Skylake GT4e neither Kabylake GT4e.

 

[DrMrLordX]

Bristol Ridge is a non-starter for anyone who actually understands PC performance and wants more of it. It might be a nice cheap bump up from Kaveri/Kaveri refresh, but that's it. Remember, it's just Excavator on the desktop.

 

[frozentundra123456]

That was leaked, Skylake-S's official introduction happens this week and at IDF.
Also the performance claims from those slides are not outrageous, certainly in the ballpark of what you would expected from an Intel tock. It's not like they are talking about huge IPC gains one and a half year before (expected) launch.

I agree. You weren't the poster I was talking about if that is what you think.

 

[nismotigerwvu]

Core 2 might have been revolutionary on desktops but it was an evolutionary product as far as mobile was concerned. Considering the top notch manufacturing process, new (tock) CPU architecture, new graphics architecture (HEVC decode/encode), 15W Iris Pro SKUs and a brand new GT4e option for high-end notebooks/desktops (no AMD APU will touch it till 2017), improved I/O, wireless charging & WiGig, integrated imaging solutions + touch + sensor hub, strong focus on mobile (lowering both chip and platform power) and the fact that server Skylake (Purley) is described as the biggest platform advancement since Nehalem (6-channel DDR4, 28-cores)... yes, I think it might be very important to the company. They could easily justify that claim.



It is, list of reasons above, your personal feelings on the matter won't change facts.

That's actually a solid point. On the desktop, Conroe was a revelation compared to the outgoing mess of Prescott derived Smithfield/Presler lineup. On the mobile side Merom was replacing Yonah, Netburst was already long gone and "the leap" had already happened. Merom fit more in line with the tick/tock we are familiar with today. Honestly though, I don't think Skylake will be that groundbreaking, more just a solid iteration on a successful formula or more or less the same level of improvement we've been seeing with each generation from Intel for the last decade.

 

[frozentundra123456]

How do people here think projects are engineered? Assuming Zen is not a complete failure, it will leapfrog multiple generations of Intel CPU architectures. If it doesn't it wouldn't be even marginally competitive, and they could maximize shareholder value by folding up .


Because every tech argument deserves an auto analogy:

Have you ever heard of Tesla? What were the state of their Sedans in the 70s? 80s? 90s? Nonexistent. And yet somehow they managed to release an automobile with all the latest stuff (winning lots of safety awards even) and arguably even moved forward the state of the art.

How could they POSSIBLY do such a thing, leapfrogging all previous generations of automobiles? Did they steal technology? Was it dark blood magic? No, it was just the fact that engineering does not happen in a vacuum. The phrase "it's easy once you know how its done" comes to mind.

This goes double if the company in question has a broad cross-licensing agreement with its competitors, which AMD does. This idea that they can only move incrementally from Bulldozer is at best ignorance on how things are actually designed and built, and at worst FUD.

All of that being said, that doesn't mean it will take the performance lead, or even be particularly competitive - because there are lots of other issues AMD is facing, such as: chronically terrible management, lack of financial resources, and employee brain drain.

Personally, I hope it blows the doors off anything Intel has - I'd spend a nice chunk of change happily if it does. I don't think it will, but I do hope it does, and I will not be cheering if it doesn't. Luckily we have the ARMy to keep Intel in check, at least for client/consumer level stuff.

I have no idea what Zen will be like, but OTOH, this very same agrument could have been made about Bulldozer too. By this reasoning, every new product that comes out would be a major advance. Doesnt necessarily work that way.

 

[2is]

If Intel says that Skylake is its "most important chip architecture in a decade", and it's nowhere near that by any metric in reality, Intel is still overhyping it no matter how you try to spin it.

If after this post you still think I'm "spinning" in any way shape or form, you shouldn't respond because it would mean that you still have no idea what I'm talking about, and I've made it as elementary as I know how.

 

[2is]

How do people here think projects are engineered? Assuming Zen is not a complete failure, it will leapfrog multiple generations of Intel CPU architectures. If it doesn't it wouldn't be even marginally competitive, and they could maximize shareholder value by folding up

You must not have kept up with AMD's CPU releases much since your join date in 2011.

 

[mrmt]

There had been rumors about AMD working on a huge APU with Zen cores and Greenland HBM graphics, something that AMD had hinted upon in its official roadmap. However, it has (finally) officially revealed details about the upcoming APU in a paper submitted to IEEE (Institute of Electrical and Electronics Engineer). T

This is just a research paper, it is meant to say where AMD is putting its money in terms of R&D, but between this and a product roadmap there's a lot of things yet to happen. It's a bit premature to say it's an upcoming product.

 

[CHADBOGA]

And i already told you SKL wont change the picture, it will just bring Intel more or less on parity features wise but it will have lower GPU perf/Watt and not a better CPU perf/Watt, at best it will barely match Carrizo in this register.

Skylake based products will continue the reduction of AMD's market share in mobile.

 

[Dresdenboy]

I have no idea what Zen will be like, but OTOH, this very same agrument could have been made about Bulldozer too. By this reasoning, every new product that comes out would be a major advance. Doesnt necessarily work that way.

R&D is always involving risks. There are ways to minimize them from the start, but there is a limit. And the more a new thing diverges from the existing ones, the higher the risk. But even having the best cycle exact simulations of a new architecture (which still are too slow to run generic applications and games so one has to prioritize) doesn't help in avoiding risks at other places like development processes (e.g. increased amount of auto placement and route, etc.), production processes (32nm SOI, 28nm Custom), and so on.

The best experience needed for planning and developing a new complex product with desired specs is to have it developed at least once before.

 

[IlllI]

for some reason I was reminded of this

 

[podspi]

I have no idea what Zen will be like, but OTOH, this very same agrument could have been made about Bulldozer too. By this reasoning, every new product that comes out would be a major advance. Doesnt necessarily work that way.

You must not have kept up with AMD's CPU releases much since your join date in 2011.

Very large difference between "can" and "will".

I am not saying it will, I am saying it can. Many people here are saying can't, as opposed to won't.

Obviously Bulldozer was a failed project, (and this is directed towards 2is' comment): Implying that I said Bulldozer was or needed to be an advancement for my argument to be true is a pretty blatant strawman.

 

[2is]

Very large difference between "can" and "will".

I am not saying it will, I am saying it can. Many people here are saying can't, as opposed to won't.

Obviously Bulldozer was a failed project, (and this is directed towards 2is' comment): Implying that I said Bulldozer was or needed to be an advancement for my argument to be true is a pretty blatant strawman.

Call it what you want. I'm basing my expectations on several years of disappointing products. You're basing it on, what, exactly? What is in the realm of being physically possible however improbable?

And you didn't say "can" you said "will" you even freakin bolded the word "will"

 

[frozentundra123456]

Very large difference between "can" and "will".

I am not saying it will, I am saying it can. Many people here are saying can't, as opposed to won't.

Obviously Bulldozer was a failed project, (and this is directed towards 2is' comment): Implying that I said Bulldozer was or needed to be an advancement for my argument to be true is a pretty blatant strawman.

Did you read your own post? You specifically posted that Zen "will" (bolded even), leapfrog several generations of Intel processors.

 

[Fjodor2001]

If after this post you still think I'm "spinning" in any way shape or form, you shouldn't respond because it would mean that you still have no idea what I'm talking about, and I've made it as elementary as I know how.

I think only an Intel-blind fanboy would consider Skylake to be the Intel's "most important chip architecture in a decade". For everyone else that statement is a complete joke and seriously overhyping it.

 

[2is]

I think only an Intel-blind fanboy would consider Skylake to be the Intel's "most important chip architecture in a decade". For everyone else that statement is a complete joke and seriously overhyping it.

Unfortunately, based on this and every other one of your replies, I've come to the conclusion that you just don't have the capacity to understand what i'm saying, despite it being quite literally, spelled out for you. I'm sorry.

 

[podspi]

Call it what you want. I'm basing my expectations on several years of disappointing products. You're basing it on, what, exactly? What is in the realm of being physically possible however improbable?

And you didn't say "can" you said "will" you even freakin bolded the word "will"

Did you read your own post? You specifically posted that Zen "will" (bolded even), leapfrog several generations of Intel processors.

I did bold will, but the other words still count - I am sorry I threw you guys for such a loop:

How do people here think projects are engineered? Assuming Zen is not a complete failure, it will leapfrog multiple generations of Intel CPU architectures. If it doesn't it wouldn't be even marginally competitive, and they could maximize shareholder value by folding up .

(Emphasis added)
Further...

All of that being said, that doesn't mean it will take the performance lead, or even be particularly competitive - because there are lots of other issues AMD is facing, such as: chronically terrible management, lack of financial resources, and employee brain drain.

Given that AMD's top-end at this point falls pretty far behind Intel's in a lot of metrics, it would be entirely possible for them to leapfrog multiple generations and *still* be behind Intel.

To be clear, I am not arguing that Zen will beat Skylake, be competitive, or any of that. I am arguing that AMD, due to their relative position to Intel, can produce a design with greater than marginal increases in performance (as opposed to Intel, who typically has been making only marginal increases, but at a much steadier pace).

None of what I said is terribly complicated, or controversial. See also: ARM and their partners. The performance gains the last few years have been HUGE relative to Intel. The reason that's happening isn't because they are working hard and Intel isn't, or that Intel has lost its magic, it is because Intel is on the forefront and the ARMy isn't.

I don't think there is much room for misinterpretation in my above statement. That being said, I personally hope Zen is competitive, and I don't understand the thinking of anyone who doesn't. I'm optimistic about it because it doesn't cost me anything - and if they come through I will be first in line to help out their balance sheet. And if they don't, I'll continue buying Intel.

 

[2is]

Bolds the word "will" claims he never said it.

Sounds like you threw yourself for a loop.

 

[majord]

Core 2 might have been revolutionary on desktops but it was an evolutionary product as far as mobile was concerned. Considering the top notch manufacturing process, new (tock) CPU architecture, new graphics architecture (HEVC decode/encode), 15W Iris Pro SKUs and a brand new GT4e option for high-end notebooks/desktops (no AMD APU will touch it till 2017), improved I/O, wireless charging & WiGig, integrated imaging solutions + touch + sensor hub, strong focus on mobile (lowering both chip and platform power) and the fact that server Skylake (Purley) is described as the biggest platform advancement since Nehalem (6-channel DDR4, 28-cores)... yes, I think it might be very important to the company. They could easily justify that claim.



It is, list of reasons above, your personal feelings on the matter won't change facts.

So, new platform features, and what seem to be pretty evolutionary architecture improvements (both x86 and GPU)..Just like you'd expect from any new product.

Bearing in mind platform features have nothing to do with the Skylake architecture itself, which is clearly what's being discussed.

 

[majord]

That's actually a solid point. On the desktop, Conroe was a revelation compared to the outgoing mess of Prescott derived Smithfield/Presler lineup. On the mobile side Merom was replacing Yonah, Netburst was already long gone and "the leap" had already happened. Merom fit more in line with the tick/tock we are familiar with today. Honestly though, I don't think Skylake will be that groundbreaking, more just a solid iteration on a successful formula or more or less the same level of improvement we've been seeing with each generation from Intel for the last decade.

It's not that simple.

Memrom was revolutonary in that not only did it increase IPC substantially from the only recently released Yonah (~20% + even more for heavy SSE) it did so whilst lengthening the pipeline, allowing far greater frequency potential.. and if that wasn't enough, it did it all without using any more power. This meant it could be used from top to bottom - Laptop-enthusiast-server. and dominate them all.

This was beyond expectation, and hasn't repeated since.. (though Sandy bridge was still significant - it came with a node shrink for 4/8c parts which contributed to its advantages over 4/8 Nehalem)

I don't think anyone even remembers what platform-level mumbo jumbo conroe/memrom brought with it (not much I don't think), because it didn't really matter in the shadow of it's performance leap

 

[Tuna-Fish]

Bolds the word "will" claims he never said it.

Sounds like you threw yourself for a loop.

Umm, are we reading the same post?

Assuming ... will ...

Is not at all the same statement as one with just "will".

 

[2is]

Umm, are we reading the same post?



Is not at all the same statement as one with just "will".

Then there was no reason for the post at all. That's like saying "if it doesn't suck it won't suck"

 

[yottabit]

AMD, touting moar coars....

Hmm...

 

[Greenlepricon]

AMD, touting moar coars....

Hmm...

Assuming this is a server/HPC chip, then yes, moar coars is better. I expect desktop Zen to have quite a few skews with notably fewer cores. Probably higher clocked and smaller dies as expected from both Intel and AMD.