Intel's getting slower and Kaby Lake processors delayed[PCPER]

[Notional]

They may say that, but I'm not going to believe it until I see it.

 

We already saw it with carrizo. Next year they move from 28nm to 14nm. That will give a huge boost.

[patrickjp93]

Oh I know, but broadwell isn't fantastic compared to Haswell. Maybe 5% better? I guess it will be better if they bump up the number of cores on all the -e versions, but the non -e versions are only up to 4 cores, so doubt it. I really hope ZEN sturs up some shit.

We haven't seen what Broadwell can really do yet anyway. The C SKUs were not built for enthusiast use or high clock rates. The HEDT platform is a different beast, and I wouldn't discount it so readily.

 

Zen won't stir up anything by my calculations unless they've got all 8 cores running at 4.2 GHz. At a promised 95W TDP, good bloody luck wishing for that. We don't have the software to dynamically scale to the number of available cores on a system, and games certainly aren't actually using all the power available to 4 threads on Sandy Bridge yet either. And it's not like Intel's lack of throwing more cores to the mainstream can be blamed for this. Intel helped develop all the programming frameworks and pushed to make scalable multithreading easy to do. It's literally easy enough you just throw OpenMP markup language around your existing C/C++ and have a ball. You can explicitly control thread counts in the program or let it scale on its own to however many cores you have available. Controlling it is easy. Unfortunately until software catches up, there's little either Intel or AMD can do to break past this performance barrier we seem to be rubbing up against it. We're nearing the limits of what we can do on SISD computing per core before we're locked down entirely to clock rates and cache speeds. At that point it's SIMD, MIMD, multicore, or bust! I suppose we could try something like VLIW as Nvidia did with Denver using an optimization cache, but then you need twice as many cores as you intend to have running the actual program, because at least 1 core has to be optimizing and translating the instructions of x86/ARM ahead of time.

[patrickjp93]

They have a monopoly position. 

No they don't. They have IBM and Oracle giving them headaches in scale-up HPC environments and ARM giving them trouble in Smartphones, tablets, microcontrollers, and embedded systems. Intel is still driven to deliver more performance at lower power and lower cost. Let's also not forget 14nm gave them fits in the first place. If 10nm goes as badly or worse everyone will be stuck at 14nm, not just Intel. ASML is currently under fire from Intel about possibly lying about the capabilities of their machines in low-resolution multipatterning, and if those accusations hold true then 14nm will be the next long-haul node like 28nm was. Intel may be equipped with more EUV machines than TSMC and Samsung combined, but 15 does not a bustling foundry make, and EUV still has tremendous obstacles to overcome regarding light sources and mask deterioration before Intel's financially comfortable using it for mass production. TSMC says it exposed 1000 wafers in a day, but do you know how many light sources burnt out and how many masks had to be switched out in that day? It's about 40 wafers before the light source burns out for EUV right now. At $4500 a piece, those light sources are not cheap regardless of your yields.

[Zeruel]

IPC has little more to grow for SISD computing. Already most of the clock latencies per instruction are between 1 and 3 cycles. There's a hard theoretical limit tied to clock speed once every instruction only requires 1 cycle, and division will never be that fast anyway due to the algorithmic requirements. The GHz race is pretty much over. The ILP race is pretty much done too. It's up to programmers and compilers to deliver on it. Branch prediction was already Intel's strongest suit at 95+% accuracy. IPC is nearing the end of its growth. At this point newer, wider instructions need to be used, or tasks need to be broken down into more threads. Perhaps Intel could do 4-way or 8-way SMT like on Xeon Phi or Power 8 respectively, but you'd still need the thread count.

The age of big IPC gains is over. No one is going to blast ahead of Intel. Everyone will get very close to each other, and then it will be a contest between SIMD, MIMD, and MPMD computing models: who can fit the most cores with the most ALUs/dedicated units, the deepest pipeline, the widest pipeline, the highest clock speed, the biggest cache, and the fastest cache? It's not going to be about IPC when every instruction is at its theoretical limits. It's going to be about which programming model wins out and who will have the hardware best suited for it. So far Intel has the best SIMD. Oracle has the best MPMD, and IBM has the best MIMD CPU. AMD has the best heterogeneous integrated solution, but that's a small distraction from the big transformative battle coming in computer science.

Damn, we all know how slowly software actually gets developed to decently utilize hardware.

[___]

-snip-

I've really liked reading your posts this thread. That's it, I don't really have a comment on this news post that hasn't already been said.

[MEC-777]

So basically, if you have an i5 or i7 from Sandy Bridge or newer, you're pretty much set for quite some time yet. At least until Zen comes along and only if Zen is powerful enough to be disruptive. 

[Curufinwe_wins]

I'm not particularly sure what people are talking about with broadwell being underwhelming or a small improvement. The 5775c is basically equivalent to the 4790k while running at 30% less power and 20% power clock speeds. Sure the current ones we have don't overclock great, but they are the midrange 65w ones anyways. I personally am extremely excited to see what a high power skylake chip could do. Btw previous comments on future of computating are dead on. #hpcmatters

[Guest Strangerbob]

Not surprising. No one is nipping at their heels or putting them on the ropes now.

Hopefully Zen gives them a good slap.

 

YEYEYEYEYEYEYE.... ZEN ZEN ZEN

 

All of you who keep saying that just want ZEN to be good so you can then go buy better and cheaper INTEL cpu`s.

 

 

I hope AMD does what they have done the last 2 years with new products - utterly disappointed everyone. So they can finally go bankrupt and the fanboys can get the world they want so much - a monopoly.

[patrickjp93]

YEYEYEYEYEYEYE.... ZEN ZEN ZEN

All of you who keep saying that just want ZEN to be good so you can then go buy better and cheaper INTEL cpu`s.

I hope AMD does what they have done the last 2 years with new products - utterly disappointed everyone. So they can finally go bankrupt and the fanboys can get the world they want so much - a monopoly.

I've said it before and will again. AMD dying sooner is better for consumers and competition than hanging on and barely competing while Intel manhandles Nvidia in the HPC space. If AMD goes down, Intel would buy up ATI in a heartbeat and be able to cancel its IP agreement with Nvidia. Intel would rapidly start applying that IP to its iGPUs and a new dGPU line.

[CommandMan7]

I've really liked reading your posts this thread. That's it, I don't really have a comment on this news post that hasn't already been said.

I was just about to comment the same thing. I always enjoy reading about detailed silicon tech and there's no better source I know of than @patrickjp93 

[patrickjp93]

I was just about to comment the same thing. I always enjoy reading about detailed silicon tech and there's no better source I know of than @patrickjp93 

Hehehe, there are better sources about the tech itself. I know a little about a lot of stuff and know how to properly build conclusions from that knowledge. I'm not super knowledgeable about the innards of many silicon products believe it or not.

[LukaP]

A lot of that is due to lower nodes, What this means is, that AMD and Intel will be on the same node again next year (14nm), and that Kaby Lake will launch at about same time as ZEN. But if Zen launches as highend processors, they might be a real threat to Intel, as the "x99" version won't launch until 2017.

 

This is very good news for us all indeed. Much more competition!

TSMC/Samsung/GloFo 14nm FF is not the same as Intels 14nm TG. the density compares more to Ivy bridge's 22nm TG. I reckon with 14/16nm FF+ TSMC can get a little under Haswell's 22nm TG density

[Notional]

TSMC/Samsung/GloFo 14nm FF is not the same as Intels 14nm TG. the density compares more to Ivy bridge's 22nm TG. I reckon with 14/16nm FF+ TSMC can get a little under Haswell's 22nm TG density

 

TSMC doesn't make 14nm, only 16nm. However you are right, that Intel 14nm is not the same as Global Foundries/Samsung 14nm. These are the stats:

 

Intel (Confirmed)

42-nm Fin Pitch

70-nm Gate Pitch

52-nm 1D Interconnect Pitch

0.0588μm² HD SRAM

 

GloFo/Samsung (Unconfirmed)

48-nm Fin Pitch

78-nm Gate Pitch

64-nm 2D Interconnect Pitch

0.064μm² HD SRAM

 

It's worth it to note that 64 nm 2D is better than 52nm 1D, as 2D makes for a higher density. So you can argue that GloFo's 14nm is a tiny bit worse than Intel, but it is not closer to 22nm at all, as that is 60 nm Fin pitch, 90 nm Gate pitch and 80 nm interconnect 1D pitch. I'd say it's close enough to be a contender in performance and power.

[Agent181]

YEYEYEYEYEYEYE.... ZEN ZEN ZEN

All of you who keep saying that just want ZEN to be good so you can then go buy better and cheaper INTEL cpu`s.

I hope AMD does what they have done the last 2 years with new products - utterly disappointed everyone. So they can finally go bankrupt and the fanboys can get the world they want so much - a monopoly.

Please go back to school and learn what monopolies are and see what happens

[Agent181]

I've said it before and will again. AMD dying sooner is better for consumers and competition than hanging on and barely competing while Intel manhandles Nvidia in the HPC space. If AMD goes down, Intel would buy up ATI in a heartbeat and be able to cancel its IP agreement with Nvidia. Intel would rapidly start applying that IP to its iGPUs and a new dGPU line.

I'd rather have amd struggle and give them time to come back than having a monopoly in the market

[patrickjp93]

I'd rather have amd struggle and give them time to come back than having a monopoly in the market

You're operating under the assumption AMD can actually come back from the edge. Moody's just dropped their credit rating again, only 2 positions above Greece. As in the likelihood it pays back its debts is predicted to be less than 60%. I do not remotely have faith that AMD can recover without a huge cash infusion. It's a member of the kind of industry where the barrier to entry is raw cash. With or without Keller and Koduri, AMD does not have a rapid enough development cycle to keep pace, and the low-hanging fruit for CPU performance gains have already been picked on Intel's side of the court. I don't understand why people think Zen will be competitive with Skylake. Theoretically in terms of IPC with a 40% gain over Excavator, they're aimed at Haswell. Unless they can pull off the clock speeds of the old FX 83xx series on 6+ cores (already doubtful at the promised 95W TDP promised), it's just not going to happen, DX 12 or not. On the GPU side AMD's making boneheaded moves like throwing the entire transistor budget into stream processors when they desperately need ROPs. They also threw money at HBM despite the fact bandwidth has not been a concern in gaming nor will be for the foreseeable future, and in doing so they limited themselves to 4GB of VRAM on a card that needs more for the kinds of resolutions it's meant to play under the current game programming model.

 

AMD is making boneheaded moves on the GPU side which it has historically been competitive in, and you really think they're going to recover when new nodes are going to be ever more expensive and risky from here on while they start nearly 2.5 billion USD in the red with 1/4 of it due at the start of 2019 while they're not even breaking even these past two years and show no signs of being able to turn it around? I'm sorry but I think it's time everyone faced facts. The likelihood of AMD surviving the decade is shrinking by the day. Sad or not, AMD dying now is a Hell of a lot better than dying in a theoretical 2022 with Intel having already swallowed Nvidia. At least Nvidia can fight back and fund Keller and Koduri to their hearts' content.

[Agent181]

You're operating under the assumption AMD can actually come back from the edge. Moody's just dropped their credit rating again, only 2 positions above Greece. As in the likelihood it pays back its debts is predicted to be less than 60%. I do not remotely have faith that AMD can recover without a huge cash infusion. It's a member of the kind of industry where the barrier to entry is raw cash. With or without Keller and Koduri, AMD does not have a rapid enough development cycle to keep pace, and the low-hanging fruit for CPU performance gains have already been picked on Intel's side of the court. I don't understand why people think Zen will be competitive with Skylake. Theoretically in terms of IPC with a 40% gain over Excavator, they're aimed at Haswell. Unless they can pull off the clock speeds of the old FX 83xx series on 6+ cores (already doubtful at the promised 95W TDP promised), it's just not going to happen, DX 12 or not. On the GPU side AMD's making boneheaded moves like throwing the entire transistor budget into stream processors when they desperately need ROPs. They also threw money at HBM despite the fact bandwidth has not been a concern in gaming nor will be for the foreseeable future, and in doing so they limited themselves to 4GB of VRAM on a card that needs more for the kinds of resolutions it's meant to play under the current game programming model.

AMD is making boneheaded moves on the GPU side which it has historically been competitive in, and you really think they're going to recover when new nodes are going to be ever more expensive and risky from here on while they start nearly 2.5 billion USD in the red with 1/4 of it due at the start of 2019 while they're not even breaking even these past two years and show no signs of being able to turn it around? I'm sorry but I think it's time everyone faced facts. The likelihood of AMD surviving the decade is shrinking by the day. Sad or not, AMD dying now is a Hell of a lot better than dying in a theoretical 2022 with Intel having already swallowed Nvidia. At least Nvidia can fight back and fund Keller and Koduri to their hearts' content.

Fuck that I'd rather have split companies than have 1 doing nothing, look at Intel right now the delayed processor's and have already stopped bringing in performance and just focused on lower power consumption, I'd rather have amd struggle and fight than get swallowed up by another company, I don't want another Comcast type problem where for a few years we all had shit until some companies started bringing out competition, we need competition in all markets, amd will get someone to help with cash flow, either by another company or the US itself that likes to bailout everything even though we go into more debt.

Also with the people of Greece they choose to figure out things by themselves, they may be in a sbithole for a while but they can figure it out if the politicians want to keep their spots or stay alive.

[patrickjp93]

Fuck that I'd rather have split companies than have 1 doing nothing, look at Intel right now the delayed processor's and have already stopped bringing in performance and just focused on lower power consumption, I'd rather have amd struggle and fight than get swallowed up by another company, I don't want another Comcast type problem where for a few years we all had shit until some companies started bringing out competition, we need competition in all markets, amd will get someone to help with cash flow, either by another company or the US itself that likes to bailout everything even though we go into more debt.

Also with the people of Greece they choose to figure out things by themselves, they may be in a sbithole for a while but they can figure it out if the politicians want to keep their spots or stay alive.

That's entirely untrue. The performance difference between Sandy Bridge and Haswell is around 85% if you can leverage the newer instructions in Haswell. The reality is software is not keeping up, and it's too focused on legacy support since so many still run Windows XP and Windows 7 which work with the freaking Pentium III.

 

If AMD went down, Intel would pick up ATI, and Nvidia would pick up x86_64. That would be better competition than we have now. Intel would jump headlong into dGPUs to try to bury Nvidia before it fully caught up in CPU and SOC designs. You'd have raging 2-sided competition.

 

And the problem is the semiconductor industry is an economy of scale, scale AMD lacks and is not going to gain back without a huge redemption.

 

And the U.S. would never bail out AMD. It's a hamper to competition as it is, and the U.S. government understands that.

[Agent181]

That's entirely untrue. The performance difference between Sandy Bridge and Haswell is around 85% if you can leverage the newer instructions in Haswell. The reality is software is not keeping up, and it's too focused on legacy support since so many still run Windows XP and Windows 7 which work with the freaking Pentium III.

If AMD went down, Intel would pick up ATI, and Nvidia would pick up x86_64. That would be better competition than we have now. Intel would jump headlong into dGPUs to try to bury Nvidia before it fully caught up in CPU and SOC designs. You'd have raging 2-sided competition.

And the problem is the semiconductor industry is an economy of scale, scale AMD lacks and is not going to gain back without a huge redemption.

And the U.S. would never bail out AMD. It's a hamper to competition as it is, and the U.S. government understands that.

Why would nvidia want to compete with Intel they are already far ahead in graphics and it's better to have more than one competitor in the market with only 2 competing with each other they can just come to a standstill, look at it again nvidia knew that the fury x was going to beat the 980 so they released the time faster. They already had this for a while but instead they decided to capitalize on the money instead of pushing for more. Like I said before amd needs to survive to give competition to the other two companies that instead of pushing to new limits and performance they would rather sit on their asses and make more money for their investors.

[patrickjp93]

Why would nvidia want to compete with Intel they are already far ahead in graphics and it's better to have more than one competitor in the market with only 2 competing with each other they can just come to a standstill, look at it again nvidia knew that the fury x was going to beat the 980 so they released the time faster. They already had this for a while but instead they decided to capitalize on the money instead of pushing for more. Like I said before amd needs to survive to give competition to the other two companies that instead of pushing to new limits and performance they would rather sit on their asses and make more money for their investors.

Imagine an Intel armed with its smaller, denser nodes AND the IP of ATI AND Raja Koduri. Intel's problem has always been access to IP. Even the license agreement they have with Nvidia lacks a lot of the fundamental underpinnings required to be able to connect the advanced functional units together into cohesive engines. Intel's problem is not experience or engineering ability when it comes to graphics. If Intel removes that obstacle, it'll go headlong into dGPU production. 

 

Nah, Nvidia's trying to kill AMD and wanted to take the air out of its tires. Take a margin loss now for long-term gains. It's a common business strategy. Nvidia left a price gap and got people on the fence with the Titan X. The 980TI got most of the stragglers to jump. It's not about worrying about the Fury X. It's just about minimizing AMD's victories.

 

AMD needs to roll over and die or get a massive cash injection. The latter is not going to happen unless IBM wants to pony up. Samsung's not going near it, and neither is Qualcomm.

 

Nvidia and Intel would be better as direct competitors than being pushed in small ways by a crippled AMD.

[ItsAFeature]

So basically, if you have an i5 or i7 from Sandy Bridge or newer, you're pretty much set for quite some time yet. At least until Zen comes along and only if Zen is powerful enough to be disruptive. 

 

Let me preface this by saying that my first rig was an AMD rig and I've always had a soft spot for them in my heart. I am willing to bet my job that Zen isn't going to save AMD. In fact, Zen probably won't make much of a difference at this point. AMD is the master of making their upcoming products sound like the bees' knees, yet for me they've been underwhelming for years. We've heard this "our next CPUs will blow the doors off Intel" rabble before. Their R&D budget is abysmal and even their own reports about Zen indicate that they expect only enough gain to stand on an equal footing with Haswell. So let's say Zen comes out and matches Haswell in per-core performance and TDP. Let's even say that Zen gets you 8 cores for the price of a quad-core Haswell.

 

Here are the problems with that:

 

1. Skylake will already be out - Zen won't be the fastest offering. And AMD said they no longer want to be thought of as the cheap option so I can't see them pricing it so that people like me won't just spring for the better performing option.

 

2. Intel is already heavily entrenched with OEMs and in the consumers' minds. AMD won't win that battle by barely being competitive - even if they are a slightly better bang for the buck. And the extras cores they have don't matter to 90% of consumers anyways since games don't really take advantage of it and professionals are going to go for the higher performing option like Skylake-E.

 

3. Servers / supercomputing. You can't win over Intel here by matching Haswell.

 

I wish it weren't so but I've mentally prepared to say goodbye to an old friend. Only hoping the name survives somehow. Oh, and maybe someone will bring back the ATI name while they're at it.

[Agent181]

Imagine an Intel armed with its smaller, denser nodes AND the IP of ATI AND Raja Koduri. Intel's problem has always been access to IP. Even the license agreement they have with Nvidia lacks a lot of the fundamental underpinnings required to be able to connect the advanced functional units together into cohesive engines. Intel's problem is not experience or engineering ability when it comes to graphics. If Intel removes that obstacle, it'll go headlong into dGPU production.

Nah, Nvidia's trying to kill AMD and wanted to take the air out of its tires. Take a margin loss now for long-term gains. It's a common business strategy. Nvidia left a price gap and got people on the fence with the Titan X. The 980TI got most of the stragglers to jump. It's not about worrying about the Fury X. It's just about minimizing AMD's victories.

AMD needs to roll over and die or get a massive cash injection. The latter is not going to happen unless IBM wants to pony up. Samsung's not going near it, and neither is Qualcomm.

Nvidia and Intel would be better as direct competitors than being pushed in small ways by a crippled AMD.

The way it's looking now if it's just nvidia and Intel then there will be no competition.

[patrickjp93]

The way it's looking now if it's just nvidia and Intel then there will be no competition.

How can you say that? Nvidia's Denver was already ahead of Kaveri in per-core performance and had the multithreaded performance of the mobile FX 7000 series chips from AMD. That was Nvidia's first big attempt, and with only 2 cores. The CPU and GPU competition would be huge! Imagine Nvidia SLI vs. Intel XFire in hybrid systems or unified systems. Nvidia also has the R7D budget to far more reasonably compete. Intel and Nvidia already widely compete in the HPC space. Why do you not think the two would compete if they got the cross-wise aspects of AMD? At least explain why you doubt this, because I can't make the logic work.

[patrickjp93]

 

 

 

Let me preface this by saying that my first rig was an AMD rig and I've always had a soft spot for them in my heart. I am willing to bet my job that Zen isn't going to save AMD. In fact, Zen probably won't make much of a difference at this point. AMD is the master of making their upcoming products sound like the bees' knees, yet for me they've been underwhelming for years. We've heard this "our next CPUs will blow the doors off Intel" rabble before. Their R&D budget is abysmal and even their own reports about Zen indicate that they expect only enough gain to stand on an equal footing with Haswell. So let's say Zen comes out and matches Haswell in per-core performance and TDP. Let's even say that Zen gets you 8 cores for the price of a quad-core Haswell.

 

Here are the problems with that:

 

1. Skylake will already be out - Zen won't be the fastest offering. And AMD said they no longer want to be thought of as the cheap option so I can't see them pricing it so that people like me won't just spring for the better performing option.

 

2. Intel is already heavily entrenched with OEMs and in the consumers' minds. AMD won't win that battle by barely being competitive - even if they are a slightly better bang for the buck. And the extras cores they have don't matter to 90% of consumers anyways since games don't really take advantage of it and professionals are going to go for the higher performing option like Skylake-E.

 

3. Servers / supercomputing. You can't win over Intel here by matching Haswell.

 

I wish it weren't so but I've mentally prepared to say goodbye to an old friend. Only hoping the name survives somehow. Oh, and maybe someone will bring back the ATI name while they're at it.

 

And you are an example of a mature, intelligent poster. Thank you. No one's happy that AMD is likely to go. I'm certainly not, but they've slowed down to the point they don't seem able to recover. If Moody's drops their credit rating again, they will be considered more than 50% likely to default on their debt. At that point no investor is going near them except the hair-brained, like maybe IBM or Richard Branson. I think people just need to learn to accept it. AMD is not what we need and cannot survive without a top to bottom overhaul.

[Agent181]

How can you say that? Nvidia's Denver was already ahead of Kaveri in per-core performance and had the multithreaded performance of the mobile FX 7000 series chips from AMD. That was Nvidia's first big attempt, and with only 2 cores. The CPU and GPU competition would be huge! Imagine Nvidia SLI vs. Intel XFire in hybrid systems or unified systems. Nvidia also has the R7D budget to far more reasonably compete. Intel and Nvidia already widely compete in the HPC space. Why do you not think the two would compete if they got the cross-wise aspects of AMD? At least explain why you doubt this, because I can't make the logic work.

That's if they decide to go that route but what's the real possibility what you're saying would happen... In this market anything can happen, it just takes one critical movement to be done that can cripple or help a company and I can't see Intel or nvidia competing against each other. I can see amd getting funds or bought out but not dieing out.