Intel is losing against AMD

[Zarathustra[H]]

Jim Keller will never say it was all him. As much as we want to praise, it wasn't just him. He led the team but it wasn't just him. As for Intel, no they are screwed. They now have to fast track to heterogeneous computing (aka bring in Keller). However, he won't be able to help them get to where AMD is since he can't use anything trademarked or work product. They will be behind the curve in arch for a while. It's not like AMD's gonna sit around and rest on their laurels are they? They have two major issues, their process is broke and they need a new arch that is not monolithic. They are slow as molasses to react because of their reliance on monolithic designs and its also partly tied to their process failure.

lol wut, did you just call AMD's Zen an "ok" design?

Yes. Zen is an OK design. Nothing amazing about it in the slightest. Certainly better than Bulldozer, Pile-driver and Excavator, but that isn't saying much.

Architecturally Intel is still ahead. Just look at how they can stay ahead of AMD 7nm chips in single threaded loads with their 14nm process, an architecture twice it's size.

Sure, the smaller node allows AMD to throw lots of cores at it within the thermal envelope, but architecturally? It's OK.

The only advantage AMD has right now is TSMC. Without TSMC the Zen architecture is an OK mid-range architecture that can't quite keep up with the competition.

Keep in mind, I just ordered a 3960x. I'm no Intel fanboy. I'm just a realist.

 

[thesmokingman]

Yes. Zen is an OK. Design. Nothing amazing about it in the slightest. Certainly better than Bulldozer, Pile-driver and Excavator, but that isn't saying much.

Architecturally Intel is still ahead. Just look at how they can stay ahead of AMD 7nm chips in single threaded loads with their 14nm process, an architecture twice it's size.

Sure, the smaller node allows AMD to throw lots of cores at it within the thermal envelope, but architecturally? It's OK.

The only advantage AMD has right now is TSMC. Without TSMC the Zen architecture is an OK mid-range architecture that can't quite keep up with the competition.

Keep in mind, I just ordered a 3960x. I'm no Intel fanboy. I'm just a realist.

Wow, ok.

 

[Mega6]

Intel needs a home run on its next chip arch. A die shrink on the latest lake core2 derivation will not be enough. Amd is now advancing quickly and by the time intel is on 10nm in volume, tsmc and Amd will be at 5nm. Should intel fail on next arch, the door is open for Amd and huge datacenter gains to lead instead of just competing.

 

[N4CR]

Right now at work we're experiencing over 1 month delays on our most popular laptops and desktops, Dell is siting processor shortages for our i5's (mobile and desktop) as the reason for delay, this is the 3rd or 4th time this year we're going through (very) extended lead times.

Of course, no AMD options available to us through our Dell Portal

Really making this Win10 upgrade project a lot more fun... think I'm going to be sleeping in my office come mid-January thanks to these shortages lol.

But anyway, it's a bummer we do not have AMD options on the professional side (for me at least), not a single offering with anything AMD in it (besides the W/FirePro line in our workstations), server-side included.

Tell them you want AMD or you will move to another vendor.
Dell seems very fast at 'procuring' AMD when that happens... they don't have solutions because you corporates don't dare pressure them enough and intel is paying them off big time. I wish this time the anti-trust lawsuits are 50billion to actually make it not worth doing that, Dell are scum like that.. Grow some balls.

 

[N4CR]

As much as I love seeing AMD having some success, everyone must acknowledge that this can never be more than temporary, right?

Intel have an army of engineers, and as soon as they recover from their production process setback, they are going to be crushing everyone again.

I love AMD, but they are a tiny company compared to Intel. No way they can keep up with Intel once they have process parity.

Don't think so. They are both fighting physics with the node shrinks and the playing field is tipped towards AMD now. That process advantage will shrink with 7nm Intel of course... in two years+ time.
10nm even when running at a better yield is still struggling with clock speed, 7nm probably will be similar. Foveros/chip stacking won't help either as the thermals are shit and the density is higher on 7nm - only way around that is to waste die space by spacing things out. Which means lower yield per wafer.. which means supply constraints or massive investments needed to increase output, leading to lower profitability and shitty shareholders and cash burning, sure they have a lot of cash to burn but they can't do it for ever when a new process node wafer fryer is in the Billion+ price range...
Either way it's not looking good for them.

Also Keller didn't hand design the whole thing, sure he's a good help but to think that he is responsible for everything is pretty disrespectful to the other great engineers involved. And Intel engineers - they are also very capable as witnessed by the ability to keep 14nm competitive in 240p gaming (with help of some compiler/market software dominance too of course).

TLDR: R&D budget won't easily fix this one this time for Intel.

P.s. the fastest CPUs in the world (sometimes by a 2x factor) are 'an okay design'?

 

[thesmokingman]

Or this is perfect too.

 

[Dan_D]

Yes. Zen is an OK design. Nothing amazing about it in the slightest. Certainly better than Bulldozer, Pile-driver and Excavator, but that isn't saying much.

Architecturally Intel is still ahead. Just look at how they can stay ahead of AMD 7nm chips in single threaded loads with their 14nm process, an architecture twice it's size.

Sure, the smaller node allows AMD to throw lots of cores at it within the thermal envelope, but architecturally? It's OK.

The only advantage AMD has right now is TSMC. Without TSMC the Zen architecture is an OK mid-range architecture that can't quite keep up with the competition.

Keep in mind, I just ordered a 3960x. I'm no Intel fanboy. I'm just a realist.

Zen was an OK design. Zen 2 is AMAZING. I don't see how you could possibly conclude otherwise. It has several advantages over Intel's architectures. The biggest disadvantage from AMD's side is no AVX-512 support. However, when you have massive core advantage, it doesn't make any difference. You can brute force your way through with AVX2. Threadripper especially has been nothing but impressive. Not just for the core counts, but for the clocks they can achieve. The chiplet design is also a fantastic way to give the chips modularity.

Wow, ok.

Yeah.....I didn't get it either.

 

[Zarathustra[H]]

However, when you have massive core advantage, it doesn't make any difference. You can brute force your way through with AVX2.

But that's just it.

The massive core advantage. That is 100% due to the process and it's lower thermals which allow for more cores inside of a sane thermal envelope. It has very little to do with the Zen2 design itself.

I agree that the new Threadripper is pretty awesome in the current landacape. That's why I bought one. As the market stands AMD's chips are the ones to have.

All I am saying is that most of their benefit is due to the process advantage, and that is a very temporary situation.

 

[Meeho]

Just like what happened when Intel screwed up with netburst and P4. It gave AMD an opening to come in, but ultimately when Intel fixes their mistake, the unequal drubbing continued.

Fixed their mistake AND used illegal tactics to permanently cripple their oposition and ride the resulting wave to this day.

 

[Meeho]

What AMD has right now is a process advantage, not a design advantage. This here is all about Intel's screwup and how when they fix it, things will be right back to normal, unless AMD really gets their shit in gear, optimizes the hell out of their new arch, and is able to survive once Intel has an OK process again.

This is not true. There is no process available that would make monolithic dies viable for the huge core counts AMD is delivering due to their chiplet design, and they have better SMT (and IPC?).

 

[SmokeRngs]

Has anyone else noticed the complete irony of the situation Intel is in right now? AMD got it's real start because IBM required Intel to find a second manufacturer for x86 CPUs. IBM required this simply because of possible shortages due to a single point of failure if Intel stumbled on the manufacturing side. Right now Intel has a massive problem with inventory because it cannot produce enough chips to satisfy demand. The irony being that AMD, one of the companies Intel farmed out CPU making to, is spanking Intel during the time Intel has massive supply problems.

I have a couple things I would like to point out. First of all, Intel must have a completely new architecture before being able to overcome AMD at this point. That's not an opinion, that's a fact. A process shrinkage with the current architecture will not suddenly propel Intel back to the front. A completely new architecture is also no guarantee that Intel will be able to take the crown back. Netburst was a complete dead end for Intel so we know Intel isn't infallible.

Second point is the manufacturing process. Even years late Intel's 10nm is a failure. Intel has admitted this and flat out stated there won't be anything of note until the 7nm node is out. There are two issues with this. Just as there is no guarantee that a totally new architecture from Intel will be able to take back a performance crown there's no guarantee Intel's 7nm node will be a full success. I'm not stating Intel's 7nm node will be a failure like the 10nm node. At the same time there many very well be issues with it. I understand the same people working on 10nm aren't necessarily the same people working on 7nm but without the success of the 10nm node Intel may not be fully prepared to roll out 7nm without issues.

Beyond that, even if Intel delivers a perfect 7nm node it will take time to get it into production with good yields and then convert additional fabs over to 7nm. Intel will not suddenly on the day of 7nm release have all fabs converted over to it. Additional fabs will have to be shut down to retool and that will take time and could cause additional supply problems.

The point is there is no guarantee Intel will come roaring back as some people seem to believe. Intel must have a lot of things to go perfectly for that to happen. Even if the stars align and everything works perfectly it will still take time before true volume production. There is a very large window of opportunity for AMD to work with here and AMD has shown it's not content to coast along like Intel did for so long.

Despite all of this, Intel is far from out of the game and I'm not predicting the demise of Intel. Inertia alone will continue to keep Intel in the game long term. Couple that with AMD's inability to produce the sort of volume Intel does and no matter how you look at it, AMD has a hard road ahead. Even if AMD quadrupled production tomorrow it still wouldn't hold a candle to what Intel can produce and quantity has a quality all its own.

 

[N4CR]

Despite all of this, Intel is far from out of the game and I'm not predicting the demise of Intel. Inertia alone will continue to keep Intel in the game long term. Couple that with AMD's inability to produce the sort of volume Intel does and no matter how you look at it, AMD has a hard road ahead. Even if AMD quadrupled production tomorrow it still wouldn't hold a candle to what Intel can produce and quantity has a quality all its own.

Intel is definitely not going to die. Their greatest strengths right now are an entrenched supplier model and their excellent software/driver/support base which is AMD's greatest weakness. That is what is stopping AMD going to 20% of the market in servers. Of course along with production.

Both AMD and Intel are production constrained so I wouldn't say it's a strength for Intel, they're both in the same boat. Intel is just a larger company with more market share, hence production (but have the same issue) but at more risk from foundry issues like we see now with 10nm. People can't get intel dCPUs for corporate orders in volume currently. They both are having the same issue (14nm and 7nm - Intel took 14nm offline to make 10nm fabs and are now reverting back some.. too late..).

Keep in mind 7nm EUV is going to significantly decrease production time for some lengthy steps (multi-patterning etc) that is currently employed by TSMC 7nm HPC. So production supply will increase for Zen3 - that's why they are jumping to it so fast, I'd bet money that's one of the main reasons.
I don't see TSMC dropping the ball again like in past - they have too much R&D, money and others waiting on their processes, Apple, Nvidia, AMD, etc etc... there is no room for failure when investing many billions in new process nodes. These new nodes are killers and if they screw one up it could severely impact their business, especially with Samsung floating around looking for customers to bring on board.

 

[GoldenTiger]

Where it counts is a little subjective isn't it? I don't see that Intel is losing in the server room and VDI took a chunk out of the desktop market as did smart phones. I am seeing reports that the virtualized desktop market is expected to double by 2024 so that doesn't sound good for the physical desktop market. The government IT sector is talking Smart Phones and Tablets over Desktops. And I just bought an ASUS ROG Phone II with a dock that allows me to connect to my computer system, displays, external drives, keyboard and mouse. I can game on it as well as do anything else including Office 365 productivity. It's an experiment to see if I can adjust to smart phone gaming over desktop PC gaming and just leave the PC behind. It's not better, but it could be and I think this is simply a limitation of the developers in realizing that PC games can run on mobile without making huge adjustments to the UI experience. In short, there isn't much reason that I can't play Skyrim or anything else on my phone in this dock configuration.

Heh, I remember an anandtech article awhile back talking about how this sort of thing could be possible in the "future". That time seems close at hand in many ways.

 

[Revdarian]

Zen was an OK design. Zen 2 is AMAZING. I don't see how you could possibly conclude otherwise. It has several advantages over Intel's architectures. The biggest disadvantage from AMD's side is no AVX-512 support. However, when you have massive core advantage, it doesn't make any difference. You can brute force your way through with AVX2. Threadripper especially has been nothing but impressive. Not just for the core counts, but for the clocks they can achieve. The chiplet design is also a fantastic way to give the chips modularity.





Yeah.....I didn't get it either.

Yes, this is exactly why I want them to add avx512 support on their next epyc - TR line, because even if one says that avx2 is a great fallback we have seen time and again that certain developers are using compilers that don't seem to recognize this/create paths that support them, and this creates an artificial deficiency higher than expected at times.

The ones still compiling with the "genuine Intel only" libraries will be a PITA if they don't want to recognize that the market is changing tho. Hopefully more ways are found like with Mathlab to activate debugging modes but this may not always be the case. (This is specially true in the sciences /research niche programs)

 

[Zarathustra[H]]

This is not true. There is no process available that would make monolithic dies viable for the huge core counts AMD is delivering due to their chiplet design, and they have better SMT (and IPC?).

I'm not convinced at all.

Right now we know AMD has a process advantage. So what is the best way to eliminate the process advantage, and just look at the merits of the architecture itself?

Well, process advantage results in less heat production, and this is what allows AMD to jam so many cores in their chips, so the best aplroximation we can do is to just look at single or low threaded loads. The reasoning for this is that Intel too will be able to jam more cores on their chips when the heat output is reduced once they go below 14nm.

And in lightly threaded loads Intel is still king. By ever smaller margins, but they still hold that lead. And what's more is that they are doing this on their 14+++++++ architecture, not benefitting from the advantages of a smaller node.

Unless AMD pulls an upset and really executes on their version of tick-tock and optimizes the crap out of this architecture in the just over a year they have left before Intel's 7nm EUV (if we believe early 2021 as the launch date) once Intel also has a smaller process node they will be able to stomp AMD again.

Yes, sure they wi have to tweak the layout a little to work on 7nm. There is always a tweak when the die shrinks, but there is no reason to believe they can't do that.

I see what you are saying about Monolithic dies, but let's be real here. Chopping up dies into smaller parts isn't rocket science. If Intel wanted to do that, they can easily do it. Probably the only reason they haven't thus far is because 14nm is so mature at this point that they don't have a major yield loss problem. I wouldn't be surprised if they also do so on their chips launched on 7nm.

AMD's IPC is currently ever so slightly higher, but it is only half the equation. Per thread performance has always been a design trade-off between IPC and clock speeds. We will have to see what happens there.

AMD has a real chance to stay in the game this time, and I really hope they do. Some of my most fun PC modding and building came during the Athlon era, and I love the underdog. I'm simply stating that this is by no way a done deal.

Intel has the better race car, but they are currently in a pit stop getting their tires changed. AMD really needs to make the most out of this momentary Intel stumble, before the Intel racecar comes roaring back. Thus far they have executed extraordinarily well, but if I were a gambling man, my money would still be on Intel. They have more resources, money, fabs, engineers, more everything to throw at this problem, and an already existing architecture that outperforms AMD on a per core basis on an inferior process.

I think the best we can reasonably hope for once Intel have regained their footing, (let's say in 2022), is Intel again back at the top in the performance wars, but AMD only very slightly behind. Similar to where they were when Phenom II was battling it out vs Core 2.

 

[Mega6]

“Chopping up dies” is not that simple. Infinity fabric makes that all possible. Intel has no counterpart. Process node shrinks and IF can easily keep Amd ahead of intel.

Let’s assume intel keeps up with tsmc In modes moving forward. (Which isn’t the case lately by far).. infinity fabric alone will keep amd ahead.

The only hope intel has to catch Amd is a

1) hell of a great new arch

2) and/or come up with a IF counter part

3) or a serious Amd stumble (which hasn’t happened on Su’s watch yet).

 

[Meeho]

I'm not convinced at all.

Right now we know AMD has a process advantage. So what is the best way to eliminate the process advantage, and just look at the merits of the architecture itself?

Well, process advantage results in less heat production, and this is what allows AMD to jam so many cores in their chips, so the best aplroximation we can do is to just look at single or low threaded loads. The reasoning for this is that Intel too will be able to jam more cores on their chips when the heat output is reduced once they go below 14nm.

I think you're overestimating the influence of node over architecture. AMD's GPUs show that it takes more than a process advantage to gain efficiency.

And in lightly threaded loads Intel is still king. By ever smaller margins, but they still hold that lead. And what's more is that they are doing this on their 14+++++++ architecture, not benefitting from the advantages of a smaller node.

If you're talking about performance, how does node figure into it? And Intel is king only in their small core count CPUs, as they require the mesh bus otherwise and lightly threaded performance suffers.

Unless AMD pulls an upset and really executes on their version of tick-tock and optimizes the crap out of this architecture in the just over a year they have left before Intel's 7nm EUV (if we believe early 2021 as the launch date) once Intel also has a smaller process node they will be able to stomp AMD again.

They will need more than a tweak, probably a new architecture or a bigger advancement in the current one to catch up. And 1H 2021 seems too optimistic for any volume production and they'll be competing with Zen 3 and much larger TSMC volume capacity at that time.

I see what you are saying about Monolithic dies, but let's be real here. Chopping up dies into smaller parts isn't rocket science. If Intel wanted to do that, they can easily do it. Probably the only reason they haven't thus far is because 14nm is so mature at this point that they don't have a major yield loss problem. I wouldn't be surprised if they also do so on their chips launched on 7nm.

There is much more to it than simple chopping. It's a complete architecture overhaul.

AMD has a real chance to stay in the game this time, and I really hope they do. Some of my most fun PC modding and building came during the Athlon era, and I love the underdog. I'm simply stating that this is by no way a done deal.

Intel has the better race car, but they are currently in a pit stop getting their tires changed. AMD really needs to make the most out of this momentary Intel stumble, before the Intel racecar comes roaring back. Thus far they have executed extraordinarily well, but if I were a gambling man, my money would still be on Intel. They have more resources, money, fabs, engineers, more everything to throw at this problem, and an already existing architecture that outperforms AMD on a per core basis on an inferior process.

I think the best we can reasonably hope for once Intel have regained their footing, (let's say in 2022), is Intel again back at the top in the performance wars, but AMD only very slightly behind. Similar to where they were when Phenom II was battling it out vs Core 2.

Realistically, due to Intel's gigantic size advantage, I don't see AMD gaining more than upper twenties marketshare even if Intel continues failing for a couple of more years.

 

[ChadD]

I think the best we can reasonably hope for once Intel have regained their footing, (let's say in 2022), is Intel again back at the top in the performance wars, but AMD only very slightly behind. Similar to where they were when Phenom II was battling it out vs Core 2.

Read an interesting rebench awhile back of sandy bridge vs FX with spectre and meltdown mitigation applied to each.

Intel remained in the lead.... but the difference between SB and FX dropped to low single digits.

We can't rewrite history of course and whatever Intel was doing they won that round... Intel has been cheating for a long while. It does make me wonder though how people would have looked at FX back at its launch if AMD was offering 8 core parts vs SB that where only 2-5% behind on IPC. (which is about what it is now when you secure SB)

 

[OrangeKhrush]

I support forced mitigation testing mandatory, leave it to the end user to decide whether or not to turn it off but reviews should be done with forced mitigation patches on. Bypassing this means computer cycles are saved not doing important checks. It is a bit like Formula One and the "Exhaust Blown Diffuser" era

 

[thesmokingman]

I support forced mitigation testing mandatory, leave it to the end user to decide whether or not to turn it off but reviews should be done with forced mitigation patches on. Bypassing this means computer cycles are saved not doing important checks. It is a bit like Formula One and the "Exhaust Blown Diffuser" era

Yea, reviews should follow best practices methodology.

 

[Snowdog]

I support forced mitigation testing mandatory, leave it to the end user to decide whether or not to turn it off but reviews should be done with forced mitigation patches on. Bypassing this means computer cycles are saved not doing important checks. It is a bit like Formula One and the "Exhaust Blown Diffuser" era

This is red herring. All the top review sites test all all current patches installed.

 

[IdiotInCharge]

We can't rewrite history of course and whatever Intel was doing they won that round... Intel has been cheating for a long while. It does make me wonder though how people would have looked at FX back at its launch if AMD was offering 8 core parts vs SB that where only 2-5% behind on IPC. (which is about what it is now when you secure SB)

These mitigations apply largely to virtualization -- which was a big thing in enterprise at the time, but very not much so for consumers, and still isn't. These issues apply when they're exploitable.

I support forced mitigation testing mandatory, leave it to the end user to decide whether or not to turn it off but reviews should be done with forced mitigation patches on. Bypassing this means computer cycles are saved not doing important checks. It is a bit like Formula One and the "Exhaust Blown Diffuser" era

For applicable use cases, unquestionably.

I think you're overestimating the influence of node over architecture. AMD's GPUs show that it takes more than a process advantage to gain efficiency.

AMD GPUs are almost never tuned for gaming, which is the primary consumer use of GPU performance. They're tuned for compute. AMD is much more interested in selling Firepros to the enterprise and Radeons with strong compute to Apple for Mac Pros than they are gaming cards where they simply have never really competed with Nvidia, and thus have to sell at lower prices and accept lower margins.

Maybe Big Navi will save us!

 

[Snowdog]

And in lightly threaded loads Intel is still king. By ever smaller margins, but they still hold that lead. And what's more is that they are doing this on their 14+++++++ architecture, not benefitting from the advantages of a smaller node.

In lightly threaded loads, Intel doesn't have a process disadvantage, it has a process advantage, because they clock higher. Power isn't an issue in lightly threaded loads.

Unless AMD pulls an upset and really executes on their version of tick-tock and optimizes the crap out of this architecture in the just over a year they have left before Intel's 7nm EUV (if we believe early 2021 as the launch date) once Intel also has a smaller process node they will be able to stomp AMD again.

AMD has kind of already pulled the upset, and they appear to be tick-tocking away from Intel. AMD looks to have Zen 3 arriving before Intel has anything new of note.

I think the best we can reasonably hope for once Intel have regained their footing, (let's say in 2022), is Intel again back at the top in the performance wars, but AMD only very slightly behind. Similar to where they were when Phenom II was battling it out vs Core 2.

Intel is going to take years digging out from it's process mess. They can't meet demand at 14nm, partially because they won't build new 14nm fabs, because 14nm is supposed to be dead by now, but what kind of huge mess is Intel going to be in migrating when they finally get something usable? 10nm barely clocks fast enough to compete in laptops.

I don't see any reason to automatically expect Intel would have the better core in 2022. It's a complete crap-shoot which core is better then. Even if it's Intel, they will probably be extremely production constrained on the 10nm++/7nm process that finally turns around the process mess. They will then have a massive amount of new Fabs to build/convert for the new process.

Here is how much AMD has upset the apple cart. Latest stats from the German retail outlet that tracks CPU sales. Intel used to be the enthusiast choice and crush AMD, like AMD is crushing Intel now with enthusiasts:

 

[ChadD]

This is red herring. All the top review sites test all all current patches installed.

This has improved over the years. I wouldn't say it was always the case.

Unfortunately for consumers and AMD... most of the major performance killing stuff that has been discovered the last few years was unknown and therefore unpatched at launch. Considering Intels track record... I would simply assume that whatever the launch benchmark says for anything they make, subtract 10% and assume that will be the real numbers 1 year post launch. lol I would say I was joking... but that is exactly the case for everything they have launched for 10 or so years now. (frankly a few skus got hit a lot harder then 10%)

https://www.phoronix.com/scan.php?page=article&item=sandy-fx-zombieload&num=8

I know we are talking about a lot of years of mitigation in this specific case I am referencing here....

Still we all remember how bad the launch reviews where for FX... and how the community basically tore AMD a new one and abandoned them for years.

I imagine the reaction to say FX-8370 would have been A LOT different if the narrative was 10-15% worse IPC, double the cores 50% the price vs an i7. Wouldn't have been a home run no doubt but AMD wouldn't have been pilloried like they where either.

 

[ChadD]

These mitigations apply largely to virtualization -- which was a big thing in enterprise at the time, but very not much so for consumers, and still isn't. These issues apply when they're exploitable.

That is the narrative Intel would love everyone to believe. Side Channel attacks only apply to Virtualization. To bad the truth is that is BS spin. Virtualization makes cloned channel attacks far easier.... but side channel attacks work very well against any form of encryption which would include the type used by things like banks paypal ect ect. Considering how less secure things where 10 years back we should all be very thankful that it doesn't seem like to many people knew Intels dirty secret back then. I have zero doubt the type of spying that snowden outed was no doubt made possible by side channel infiltration of supposedly secure systems.

 

[IdiotInCharge]

AMD has kind of already pulled the upset, and they appear to be tick-tocking away from Intel. AMD looks to have Zen 3 arriving before Intel has anything new of note.

Key word: appearing

Consider that Intel hasn't stopped their tick-tock cadence, but rather had a delay. Which means that their technologies further in the pipe are still coming.

Intel is going to take years digging out from it's process mess. They can't meet demand at 14nm, partially because they won't build new 14nm fabs, because 14nm is supposed to be dead by now, but what kind of huge mess is Intel going to be in migrating when they finally get something usable? 10nm barely clocks fast enough to compete in laptops.

14nm isn't supposed to be dead; it's supposed to be producing stuff other than CPUs. Chipsets and so on. Intel just doesn't turn off all of their fabs one night and then reopen them the next day to produce a different node. The reason they're having issues with 14nm yes is that they're not standing more 14nm production back up, but also because they're actually pushing out 10nm products as well as standing up 7nm.

I don't see any reason to automatically expect Intel would have the better core in 2022. It's a complete crap-shoot which core is better then. Even if it's Intel, they will probably be extremely production constrained on the 10nm++/7nm process that finally turns around the process mess. They will then have a massive amount of new Fabs to build/convert for the new process.

Intel already has 'the better core', and that core isn't nearly the best that they could do today if they had their fabs in line today.

AMD, for all the ground that they've made up, is still behind in terms of architecture design. Remember that all of AMDs recent 'wins' are against an Intel architecture that is six years old. Intel's fabs may have stumbled, but the rest of the company certainly hasn't skipped a breath.

Here is how much AMD has upset the apple cart. Latest stats from the German retail outlet that tracks CPU sales. Intel used to be the enthusiast choice and crush AMD, like AMD is crushing Intel now with enthusiasts:

Now you're talking about aftermarket sales at a single vendor in a single country.

On its own, this data is only relevant to the fraction of the market that it represents. It's statistically irrelevant to the discussion at hand. And please don't go screenshot Amazon's list that puts a CPU with < 1,000 units shipped up as 'best seller', lol.

 

[IdiotInCharge]

That is the narrative Intel would love everyone to believe. Side Channel attacks only apply to Virtualization. To bad the truth is that is BS spin. Virtualization makes cloned channel attacks far easier.... but side channel attacks work very well against any form of encryption which would include the type used by things like banks paypal ect ect.

So, not consumers? And you're not going to even hint at the defense in depth involved in those enterprises?

Considering how less secure things where 10 years back we should all be very thankful that it doesn't seem like to many people knew Intels dirty secret back then.

No one knew except Intel, who were right in classifying the possibilities as extremely remote. Guess what?

They still are. But they make for good counter-marketing!

I have zero doubt the type of spying that snowden outed was no doubt made possible by side channel infiltration of supposedly secure systems.

Well, you're the only one I know of that would rate their doubt as 'zero'. Snowden did significant damage, but the data collection he used to justify his desperate attention grab was fully vetted legally, with the biggest issues being lax oversight.

Which is basically what you'd expect when you consider the goal and the scale involved.

And they didn't need to break any hardware to do it. They just installed their own.

 

[Jim Kim]

That is the narrative Intel would love everyone to believe. Side Channel attacks only apply to Virtualization. To bad the truth is that is BS spin. Virtualization makes cloned channel attacks far easier.... but side channel attacks work very well against any form of encryption which would include the type used by things like banks paypal ect ect. Considering how less secure things where 10 years back we should all be very thankful that it doesn't seem like to many people knew Intels dirty secret back then. I have zero doubt the type of spying that snowden outed was no doubt made possible by side channel infiltration of supposedly secure systems.

I thought the patriot just had a big usb drive and started copying the illegal activities our government does everyday.
Cut and Paste stuff.
There should be statues of Edward Snowden in every town square.

 

[ChadD]

Intel already has 'the better core', and that core isn't nearly the best that they could do today if they had their fabs in line today.

AMD, for all the ground that they've made up, is still behind in terms of architecture design. Remember that all of AMDs recent 'wins' are against an Intel architecture that is six years old. Intel's fabs may have stumbled, but the rest of the company certainly hasn't skipped a breath.

Now you're talking about aftermarket sales at a single vendor in a single country.

On its own, this data is only relevant to the fraction of the market that it represents. It's statistically irrelevant to the discussion at hand. And please don't go screenshot Amazon's list that puts a CPU with < 1,000 units shipped up as 'best seller', lol.

There is nothing about the Intel core design that is any better then AMDs. AMD did update their speculation engine with Zen... and bump it again with Zen2. You could argue that Intel had an advantage there. But the Speculation engine of both companies since Zen launched is based on the exact same research paper. There are newer potentially faster more efficient designs out there now in research papers... but they aren't revolutionary more evolutionary designs.

As far as sales numbers posted by SD your not wrong that is only once country. Still I seem to remember Americans lined up for Zen2.... and I know in Canada our local store is still running out of stock on Zen2 parts. Can watch their stock online yoyo from 50+ to zero on all the mainstream skus every week.

 

[ChadD]

So, not consumers? And you're not going to even hint at the defense in depth involved in those enterprises?

No one knew except Intel, who were right in classifying the possibilities as extremely remote. Guess what?

They still are. But they make for good counter-marketing!

Well, you're the only one I know of that would rate their doubt as 'zero'. Snowden did significant damage, but the data collection he used to justify his desperate attention grab was fully vetted legally, with the biggest issues being lax oversight.

Which is basically what you'd expect when you consider the goal and the scale involved.

And they didn't need to break any hardware to do it. They just installed their own.

No side channel attacks can and are carried out against consumer devices. The narrative that side channel attacks can only be carried out on servers and virtual machines... so they only effect big companies and not everyday folk is stupid and dangerous.

As for snowden I don't care about him or what he did or didn't do. I am simply saying we all now know what the US gov was able to do to compromise consumer devices. The tools we know they used in that Era took advantage of multiple attacks one of which was a side channel based cache attack. Intel was clearly NOT the only people in the know about their design choice.

 

[IdiotInCharge]

There is nothing about the Intel core design that is any better then AMDs. AMD did update their speculation engine with Zen... and bump it again with Zen2. You could argue that Intel had an advantage there. But the Speculation engine of both companies since Zen launched is based on the exact same research paper. There are newer potentially faster more efficient designs out there now in research papers... but they aren't revolutionary more evolutionary designs.

-- clocks higher
-- still higher IPC
-- still on 14nm
-- still six years old, unless you're talking about Ice Lake (which isn't their latest design, just latest released)

As far as sales numbers posted by SD your not wrong that is only once country. Still I seem to remember Americans lined up for Zen2.... and I know in Canada our local store is still running out of stock on Zen2 parts. Can watch their stock online yoyo from 50+ to zero on all the mainstream skus every week.

The bigger issue is that this is all talking about retail parts. Retail sales worldwide are still a drop in the bucket. AMD could outsell Intel 100:1 at retail and still not gain notable marketshare.

It's a reminder that while enthusiasts are vocal, the purchasing power we represent doesn't even rise to the level of a rounding error.

 

[ChadD]

I thought the patriot just had a big usb drive and started copying the illegal activities our government does everyday.
Cut and Paste stuff.
There should be statues of Edward Snowden in every town square.

I didn't explain my point very well. Snowden is not the point. The point was the projects he outed. The mass data collection of everyone. Yes a lot of that came from server cloud data... but they used multiple tools. One of the tools they had was capable of breaking consumer level encryption tech via side channel exploits. Made possible by Intels either stupidity or complicity. Not sure we will ever know which.

 

[IdiotInCharge]

No side channel attacks can and are carried out against consumer devices. The narrative that side channel attacks can only be carried out on servers and virtual machines... so they only effect servers is stupid and dangerous.

Agreed -- I'm not talking about the devices, but the applications.

Demonstrated vulnerabilities that have been patched, that would still be exploitable through the full stack. With respect to consumers, this hasn't been shown; with respect to the enterprise, the same, though the mitigations for enterprise applications and hardware do incur a greater performance penalty.

Intel was clearly NOT the only people in the know about their design choice.

No, they were the only ones with public knowledge until security researchers released their papers. The spooks know what the spooks know.

 

[ChadD]

The bigger issue is that this is all talking about retail parts. Retail sales worldwide are still a drop in the bucket. AMD could outsell Intel 100:1 at retail and still not gain notable marketshare.

It's a reminder that while enthusiasts are vocal, the purchasing power we represent doesn't even rise to the level of a rounding error.

That is fair... OEM sales will take a bit to trickle up. Having said that I just picked up a R5 based laptop for my mother. I see your point and ya while I was shopping there is still a lot more Intel stuff around. Having said that I did see a good number of mid range and even gaming laptops with zen in them. AMD is picking up some wins no doubt.

Point taken though. I agree with you.

 

[ManofGod]

-- clocks higher
-- still higher IPC
-- still on 14nm
-- still six years old, unless you're talking about Ice Lake (which isn't their latest design, just latest released)



The bigger issue is that this is all talking about retail parts. Retail sales worldwide are still a drop in the bucket. AMD could outsell Intel 100:1 at retail and still not gain notable marketshare.

It's a reminder that while enthusiasts are vocal, the purchasing power we represent doesn't even rise to the level of a rounding error.

The truth is, all I am seeing is a bunch of wordy responses in the hope that they can make plausible excuses for Intel and bury the message that AMD is seriously kicking some butt. But the reality is that 2.5 or so years ago, AMD was in serious financial and market share trouble and now they are not. The good thing is, the facts will speak for themselves and those who try to bury message are only embarrassing themselves.

As for clock speed, it is the only advantage they have now and that will not hold out much longer. Clock for clock, AMD is actually ahead now. Oh, and you mentioned AMD graphics at one point, I believe, and even in that realm, they are doing much, much better. Competition is good, unless you own Intel Stock. Now, if you owned AMD stock earlier, you are definitely making much more money now.

 

[Zarathustra[H]]

In lightly threaded loads, Intel doesn't have a process disadvantage, it has a process advantage, because they clock higher. Power isn't an issue in lightly threaded loads.

I guess that's why chipmakers never shrunk process nodes in the single core era?

Smaller node produces less heat, and requires less voltage.

Less heat means you can crank up voltage and gain higher clocks OR build a more complex core with higher IPC inside the same thermal envelope.

Don't mistake AMD's difficulties with hitting the same clocks as Intel for an inherent problem with smaller processes.

 

[IdiotInCharge]

The truth is, all I am seeing is a bunch of wordy responses in the hope that they can make plausible excuses for Intel and bury the message that AMD is seriously kicking some butt.

If you could get actual shipment numbers -- you can't, but if you could -- you could show a difference. What we do have is stock price and so on, as well as financials, and both companies are doing well because demand is high. Of course, the topic isn't 'is AMD doing well'.

The good thing is, the facts will speak for themselves and those who try to bury message or only embarrassing themselves.

So sensitive

Competition is good, unless you own Intel Stock. Now, if you owned AMD stock earlier, you are definitely making much more money now.

If you own stock from either, you're not doing bad -- but Intel still has a better market outlook. AMD is downright precarious especially when you consider that they're essentially releasing tweaks of an architecture that only competes with their competitors six year old designs, and that they depend on others for manufacturing volume.

If I were looking at moving money in stock, it'd be from AMD to Intel.

 

[Zarathustra[H]]

There is nothing about the Intel core design that is any better then AMDs. AMD did update their speculation engine with Zen... and bump it again with Zen2. You could argue that Intel had an advantage there. But the Speculation engine of both companies since Zen launched is based on the exact same research paper. There are newer potentially faster more efficient designs out there now in research papers... but they aren't revolutionary more evolutionary designs.

As far as sales numbers posted by SD your not wrong that is only once country. Still I seem to remember Americans lined up for Zen2.... and I know in Canada our local store is still running out of stock on Zen2 parts. Can watch their stock online yoyo from 50+ to zero on all the mainstream skus every week.

Yet Intel is still beating AMD in lightly threaded loads on a process node twice the size of AMD...

That to me is the only fact that matters in this discussion, and evidence that their architecture is leaps and bounds ahead of even Zen2. All they are missing is a good process, which they will have in 2021.

 

[Mega6]

A reality check. Amd was at 2 and now close to 40. Amd share price will surpass intel soon.

 

[IdiotInCharge]

A reality check. Amd was at 2 and now close to 40. Amd share price will surpass intel soon.

Further reality check: considering their history, would you put your life savings into AMD stock, knowing what it will take for them to even remain competitive?

You can scream 'yes' on an internet forum, but we all know what you'd really do.