Really? Remember Bulldozer? That wasn't that long ago and AMD came back in essentially one architecture and Intel has a lot more resources at their disposal.
Personally I just hope 2024/25 is enough time for AMD to gain enough market share and reach more of a market parity with Intel so their aren't perpetually at a disadvantage. Its super hard to compete technologically when your main competitor operates at scale several times what you do.
Bulldozer was a bad architecture and it succeeded in getting process node upgrades. It's not relevant to Intel's track record on not getting new process node upgrades, which is what derailed Tick-Tock.
Also, Zen was aided by Jim Keller. Keller left AMD for Intel, where he basically butted heads with their management and left without Intel making good use of him.
Most likely the design of Golden Cove (the "big" cores of Alder Lake) was largely "locked" when Jim Keller joined Intel - at most it required optimizations and "trimming". Back when Keller's hiring was announced Intel also published a LinkedIn ad asking for microprocessor engineers for the design of "Ocean Cove", which was first mentioned by name then.
The ad was soon pulled (it appears that the Ocean Cove name was leaked prematurely by mistake) but since it was published just a few days after (or before, cannot quite recall) Intel announced Keller was joining the company the link between the two was unstated but obvious.
Jim Keller is not in the habit of optimizing designs of others; his expertise is in designing / leading the design of *new* designs. When Keller was hired by Intel the design of Ocean Cove was just starting but the design of Golden Cove was largely complete. While I have no proof of this, based on Keller's track record it stands to reason that he would clearly want to work on Ocean Cove, not polish Golden Cove; whether his presumed request was granted by Intel's upper management is unknown though.
That makes no sense, you don't lock design more than 3 years ahead of release. Golden Cove most likely just started (a couple of months at most) back when Jim joined Intel.
It takes roughly 4 years to bring a new design to life. By early 2018 (T minus 3.5 years), Alder Lake would have just come off the initial phase (design specification) and moved to the next phase - behavioral design (including microarchitecture and logic) not long ago.
Of course, I've got no idea what went on in AMD's design rooms, but I see this recurring idea that Jim Keller's return saved AMD. I'm not trying to put down Mr. Keller, who is highly respected and certainly a genius (lead architect of the K8 and excellent credits elsewhere); but the thing is, isn't it more likely that he was just one of many minds pouring their ideas into Zen? Wasn't Michael Clark the chief architect of Zen? Didn't hundreds of engineers work on it? If he had a big hand in it, we are more than willing to give him due credit, but if not, it is downplaying the work of the other engineers and designers.
Jim Keller led designs on Athlon 64, helped build x86-64 and dominated in performance. Went over to Apple and helped change their whole lineup with the high performing A4. He returned when AMD was at its worst, then Zen ships out and AMD is back on the map.
Either he is the luckiest engineer alive, dipping in right as multiple different companies were about to make a breakthrough, or he is a critical part in the success of these architectures. It's obviously a team effort and he is not solely responsible but no one just wanders around the industry and continuously hits home runs.
Given lead times for CPUs and how A6/A7 were Apple's first in-house 32-bit and 64-bit cores, I honestly think Keller helped them with the early designs of those chips, rather than the ARM-licensed Cortex cores in A4 and A5. It kinda shows in that Apple's in-house ARM is going toe-to-toe with x86 at this point.
AMD made a claim of a 40% IPC uplift in a single generation on a brand new node with a massive improvement in efficiency, which is pretty damn ambitious.
Sunny cove had an IPC lead over Zen 2, so it is likely that Golden Cove will have an IPC lead. over Zen 3. If this pattern follow in 3-4 years Intel will have a new microarchitecture leading Zen 4, and possibly releasing their next one that leads Zen 5.
Intel has performance now, but tremendously lacks efficiency with 14nm. If 7nm is as good as Intel claims it is, then it is likely that Intel could be back in the efficiency and IPC lead (also if 7nm is that good then Intel can hit 5+ Ghz, giving them the single thread performance crown). With "tiles" Intel could combat AMD's chiplets, taking the multi-threaded crown.
With this in mind, I don't think Intel is as ambitious as AMD for Zen. Remember AMD went from a waste of silicon to viable to leader. Intel plans to go from viable to leader, which isn't that hard for a company of their scope (as long as there is good management).
"If 7nm is as good as Intel claims it is" That's the big "if". They claimed 10nm was great long past the point where it was clear it wasn't even hitting their claimed density, let alone performance. That's no reason to assume 7nm is as bad as 10nm, but it does give me pause on taking their word on how good it is.
They can do 5Ghz on 10nm now, albeit at a rather ugly point on the voltage/power curve. I'll be interested to see whether that gets better or worse with 7nm.
They screwed up on EUV timing in both cases. Their new CEO seems more willing and interested in branching out and using standardized tools, rather than going all custom while making crazy speculative goals like they did under people like Krzanich. 7nm will depend a lot on how quickly they can adopt EUV and if they can even get enough EUV equipment to hit their volume targets. I'd say its doable but not under their old corporate culture, going to need a lot of money and sleepless nights.
AMD was coming from a garbage architecture on a garbage node (28nm bulk, density-optimized not performance-optimized, with the sloppy FM socket standard).
Not difficult to make a massive improvement from that.
Had AMD aggressively improved upon Phenom, rather than releasing Bulldozer (which had worse, not better, IPC) — and had the company not decided to leave the architecture blowing in the wind (an architecture that was far too slow in 2011) until Zen it wouldn't have been able to make such a great improvement. Remember, not only did AMD refuse to improve upon Piledriver (basically Bulldozer with minor tweaks, including the breaking of AVX) for the 'performance desktop' and supercomputer markets, it cut the number of cores way down for Steamroller and Excavator. AND, it cut cache size to the bone. All of that was to make them cheap to produce, not performant. AMD could have taken the improvements from Steamroller and Excavator, married them to the 32nm SOI node rather than 28nm dense-library and the anemic FM socket, and further improved the 'construction core/CMT' design. It didn't do that. The company left it completely stagnant, an architecture that was mostly dead on arrival in 2011, as it wasn't fast enough versus Phenom and was destroyed by Sandy Bridge quads that had far fewer transistors and much less power consumption.
AMD would have only looking better with Zen had the previous architecture been something from VIA.
It reduced area, yes, but more importantly, dropped power consumption. And if I had been AMD, I think I would have done exactly the same, concerning Steamroller and Excavator, knowing that Zen was almost ready to knock Intel from its high horse.
The bottom line is that the architecture was designed for high clocks to compensate for the low IPC.
That AMD chose to go super-cheap (28nm density library, 2 FPU cores, flaccid socket standards) is not the point. It's just historical anecdote, although it also helped to make Zen seem more impressive than it would have had AMD bothered to offer anything after Piledriver in the performance space.
"The bottom line is that the architecture was designed for high clocks to compensate for the low IPC."
True, can't disagree with that.
"it also helped to make Zen seem more impressive"
Agree there too. I recall quite strongly that when Zen's IPC was being unveiled in 2016/7, you know that 52%, I thought, "Well, that's fantastic, but the comparison is against Bulldozer, so obviously the value will be high." Having said that, I still believe that Zen was an impressive product. In a single iteration, you've got a design from scratch, reaching just short of Skylake IPC, the latter going all the way back to P6 in the 90s. How many iterations of polish is that? Zen was impressive to cover the ground in one stroke, and Zen 2, as I expected, being the first touching up, surpassed Skylake just like that.
Phenom2 couldn't really scale further and on servers AMD had MC that was great with highly parallel workloads and not much else; so they were competitive but they were clearly underdogs and sliding behind. So, much before launch, AMD was bragging about how BD will drive over Intel in everything and offer tremendous gains over their MC. It was a spectacular disaster, significantly worse than promised. Zen launched to way less fanfare and was a much better chip than promised.
Obviously, there is no indication history will repeat itself, but if you want to draw any parallels from the history, it resembles era before BD launch, not before Zen.
Zen was exactly as promised. They laid the whole thing out at hotchips. Fanfare was just low due to AMD's low budget for marketing and years of bulldozer disappointments. Fanfare doesn't really seem that high for intel either, plenty of people are rightfully skeptical since 10nm is something like 5 years late now.
• slower than Intel (for gaming at least) • quite limited in upper-end clockrate • more latency than Intel, due to chiplets and design • unstable with 3200-speed RAM sometimes
It was a tremendous improvement but also imperfect. One of the drawbacks, for consumers, is that it was designed to improve profit margin, via keeping the number of cores low in the chiplets, to maximize yield. That increases latency.
Not in terminology or exact philosophy, but the idea crudely existed back then. The Pentium, starting off at 0.8 um, was produced on a range of processes, probably 0.35 um being the most common. Pentium Pro on both 0.35 and 0.5 um. By the Pentium II's time it was starting to be differentiated better: Klamath, carrying the older process, was 0.35 and Deschutes 0.25 um. Pentium III Katmai started off at 0.25 and Coppermine shrunk that to 0.18 um.
Pentium 4 Willamette took off at 0.18 um/180 nm, and Northwood went down to 130. Prescott, notably, alterered both microarchitecture and process (90 nm), but Cedar Mill was just a 65 nm die shrink; and Conroe, coming out shortly afterwards, launched at 65 nm.
Prescott was a giant band aid fix to the Pentium 4 choking due to lack of FSB & Memory bandwidth as core clock speeds rose. Intel just piled on a tons of cache memory hoping it would hid the issue without have to engineer a new chipset with a higher bus speed.
I had a Pentium 4 HT 3.0 at the time, and sold it a purchased a Pentium 4 HT 2.8C and matched it with Corsair Dominator 2 1066 memory, I overclocked the FSB to 1066MT/s from 800MT/s and lower the CPU multiplier from 14x to 13x, giving me a clock speed just under 3.5ghz , my flat mate at the time had a Pentium 4 HT 560 (Prescott 3.6ghz) my system was much faster than his, despite his CPU have twice the cache memory and nearly a 150hz clock speed advantage.
The band aid didn't work, the CPU ended up almost 50% cache memory by die size, which cranked the power consumption and heat output to 11. The real fix was to get rid of the system bottle neck.
Even if Intel was able to become the CPU performance leader I'd never allow the purchase of any of their products by our company due to Intel's documented history of crime and anti-trust law violations in their efforts to prevent a free, level marketplace for consumers. Intel's evil management mentality almost eliminated AMD and choice for all consumers. Intel could not pay me enough to use their products and they certainly don't deserve my business. Voting with you wallet can prevent exploitation.
that's like saying "i don't want to live in usa because half the states legalized weed". continue buying your underperforming Chinese-built AMD while I stick to quality Intel.
How is AMD "Chinese-built"? The company is an American company who designs their microarchitectures in America. Their foundry is TSMC, a Taiwanese company, with fabs in Taiwan and America. Also, right now AMD is "quality" given that they are more performant, less power hungry, and more feature rich.
1) weed is good for you. may be try it sometime so you are not so angry all the time.
2) TSMC is Taiwanese, not Chinese. please educate yourself on the difference.
3) "quality" Intel? is that why Intel cancelled their overclocking warranty? are Apple and Qualcomm silicon not "quality" because TSMC also makes those?
Totally Agree Techie2, all my many compute devices are all ether AMD silicon or Apple silicon currently. My last 6 PC's built for me, my wife and my HTPC are all Ryzen and my wife has a Ryzen Laptop as well. AIso just got an M1 Mac for myself which is really good. not an Intel CPU in sight.
You're in luck. I believe the official PR material style guide was leaked in the huge data dump. So actually, someone could make their own version very easily.
Mobile also got Kaby Lake-R because they needed an emergency response to Zen Mobile. So depending on how you count it, there's up to 5 "optimizations" in there
I can't wait for next financial results, only to hear Pat's high attitude statements :) Tick-tock won't help to gain leadership products. AMD is 'firing on all cylinders', what means they can jump into new process and new architecture in single shot, Zen 2 and upcoming Zen 4 are prime example of that, while Zen 3 is 'only' new architecture.
Amen. And TSMC will grow capacity for AMD into late this year and next. If Intel executes on their "dreams" perfectly, then they have a chance to get back in the game. If not, AMD will stay ahead and the lead will become insurmountable by 2024/2025.
I've always found the assumption that AMD's lead will be insurmountable fascinating. AMD was much, much farther behind Intel with the launch of Bulldozer and its subsequent products than Intel is behind AMD now. Have people forgotten the 2500k being comparable in MT to Bulldozer with twice as many cores?
Intel's core design is still reasonably good; if Golden Cove has a nice improvement, they'll likely have a higher IPC design than AMD. If Rocket Lake needed twice as many cores as Zen 3 to match it, that'd be one thing, but it's really only at a major disadvantage in one area, efficiency.
The most common response I see to this is that AMD is firing on all cylinders now, but while Intel wasn't making the kind of improvements AMD is making now, they were still turning out good products several years after Bulldozer came out.
Yeah, I entirely agree with you that AMD are in no way "unassailable" at present - especially not by the juggernaut that is Intel. When people will buy your products in huge quantities whether good or bad, you can guarantee the revenue streams required to put good designs together in the current era of CPU design.
Yes. My belief is that Intel will be back on track one of these years, but it won't be a repeat of the Phenom/Bulldozer era. AMD, having learnt the hard way, is fighting like Goku against Frieza, and I think (I hope?) they'll keep it up. Just watch out for dirty tricks, AMD! Most likely, once Intel's process troubles are mended, we'll see Ryzen and Core toe to toe, each leading slightly at different times. "Core leads, 2030. Ryzen back on top, 2031." Of course, ARM could muddle this nice picture but we'll see what happens.
I think we're on the verge of that now, quantum tunnelling effects kicking in at <= 3 nm. I'd like to see what TSMC, Intel, AMD, and Apple do once the limit is reached. As for quantum computing, I sometimes wonder if these computers are really doing anything more than just being indeterministic---even a classical computer can simulate a QC, I believe, but orders of magnitude slower. At any rate, I need to get a better grounding on this topic.
AMD embraced being the "value option" once long ago, I think it was like 2013? They didn't get on top with Zen (really Zen 2 and 3) by embracing the value option though. AMD ultimately didn't settle for mediocrity, and we shouldn't really want Intel to either. AMD is probably driven right now by the knowledge that Intel *can* put out capable products and execute a capable roadmap - the fact that they haven't for the last few years might inform expectations, but neither AMD nor us should assume Intel will never get a competent roadmap executed again.
"Intel stated that its 7nm process is now comfortably on track to deliver Meteor Lake, a client CPU using tiles/chiplets, in 2023, however we are likely looking to a 7nm variant or even external processes for a 2024/2025 product."
Meteor Lake (on a brand new 7nm process node with brand new Ocean Cove "big" cores - the tiles thing is also brand new) will almost certainly be released on Q1 2024 at the earliest, thus the 2024/2025 timeline for ‘unquestioned CPU leadership performance’. If Ocean Cove cores on a new very efficient 7nm process node that is slightly denser than TSMC's 5nm process node (at the maximum density variant of both) cannot beat Zen 5 -they will almost certainly beat Zen 4 but I strongly doubt Ocean Cove will compete with that- then nothing can.
I expect Intel *will* fab 7nm parts in 2023 (though nothing in 2022), just not Meteor Lake. They never fab their premiere desktop parts on new nodes. They tend to fab mobile parts with up to 4 cores and up to 15 - 20W TDPs, so that the customers themselves do the final testing and debugging of the new node "in the wild". They might also fab some 7nm GPUs in 2023, though not the enthusiast or server ones. Up to mainstream class tops.
AMD doesn't have the resources to build a new microarchitecture every year. If Zen 4 is 2022, then the next major microarchitecture update would be 2024. If Ocean/Redwood cove is 2023, then Intel can bring a new microarchitecture 2025, which would beat Zen5+/6 (whatever AMD calls it). Remember that the microarchitecture for 7nm would be complete in 2022/2023, so even if Ocean Cove is delayed to 2024, it's possible for Intel to tock for 2025. I do agree that it is unlikely that Intel beats AMD's post Zen 4 big architecture in 2024, but 2025 should be competitive.
Aren't Intel supposed to be putting their Ponte Vecchio HPC GPU out as one of the (if not the) first products on 7nm?
My best guess on why is that, after the abject failure and cost-soak of trying to manufacture tiny dual-core mobile chips on OG 10nm and failing, they've gone for a strategy that will probably pay for itself no matter how few GPU tiles they can yield from each wafer.
The big tell will be how fast they can move from pushing out a trickle of server-grade parts for a supercomputer project to ramping consumer parts that actually require decent yields to be profitable.
I want what Gelsinger is smoking right now. What a comedy show... Tick Tock ? As much as I wanted to see it come. Its simply not possible now. Intel squandered their 5 year advantage of Fab Lithography edge long time back. Their true age was FinFet Haswell and 14nm debut in 2014. That's 6 years ago we are still in 14nm++ yes its very much optimized but its over for them.
This guy saying x86 licensing, IP like EMIB (AMD G processors does anyone remember ?) And Foundry business. What the fuck ? Intel backport failed with RKL on Cypress Cove and 14nm++ now they want all of that licensing money by selling out their top x86.
Next this guy says High performance in 2024, by that Time even Apple might ship a server ARM alternative if we want to use ladders in sky type of talk. To be realistic AMD in 2024 is Zen5 probably on TSMC 3nm. And Intel doesn't even have 7nm by that time. Add Samsung GAAFET 3nm so who wants 7nm ? ARM processors will not for sure. Apple won't either. Altera ? Graviton ? Or that MS ARM server CPU ? Haha.
Man this guy is really all talk. First show us a successor for X299 then we talk.
Haswell was neither the first 14nm processor, nor was it the first finfet
IVB was 22nm and finfet debut, and Broadwell was 14nm debut. That Intel initially had so many problems with 14nm at high power was an ill omen for their 10nm process (Broadwell-C was the only official desktop variant, and it was capped at 65w)
Yeah, I didn't mention Broadwell which is a mistake by omission, I was in bed. BWL was 14nm debut yes and 22nm FinFet on IVB which I completely forgot.
14nm was the early warning signs that all was not well in paradise - it arrived later than expected and the initial results were underwhelming.
Sometimes I wonder how good 22nm would have looked if they'd kept improving it like they eventually did with 14nm, rather than trying to kick it out the door ASAP.
" Intel stated that its 7nm process is now comfortably on track to deliver Meteor Lake, a client CPU using tiles/chiplets, in 2023...."
Let me check the TSMC process node roadmap as reported here on Anandtech real quick.. TSMC expects to enter full volume production on 3nm in 22H2.
Auch Intel. Even if we assume the Intel 7nm is comparable to one form of TSMC 5nm that still means they're at least a full process node behind in 2023.
"TSMC expects to enter full volume production on 3nm in 22H2."
With only enough supply for Apple and possibly Huawei. There won't be capacity available from AMD until at least 2023.
You are right they are still a node behind even if 7nm comes out in 2023. There's a lot of catching up for Intel to do if they want to take back the processing lead.
Well the show is there, that was needed off course from CEO perspective, the show must go on. Now the facts, the bleeding is hurting a lot....and will continue for a very long time.
Bad 10nm process, bad core re-designs, bad roadmap. Desktop is waiting for Alderlake which is again not scale the Core count, Ice lake is hardly improving anything (few cores more and finally more memory lanes) and Eagle is another year out.
So its 2023-2024 to get back on track with no idea where the competition will be at that point.
Lost apple as a customer by that time in all the segments, lost at least 30% marketshare by that time to AMD. Have no footage in mobile phone devices, GPU has yet to be seen if it will any change of taking off....
Adding 2 fabs? they will have spare capacity by the time they are ready... but they will always sell lots of stuff even if it is bad since the OEM market is so broken ECO system and R&D sponsoring.
"Tiles by this definition are more costly to implement than chiplets, and have additional thermal considerations by having high-powered silicon close together"
Through-substrate interconnects have their own thermal issues: both ends of every link require drivers, which are pure power waste. The more links (i.e. the more chiplets you split the design into) the more power is spent for no useful output: https://www.anandtech.com/show/13124/the-amd-threa... . Inter-die (EMIB, Foveros, etc) routing has minimal overhead over routing over the same area monolithic die. While stacked dies complicate internal thermal management (Z- conduction path), they also mean total package thermal management is simpler (reduced total thermal power, smaller physical package).
"What would you call the more than 40 pieces in the xe-hpc?"
Tiles.
Chiplet = interconnect through substrate, requires TX/RX at either end of link and protocol translation (die <> link <> die). Power per bit high (though lower than through PCB).
Tile = interconnect through silicon, does not require dedicated drivers and may or may not not have specific inter-tile protocol (protocol more for interchangeability when desired, e.g. HBM, otherwise tiles can be treated as if silicon were monolithic). Power per bit low (though higher than monolithic silicon due to TSV resistance).
I think Pat said a lot of great things, but talk is cheap. Intel did not voluntarily stop their Tick Tock cycle. They were forced to because their could not deliver 10nm on time. So unless him joining Intel can miraculously overcome all these obstacles, then I will buy his marketing. 2024/25 is really not that far away when you consider the fact that it takes many years to build a quality CPU from scratch. Moreover without their fab advantage, Intel needs to work a lot harder to keep themselves competitive. So looking forward to see if Pat can bring these supposed miracles to turn this oversized “lorry” around in 4 years.
There was some announced plan to simplify the design rules so they could push out the decision on process. Perhaps they discovered this is not so easy, or it is still a work in progress.
One reason why Intel found it hard to shrink their node was the massive increases in density. 14nm was 2.7x as dense as 22nm+, and 10nm was 2.7x as dense as 14nm++. Most foundries go for 1.6x to 2x density improvements per major node shrink. Being less aggressive on the density improvements would allow for Intel to tick more frequently.
It took the "tick-tick-tick-tock" joke to be at least half the age of the "tick-tock" model for Intel to even acknowledge that tick-tock is currently a dream.
The first step to fixing a problem is admitting you have one. I hope to see less bullshit and more competition.
"One of Gelsinger’s mantras seems to be that unquestioned leadership products bring unquestioned leadership margins for those products, and for Intel to execute, it needs to return to its days of old."
So I guess this Gelsinger guy believes that AMD will simply stand still and watch them pass by. 🤦🏽♂️ As the saying goes: you only get one! 😉
https://www.forbes.com/sites/patrickmoorhead/2021/... Perhaps far more important than previous link, but I might be biased. I've been following patrick since, jeez, never mind, I'm old. Let's just say decades and leave it at that. He's #1. Just saying. Make of those two links what you will and remember Anandtech was (is? don't come here much today) an AMD portal site. ;)
"remember Anandtech was (is? don't come here much today) an AMD portal site" You seem to be casually referring to when they had a specific AMD section for AMD-sponsored news. Do you see that here anymore?
It's been good not seeing your posts around here. Don't let us persuade you not to stay away!
He could have added 'and we're going to make overclocking really really good again, too'.
After all, if one can magically restore the cadence with which process nodes can shrink to be what it used to be (ignoring those pesky laws of physics), why not add that symptom of that cadence?
I am far from an expert in node tech but I recall the statements from experts, like:
• The complexity goes way up as the nodes shrink. This manifests as 'design rules'. That means added cost (financial and time) and difficulty. More and more design rules appear with each node shrink. This obviously means that unless the company becomes bigger and is able to execute all of this added complexity just as quickly as it could execute the less complexity before, things will slow down.
• Silicon is marching toward the point of returns so diminished that a different material will have to be used. This angle on the problem is more speculative but we have seen the death of overclocking, the spewing of cores/chiplets as a kludge, the lack of progress past 5 GHz, increasingly power-hogging parts requiring now only solder but thinner substrates, etc.
I am not sure if tick tock is still feasible when its going to be even more challenging shrinking from 7nm to 5nm, so on and so forth. They already struggled with 10nm and with the experience gained, still struggled to deliver 7nm in 3 years, so I don't expect any miracle again with the new CEO. In addition, considering that AMD's Zen 2 and Zen 3 are utilizing the same 7nm, and introduced performance improvements more on the "tock" cycle in both instances, I think this tick-tock strategy may likely struggle.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
109 Comments
Back to Article
atomicWAR - Tuesday, March 23, 2021 - link
I would love to see that happen. But with Intel's track record as of late. I'll believe it when I see it.Operandi - Tuesday, March 23, 2021 - link
Really? Remember Bulldozer? That wasn't that long ago and AMD came back in essentially one architecture and Intel has a lot more resources at their disposal.Personally I just hope 2024/25 is enough time for AMD to gain enough market share and reach more of a market parity with Intel so their aren't perpetually at a disadvantage. Its super hard to compete technologically when your main competitor operates at scale several times what you do.
Gigaplex - Tuesday, March 23, 2021 - link
Bulldozer was a bad architecture and it succeeded in getting process node upgrades. It's not relevant to Intel's track record on not getting new process node upgrades, which is what derailed Tick-Tock.Also, Zen was aided by Jim Keller. Keller left AMD for Intel, where he basically butted heads with their management and left without Intel making good use of him.
drothgery - Tuesday, March 23, 2021 - link
I'd read somewhere that Adler Lake has a lot of Keller's fingerprints on it, so we'll see there. He left AMD before Zen launched, too.silencer12 - Wednesday, March 24, 2021 - link
Keller had a 3 year contract with AMD to develop the "zen" architecture. His contract expired and thus went elsewhere.Santoval - Wednesday, March 24, 2021 - link
Most likely the design of Golden Cove (the "big" cores of Alder Lake) was largely "locked" when Jim Keller joined Intel - at most it required optimizations and "trimming". Back when Keller's hiring was announced Intel also published a LinkedIn ad asking for microprocessor engineers for the design of "Ocean Cove", which was first mentioned by name then.The ad was soon pulled (it appears that the Ocean Cove name was leaked prematurely by mistake) but since it was published just a few days after (or before, cannot quite recall) Intel announced Keller was joining the company the link between the two was unstated but obvious.
Jim Keller is not in the habit of optimizing designs of others; his expertise is in designing / leading the design of *new* designs. When Keller was hired by Intel the design of Ocean Cove was just starting but the design of Golden Cove was largely complete. While I have no proof of this, based on Keller's track record it stands to reason that he would clearly want to work on Ocean Cove, not polish Golden Cove; whether his presumed request was granted by Intel's upper management is unknown though.
dotjaz - Sunday, April 4, 2021 - link
That makes no sense, you don't lock design more than 3 years ahead of release. Golden Cove most likely just started (a couple of months at most) back when Jim joined Intel.It takes roughly 4 years to bring a new design to life. By early 2018 (T minus 3.5 years), Alder Lake would have just come off the initial phase (design specification) and moved to the next phase - behavioral design (including microarchitecture and logic) not long ago.
GeoffreyA - Wednesday, March 24, 2021 - link
"Zen was aided by Jim Keller"Of course, I've got no idea what went on in AMD's design rooms, but I see this recurring idea that Jim Keller's return saved AMD. I'm not trying to put down Mr. Keller, who is highly respected and certainly a genius (lead architect of the K8 and excellent credits elsewhere); but the thing is, isn't it more likely that he was just one of many minds pouring their ideas into Zen? Wasn't Michael Clark the chief architect of Zen? Didn't hundreds of engineers work on it? If he had a big hand in it, we are more than willing to give him due credit, but if not, it is downplaying the work of the other engineers and designers.
whatthe123 - Wednesday, March 24, 2021 - link
Jim Keller led designs on Athlon 64, helped build x86-64 and dominated in performance. Went over to Apple and helped change their whole lineup with the high performing A4. He returned when AMD was at its worst, then Zen ships out and AMD is back on the map.Either he is the luckiest engineer alive, dipping in right as multiple different companies were about to make a breakthrough, or he is a critical part in the success of these architectures. It's obviously a team effort and he is not solely responsible but no one just wanders around the industry and continuously hits home runs.
mukiex - Friday, March 26, 2021 - link
Given lead times for CPUs and how A6/A7 were Apple's first in-house 32-bit and 64-bit cores, I honestly think Keller helped them with the early designs of those chips, rather than the ARM-licensed Cortex cores in A4 and A5. It kinda shows in that Apple's in-house ARM is going toe-to-toe with x86 at this point.GeoffreyA - Friday, March 26, 2021 - link
No doubt, Mr. Keller is a brilliant mind.Oxford Guy - Wednesday, March 31, 2021 - link
'Bulldozer was a bad architecture and it succeeded in getting process node upgrades.'No. The high-performance desktop and supercomputer areas were never upgraded after Piledriver, which has 2011 stamped on the spreaders.
That was a long time to wait for Zen 1.
Railander - Tuesday, March 23, 2021 - link
to be fair, i don't remember AMD making claims as ambitious as these when they were developing zen.Otritus - Wednesday, March 24, 2021 - link
AMD made a claim of a 40% IPC uplift in a single generation on a brand new node with a massive improvement in efficiency, which is pretty damn ambitious.Sunny cove had an IPC lead over Zen 2, so it is likely that Golden Cove will have an IPC lead. over Zen 3. If this pattern follow in 3-4 years Intel will have a new microarchitecture leading Zen 4, and possibly releasing their next one that leads Zen 5.
Intel has performance now, but tremendously lacks efficiency with 14nm. If 7nm is as good as Intel claims it is, then it is likely that Intel could be back in the efficiency and IPC lead (also if 7nm is that good then Intel can hit 5+ Ghz, giving them the single thread performance crown). With "tiles" Intel could combat AMD's chiplets, taking the multi-threaded crown.
With this in mind, I don't think Intel is as ambitious as AMD for Zen. Remember AMD went from a waste of silicon to viable to leader. Intel plans to go from viable to leader, which isn't that hard for a company of their scope (as long as there is good management).
grant3 - Wednesday, March 24, 2021 - link
Great take. Despite Intel's recent failures, it still commands massive resources. It's entirely up to management.Spunjji - Friday, March 26, 2021 - link
"If 7nm is as good as Intel claims it is"That's the big "if". They claimed 10nm was great long past the point where it was clear it wasn't even hitting their claimed density, let alone performance. That's no reason to assume 7nm is as bad as 10nm, but it does give me pause on taking their word on how good it is.
They can do 5Ghz on 10nm now, albeit at a rather ugly point on the voltage/power curve. I'll be interested to see whether that gets better or worse with 7nm.
whatthe123 - Friday, March 26, 2021 - link
They screwed up on EUV timing in both cases. Their new CEO seems more willing and interested in branching out and using standardized tools, rather than going all custom while making crazy speculative goals like they did under people like Krzanich. 7nm will depend a lot on how quickly they can adopt EUV and if they can even get enough EUV equipment to hit their volume targets. I'd say its doable but not under their old corporate culture, going to need a lot of money and sleepless nights.Oxford Guy - Saturday, March 27, 2021 - link
AMD was coming from a garbage architecture on a garbage node (28nm bulk, density-optimized not performance-optimized, with the sloppy FM socket standard).Not difficult to make a massive improvement from that.
Had AMD aggressively improved upon Phenom, rather than releasing Bulldozer (which had worse, not better, IPC) — and had the company not decided to leave the architecture blowing in the wind (an architecture that was far too slow in 2011) until Zen it wouldn't have been able to make such a great improvement. Remember, not only did AMD refuse to improve upon Piledriver (basically Bulldozer with minor tweaks, including the breaking of AVX) for the 'performance desktop' and supercomputer markets, it cut the number of cores way down for Steamroller and Excavator. AND, it cut cache size to the bone. All of that was to make them cheap to produce, not performant. AMD could have taken the improvements from Steamroller and Excavator, married them to the 32nm SOI node rather than 28nm dense-library and the anemic FM socket, and further improved the 'construction core/CMT' design. It didn't do that. The company left it completely stagnant, an architecture that was mostly dead on arrival in 2011, as it wasn't fast enough versus Phenom and was destroyed by Sandy Bridge quads that had far fewer transistors and much less power consumption.
AMD would have only looking better with Zen had the previous architecture been something from VIA.
GeoffreyA - Sunday, March 28, 2021 - link
"density-optimized not performance-optimized"It reduced area, yes, but more importantly, dropped power consumption. And if I had been AMD, I think I would have done exactly the same, concerning Steamroller and Excavator, knowing that Zen was almost ready to knock Intel from its high horse.
Oxford Guy - Wednesday, March 31, 2021 - link
It also reduced maximum clocks — the thing that entire architecture hinged on.GeoffreyA - Monday, April 5, 2021 - link
That's right, but I suppose that's the tradeoff they were willing go for.Oxford Guy - Tuesday, April 6, 2021 - link
You're debating apples and oranges.The bottom line is that the architecture was designed for high clocks to compensate for the low IPC.
That AMD chose to go super-cheap (28nm density library, 2 FPU cores, flaccid socket standards) is not the point. It's just historical anecdote, although it also helped to make Zen seem more impressive than it would have had AMD bothered to offer anything after Piledriver in the performance space.
GeoffreyA - Tuesday, April 6, 2021 - link
"The bottom line is that the architecture was designed for high clocks to compensate for the low IPC."True, can't disagree with that.
"it also helped to make Zen seem more impressive"
Agree there too. I recall quite strongly that when Zen's IPC was being unveiled in 2016/7, you know that 52%, I thought, "Well, that's fantastic, but the comparison is against Bulldozer, so obviously the value will be high." Having said that, I still believe that Zen was an impressive product. In a single iteration, you've got a design from scratch, reaching just short of Skylake IPC, the latter going all the way back to P6 in the 90s. How many iterations of polish is that? Zen was impressive to cover the ground in one stroke, and Zen 2, as I expected, being the first touching up, surpassed Skylake just like that.
GeoffreyA - Wednesday, March 24, 2021 - link
"Remember Bulldozer?"Exactly. Never underestimate.
Zizy - Wednesday, March 24, 2021 - link
Bulldozer lesson goes the other way around.Phenom2 couldn't really scale further and on servers AMD had MC that was great with highly parallel workloads and not much else; so they were competitive but they were clearly underdogs and sliding behind. So, much before launch, AMD was bragging about how BD will drive over Intel in everything and offer tremendous gains over their MC.
It was a spectacular disaster, significantly worse than promised. Zen launched to way less fanfare and was a much better chip than promised.
Obviously, there is no indication history will repeat itself, but if you want to draw any parallels from the history, it resembles era before BD launch, not before Zen.
whatthe123 - Wednesday, March 24, 2021 - link
Zen was exactly as promised. They laid the whole thing out at hotchips. Fanfare was just low due to AMD's low budget for marketing and years of bulldozer disappointments. Fanfare doesn't really seem that high for intel either, plenty of people are rightfully skeptical since 10nm is something like 5 years late now.Oxford Guy - Wednesday, March 31, 2021 - link
Zen was also:• slower than Intel (for gaming at least)
• quite limited in upper-end clockrate
• more latency than Intel, due to chiplets and design
• unstable with 3200-speed RAM sometimes
It was a tremendous improvement but also imperfect. One of the drawbacks, for consumers, is that it was designed to improve profit margin, via keeping the number of cores low in the chiplets, to maximize yield. That increases latency.
Marlin1975 - Tuesday, March 23, 2021 - link
"Intel stated that its 7nm process is now comfortably on track..."They said the same thing about 10nm for years. And the intel 10nm of today is not what they started with.
Santoval - Wednesday, March 24, 2021 - link
The above quoted sentence is meant for the average Intel investor, not impartial people with thinking brains.yeeeeman - Wednesday, March 24, 2021 - link
Quite hilarious actually since 7nm was in talks just a month ago to be changed to tsmc instead and now it is on track. DamnColin1497 - Tuesday, March 23, 2021 - link
Tick Tock doesn't really date back to the 1990s.quorm - Tuesday, March 23, 2021 - link
Yeah, didn't it start around the same time as "core" ?DigitalFreak - Tuesday, March 23, 2021 - link
2007, so roughly a year before Nehalem launched.GeoffreyA - Wednesday, March 24, 2021 - link
Not in terminology or exact philosophy, but the idea crudely existed back then. The Pentium, starting off at 0.8 um, was produced on a range of processes, probably 0.35 um being the most common. Pentium Pro on both 0.35 and 0.5 um. By the Pentium II's time it was starting to be differentiated better: Klamath, carrying the older process, was 0.35 and Deschutes 0.25 um. Pentium III Katmai started off at 0.25 and Coppermine shrunk that to 0.18 um.Pentium 4 Willamette took off at 0.18 um/180 nm, and Northwood went down to 130. Prescott, notably, alterered both microarchitecture and process (90 nm), but Cedar Mill was just a 65 nm die shrink; and Conroe, coming out shortly afterwards, launched at 65 nm.
Spunjji - Friday, March 26, 2021 - link
"Prescott, notably, alterered both microarchitecture and process"Notably indeed :D
GeoffreyA - Friday, March 26, 2021 - link
And now, the venerable Prescott passes on its (heated) baton to the Lake Rocket.taz-nz - Saturday, April 3, 2021 - link
Prescott was a giant band aid fix to the Pentium 4 choking due to lack of FSB & Memory bandwidth as core clock speeds rose. Intel just piled on a tons of cache memory hoping it would hid the issue without have to engineer a new chipset with a higher bus speed.I had a Pentium 4 HT 3.0 at the time, and sold it a purchased a Pentium 4 HT 2.8C and matched it with Corsair Dominator 2 1066 memory, I overclocked the FSB to 1066MT/s from 800MT/s and lower the CPU multiplier from 14x to 13x, giving me a clock speed just under 3.5ghz , my flat mate at the time had a Pentium 4 HT 560 (Prescott 3.6ghz) my system was much faster than his, despite his CPU have twice the cache memory and nearly a 150hz clock speed advantage.
The band aid didn't work, the CPU ended up almost 50% cache memory by die size, which cranked the power consumption and heat output to 11. The real fix was to get rid of the system bottle neck.
Techie2 - Tuesday, March 23, 2021 - link
Even if Intel was able to become the CPU performance leader I'd never allow the purchase of any of their products by our company due to Intel's documented history of crime and anti-trust law violations in their efforts to prevent a free, level marketplace for consumers. Intel's evil management mentality almost eliminated AMD and choice for all consumers. Intel could not pay me enough to use their products and they certainly don't deserve my business. Voting with you wallet can prevent exploitation.Kamen Rider Blade - Tuesday, March 23, 2021 - link
I'm with you, I'll never purchase a CPU from the convicted monopolist known to Lie/Cheat/Steal ever again.timecop1818 - Tuesday, March 23, 2021 - link
that's like saying "i don't want to live in usa because half the states legalized weed". continue buying your underperforming Chinese-built AMD while I stick to quality Intel.DigitalFreak - Tuesday, March 23, 2021 - link
Oh child. You have much to learn.Qasar - Tuesday, March 23, 2021 - link
nope, timecrap1818, just loves intel, and hates everything elseGeoffreyA - Wednesday, March 24, 2021 - link
Go on, stick with quality Intel.Otritus - Wednesday, March 24, 2021 - link
How is AMD "Chinese-built"? The company is an American company who designs their microarchitectures in America. Their foundry is TSMC, a Taiwanese company, with fabs in Taiwan and America. Also, right now AMD is "quality" given that they are more performant, less power hungry, and more feature rich.Bagheera - Friday, March 26, 2021 - link
1) weed is good for you. may be try it sometime so you are not so angry all the time.2) TSMC is Taiwanese, not Chinese. please educate yourself on the difference.
3) "quality" Intel? is that why Intel cancelled their overclocking warranty? are Apple and Qualcomm silicon not "quality" because TSMC also makes those?
Oxford Guy - Wednesday, March 31, 2021 - link
'TSMC is Taiwanese, not Chinese. please educate yourself on the difference.'The people of Hong Kong say hello.
Spunjji - Friday, March 26, 2021 - link
Someone's salty that TSMC are at least 1.5 fab generations ahead of Intel 😬Trenteth - Tuesday, March 30, 2021 - link
Totally Agree Techie2, all my many compute devices are all ether AMD silicon or Apple silicon currently. My last 6 PC's built for me, my wife and my HTPC are all Ryzen and my wife has a Ryzen Laptop as well. AIso just got an M1 Mac for myself which is really good. not an Intel CPU in sight.inighthawki - Tuesday, March 23, 2021 - link
They forgot a few of the "Optimization"s in the picture.Hifihedgehog - Tuesday, March 23, 2021 - link
TODAYProcess (Broadwell) -> Architecture (Skylake) -> Optimization (Kaby Lake) -> Optimization (Coffee Lake) -> Optimization (Whiskey Lake) -> Optimization (Comet Lake) -> Last-Ditch Effort (Rocket Lake)
DigitalFreak - Tuesday, March 23, 2021 - link
LMAO. I'd pay someone in Intel marketing to update their official slides with "Last-Ditch Effort".Hifihedgehog - Wednesday, March 24, 2021 - link
You're in luck. I believe the official PR material style guide was leaked in the huge data dump. So actually, someone could make their own version very easily.Rudde - Wednesday, March 24, 2021 - link
Small nit-pick, Whiskey Lake was for mobile, while desktop got Coffee Lake Refresh.Hifihedgehog - Wednesday, March 24, 2021 - link
I know. I just like the sound of Whiskey Lake more. It sounds like Intel is probably in recovery at an Alano Club.Spunjji - Friday, March 26, 2021 - link
Mobile also got Kaby Lake-R because they needed an emergency response to Zen Mobile.So depending on how you count it, there's up to 5 "optimizations" in there
Spunjji - Friday, March 26, 2021 - link
enhanceEnhance
ENHANCE
@#"£&%£#$*^@£^%^$@£^
Cullinaire - Tuesday, March 23, 2021 - link
"With any luck, if Intel can get a headwind with 7nm, when 2024 rolls around it might all come thick and fast."Pretty sure you meant to write "tailwind" there.
Spunjji - Friday, March 26, 2021 - link
Intel 2024: coming thick and fast all over your laptop(sorry)
TristanSDX - Tuesday, March 23, 2021 - link
I can't wait for next financial results, only to hear Pat's high attitude statements :)Tick-tock won't help to gain leadership products. AMD is 'firing on all cylinders', what means they can jump into new process and new architecture in single shot, Zen 2 and upcoming Zen 4 are prime example of that, while Zen 3 is 'only' new architecture.
outsideloop - Tuesday, March 23, 2021 - link
Amen. And TSMC will grow capacity for AMD into late this year and next. If Intel executes on their "dreams" perfectly, then they have a chance to get back in the game. If not, AMD will stay ahead and the lead will become insurmountable by 2024/2025.Drumsticks - Wednesday, March 24, 2021 - link
I've always found the assumption that AMD's lead will be insurmountable fascinating. AMD was much, much farther behind Intel with the launch of Bulldozer and its subsequent products than Intel is behind AMD now. Have people forgotten the 2500k being comparable in MT to Bulldozer with twice as many cores?Intel's core design is still reasonably good; if Golden Cove has a nice improvement, they'll likely have a higher IPC design than AMD. If Rocket Lake needed twice as many cores as Zen 3 to match it, that'd be one thing, but it's really only at a major disadvantage in one area, efficiency.
The most common response I see to this is that AMD is firing on all cylinders now, but while Intel wasn't making the kind of improvements AMD is making now, they were still turning out good products several years after Bulldozer came out.
Spunjji - Friday, March 26, 2021 - link
Yeah, I entirely agree with you that AMD are in no way "unassailable" at present - especially not by the juggernaut that is Intel. When people will buy your products in huge quantities whether good or bad, you can guarantee the revenue streams required to put good designs together in the current era of CPU design.GeoffreyA - Friday, March 26, 2021 - link
Yes. My belief is that Intel will be back on track one of these years, but it won't be a repeat of the Phenom/Bulldozer era. AMD, having learnt the hard way, is fighting like Goku against Frieza, and I think (I hope?) they'll keep it up. Just watch out for dirty tricks, AMD! Most likely, once Intel's process troubles are mended, we'll see Ryzen and Core toe to toe, each leading slightly at different times. "Core leads, 2030. Ryzen back on top, 2031." Of course, ARM could muddle this nice picture but we'll see what happens.GeoffreyA - Wednesday, March 24, 2021 - link
Agreed. Tick-Tock might not be enough this time. As AMD's Mike Clark, I think, said: "We're going tock, tock, tock."FunBunny2 - Wednesday, March 24, 2021 - link
"We're going tock, tock, tock."one has to wonder when the laws of God make ever smaller nodes impossible. quantum computing is cool, but doesn't fit easily with deterministic tock.
Fulljack - Wednesday, March 24, 2021 - link
I think the point is since AMD are fabless, they'd focus solely on the uArch design and not about the fabrication process.GeoffreyA - Wednesday, March 24, 2021 - link
"ever smaller nodes impossible"I think we're on the verge of that now, quantum tunnelling effects kicking in at <= 3 nm. I'd like to see what TSMC, Intel, AMD, and Apple do once the limit is reached. As for quantum computing, I sometimes wonder if these computers are really doing anything more than just being indeterministic---even a classical computer can simulate a QC, I believe, but orders of magnitude slower. At any rate, I need to get a better grounding on this topic.
GeoffreyA - Wednesday, March 24, 2021 - link
* just being probabilisticzodiacfml - Tuesday, March 23, 2021 - link
wow 2024. we're not even half of 2021. if they could just take the hit and be the value option.Drumsticks - Wednesday, March 24, 2021 - link
AMD embraced being the "value option" once long ago, I think it was like 2013? They didn't get on top with Zen (really Zen 2 and 3) by embracing the value option though. AMD ultimately didn't settle for mediocrity, and we shouldn't really want Intel to either. AMD is probably driven right now by the knowledge that Intel *can* put out capable products and execute a capable roadmap - the fact that they haven't for the last few years might inform expectations, but neither AMD nor us should assume Intel will never get a competent roadmap executed again.Spunjji - Friday, March 26, 2021 - link
100%shabby - Tuesday, March 23, 2021 - link
What is this new ceo on? 😂Techie2 - Tuesday, March 23, 2021 - link
He thinks he's back in the 80's. ;)Santoval - Wednesday, March 24, 2021 - link
"Intel stated that its 7nm process is now comfortably on track to deliver Meteor Lake, a client CPU using tiles/chiplets, in 2023, however we are likely looking to a 7nm variant or even external processes for a 2024/2025 product."Meteor Lake (on a brand new 7nm process node with brand new Ocean Cove "big" cores - the tiles thing is also brand new) will almost certainly be released on Q1 2024 at the earliest, thus the 2024/2025 timeline for ‘unquestioned CPU leadership performance’. If Ocean Cove cores on a new very efficient 7nm process node that is slightly denser than TSMC's 5nm process node (at the maximum density variant of both) cannot beat Zen 5 -they will almost certainly beat Zen 4 but I strongly doubt Ocean Cove will compete with that- then nothing can.
I expect Intel *will* fab 7nm parts in 2023 (though nothing in 2022), just not Meteor Lake. They never fab their premiere desktop parts on new nodes. They tend to fab mobile parts with up to 4 cores and up to 15 - 20W TDPs, so that the customers themselves do the final testing and debugging of the new node "in the wild". They might also fab some 7nm GPUs in 2023, though not the enthusiast or server ones. Up to mainstream class tops.
Otritus - Wednesday, March 24, 2021 - link
AMD doesn't have the resources to build a new microarchitecture every year. If Zen 4 is 2022, then the next major microarchitecture update would be 2024. If Ocean/Redwood cove is 2023, then Intel can bring a new microarchitecture 2025, which would beat Zen5+/6 (whatever AMD calls it). Remember that the microarchitecture for 7nm would be complete in 2022/2023, so even if Ocean Cove is delayed to 2024, it's possible for Intel to tock for 2025. I do agree that it is unlikely that Intel beats AMD's post Zen 4 big architecture in 2024, but 2025 should be competitive.Spunjji - Friday, March 26, 2021 - link
Aren't Intel supposed to be putting their Ponte Vecchio HPC GPU out as one of the (if not the) first products on 7nm?My best guess on why is that, after the abject failure and cost-soak of trying to manufacture tiny dual-core mobile chips on OG 10nm and failing, they've gone for a strategy that will probably pay for itself no matter how few GPU tiles they can yield from each wafer.
The big tell will be how fast they can move from pushing out a trickle of server-grade parts for a supercomputer project to ramping consumer parts that actually require decent yields to be profitable.
Silver5urfer - Wednesday, March 24, 2021 - link
I want what Gelsinger is smoking right now. What a comedy show... Tick Tock ? As much as I wanted to see it come. Its simply not possible now. Intel squandered their 5 year advantage of Fab Lithography edge long time back. Their true age was FinFet Haswell and 14nm debut in 2014. That's 6 years ago we are still in 14nm++ yes its very much optimized but its over for them.This guy saying x86 licensing, IP like EMIB (AMD G processors does anyone remember ?) And Foundry business. What the fuck ? Intel backport failed with RKL on Cypress Cove and 14nm++ now they want all of that licensing money by selling out their top x86.
Next this guy says High performance in 2024, by that Time even Apple might ship a server ARM alternative if we want to use ladders in sky type of talk. To be realistic AMD in 2024 is Zen5 probably on TSMC 3nm. And Intel doesn't even have 7nm by that time. Add Samsung GAAFET 3nm so who wants 7nm ? ARM processors will not for sure. Apple won't either. Altera ? Graviton ? Or that MS ARM server CPU ? Haha.
Man this guy is really all talk. First show us a successor for X299 then we talk.
Rudde - Wednesday, March 24, 2021 - link
Zen 5 is unlikely to be on 3 nm. Zen 6 on the other hand.drexnx - Wednesday, March 24, 2021 - link
Haswell was neither the first 14nm processor, nor was it the first finfetIVB was 22nm and finfet debut, and Broadwell was 14nm debut. That Intel initially had so many problems with 14nm at high power was an ill omen for their 10nm process (Broadwell-C was the only official desktop variant, and it was capped at 65w)
Silver5urfer - Wednesday, March 24, 2021 - link
Yeah, I didn't mention Broadwell which is a mistake by omission, I was in bed. BWL was 14nm debut yes and 22nm FinFet on IVB which I completely forgot.drothgery - Wednesday, March 24, 2021 - link
If Intel only had their early 14nm-level problems with 10nm and 7nm, they'd have 7nm mobile chips out *now*.Spunjji - Friday, March 26, 2021 - link
14nm was the early warning signs that all was not well in paradise - it arrived later than expected and the initial results were underwhelming.Sometimes I wonder how good 22nm would have looked if they'd kept improving it like they eventually did with 14nm, rather than trying to kick it out the door ASAP.
SaturnusDK - Wednesday, March 24, 2021 - link
" Intel stated that its 7nm process is now comfortably on track to deliver Meteor Lake, a client CPU using tiles/chiplets, in 2023...."Let me check the TSMC process node roadmap as reported here on Anandtech real quick..
TSMC expects to enter full volume production on 3nm in 22H2.
Auch Intel. Even if we assume the Intel 7nm is comparable to one form of TSMC 5nm that still means they're at least a full process node behind in 2023.
mattbe - Wednesday, March 24, 2021 - link
"TSMC expects to enter full volume production on 3nm in 22H2."With only enough supply for Apple and possibly Huawei. There won't be capacity available from AMD until at least 2023.
You are right they are still a node behind even if 7nm comes out in 2023. There's a lot of catching up for Intel to do if they want to take back the processing lead.
Bagheera - Friday, March 26, 2021 - link
Huawei for kicked off TSMC client list last year. AMD is on track to become TSMC's second largest customer.Danvelopment - Wednesday, March 24, 2021 - link
The "Today" on that image is missing about 15 "Optimisation" items.Rudde - Wednesday, March 24, 2021 - link
Kaby Lake, Coffee Lake, Coffee Lake R, Comet Lake. Rocket lake is tock. Tick tock opt opt opt opt tock.duploxxx - Wednesday, March 24, 2021 - link
Well the show is there, that was needed off course from CEO perspective, the show must go on.Now the facts, the bleeding is hurting a lot....and will continue for a very long time.
Bad 10nm process, bad core re-designs, bad roadmap. Desktop is waiting for Alderlake which is again not scale the Core count, Ice lake is hardly improving anything (few cores more and finally more memory lanes) and Eagle is another year out.
So its 2023-2024 to get back on track with no idea where the competition will be at that point.
Lost apple as a customer by that time in all the segments, lost at least 30% marketshare by that time to AMD. Have no footage in mobile phone devices, GPU has yet to be seen if it will any change of taking off....
Adding 2 fabs? they will have spare capacity by the time they are ready... but they will always sell lots of stuff even if it is bad since the OEM market is so broken ECO system and R&D sponsoring.
edzieba - Wednesday, March 24, 2021 - link
"Tiles by this definition are more costly to implement than chiplets, and have additional thermal considerations by having high-powered silicon close together"Through-substrate interconnects have their own thermal issues: both ends of every link require drivers, which are pure power waste. The more links (i.e. the more chiplets you split the design into) the more power is spent for no useful output: https://www.anandtech.com/show/13124/the-amd-threa... . Inter-die (EMIB, Foveros, etc) routing has minimal overhead over routing over the same area monolithic die.
While stacked dies complicate internal thermal management (Z- conduction path), they also mean total package thermal management is simpler (reduced total thermal power, smaller physical package).
JayNor - Wednesday, March 24, 2021 - link
What would you call the more than 40 pieces in the xe-hpc?Most of it is foveros stacked chiplets. Perhaps he refers to the emib connected units as tiles.
edzieba - Thursday, March 25, 2021 - link
"What would you call the more than 40 pieces in the xe-hpc?"Tiles.
Chiplet = interconnect through substrate, requires TX/RX at either end of link and protocol translation (die <> link <> die). Power per bit high (though lower than through PCB).
Tile = interconnect through silicon, does not require dedicated drivers and may or may not not have specific inter-tile protocol (protocol more for interchangeability when desired, e.g. HBM, otherwise tiles can be treated as if silicon were monolithic). Power per bit low (though higher than monolithic silicon due to TSV resistance).
Spunjji - Friday, March 26, 2021 - link
So far we have two great examples of chiplets on the market (Zen 2 and 3) and one astonishingly mediocre example of stacked dies (Lakefield).EMIB certainly looks promising, but Foveros is not yet something I can bring myself to get enthusiastic about.
watzupken - Wednesday, March 24, 2021 - link
I think Pat said a lot of great things, but talk is cheap. Intel did not voluntarily stop their Tick Tock cycle. They were forced to because their could not deliver 10nm on time. So unless him joining Intel can miraculously overcome all these obstacles, then I will buy his marketing. 2024/25 is really not that far away when you consider the fact that it takes many years to build a quality CPU from scratch. Moreover without their fab advantage, Intel needs to work a lot harder to keep themselves competitive. So looking forward to see if Pat can bring these supposed miracles to turn this oversized “lorry” around in 4 years.JayNor - Wednesday, March 24, 2021 - link
There was some announced plan to simplify the design rules so they could push out the decision on process. Perhaps they discovered this is not so easy, or it is still a work in progress.Otritus - Wednesday, March 24, 2021 - link
One reason why Intel found it hard to shrink their node was the massive increases in density. 14nm was 2.7x as dense as 22nm+, and 10nm was 2.7x as dense as 14nm++. Most foundries go for 1.6x to 2x density improvements per major node shrink. Being less aggressive on the density improvements would allow for Intel to tick more frequently.JayNor - Wednesday, March 24, 2021 - link
Intel is already producing 7nm xe-hpc compute chiplets.Bagheera - Friday, March 26, 2021 - link
Intel has been "producing" 10nm since 2016.there's a thing called *yield*.
Spunjji - Friday, March 26, 2021 - link
Yup. It's no surprise they tried to memory hole Cannon Lake.I'd hate to see how many functional dies they get per wafer on 7nm right now.
willis936 - Wednesday, March 24, 2021 - link
It took the "tick-tick-tick-tock" joke to be at least half the age of the "tick-tock" model for Intel to even acknowledge that tick-tock is currently a dream.The first step to fixing a problem is admitting you have one. I hope to see less bullshit and more competition.
oklukar - Wednesday, March 24, 2021 - link
"One of Gelsinger’s mantras seems to be that unquestioned leadership products bring unquestioned leadership margins for those products, and for Intel to execute, it needs to return to its days of old."So I guess this Gelsinger guy believes that AMD will simply stand still and watch them pass by. 🤦🏽♂️
As the saying goes: you only get one! 😉
Sivar - Wednesday, March 24, 2021 - link
"In the past, through the 1990s, 2000s, and into the 2010s, Intel’s manufacturing philosophy was known as ‘Tick-Tock’."Tick-Tock wasn't named until the Core architecture in 2006 not 2007, was it?
Spunjji - Friday, March 26, 2021 - link
CorrectTheJian - Friday, March 26, 2021 - link
https://www.tweaktown.com/news/77583/tsmc-will-mak...Click the pics to blow them up.
https://www.forbes.com/sites/patrickmoorhead/2021/...
Perhaps far more important than previous link, but I might be biased. I've been following patrick since, jeez, never mind, I'm old. Let's just say decades and leave it at that. He's #1. Just saying. Make of those two links what you will and remember Anandtech was (is? don't come here much today) an AMD portal site. ;)
Spunjji - Friday, March 26, 2021 - link
"remember Anandtech was (is? don't come here much today) an AMD portal site"You seem to be casually referring to when they had a specific AMD section for AMD-sponsored news. Do you see that here anymore?
It's been good not seeing your posts around here. Don't let us persuade you not to stay away!
dsplover - Saturday, March 27, 2021 - link
They’re trolling AMD.In 2022 tick-tock will happen and take the world by surprise.
Until then if you absolutely need something new Intel CPUs are cheap cheap cheap.
Qasar - Saturday, March 27, 2021 - link
i will believe anything intel says, if and when it happens.Oxford Guy - Saturday, March 27, 2021 - link
He could have added 'and we're going to make overclocking really really good again, too'.After all, if one can magically restore the cadence with which process nodes can shrink to be what it used to be (ignoring those pesky laws of physics), why not add that symptom of that cadence?
I am far from an expert in node tech but I recall the statements from experts, like:
• The complexity goes way up as the nodes shrink. This manifests as 'design rules'. That means added cost (financial and time) and difficulty. More and more design rules appear with each node shrink. This obviously means that unless the company becomes bigger and is able to execute all of this added complexity just as quickly as it could execute the less complexity before, things will slow down.
• Silicon is marching toward the point of returns so diminished that a different material will have to be used. This angle on the problem is more speculative but we have seen the death of overclocking, the spewing of cores/chiplets as a kludge, the lack of progress past 5 GHz, increasingly power-hogging parts requiring now only solder but thinner substrates, etc.
watzupken - Sunday, March 28, 2021 - link
I am not sure if tick tock is still feasible when its going to be even more challenging shrinking from 7nm to 5nm, so on and so forth. They already struggled with 10nm and with the experience gained, still struggled to deliver 7nm in 3 years, so I don't expect any miracle again with the new CEO. In addition, considering that AMD's Zen 2 and Zen 3 are utilizing the same 7nm, and introduced performance improvements more on the "tock" cycle in both instances, I think this tick-tock strategy may likely struggle.