In-Depth: Intel's 10nm was definitely NOT too ambitious

[moinmoin]

Intel is lucky that they'll have more time to recover.

I wonder if the future tense is still applicable here due to the multi year lead time necessary. Intel had more time to recover, the coming years we will see whether they made any actual use of it aside all the filler they managed to push out until now.

 

[dacostafilipe]

Any company only needs to release a new product if their old product stops making them (enough) money AMD already released 3 different ZEN spins while intel still makes more money with their 14nm CPUs,intel just had their best profit quarter since forever...

This is how big companies die ... we don't want that for Intel!

 

[Zucker2k]

Intel still has the superior IPC, it just happens to be on mobile. Just as in the past, they just have to find a way to bring it to the desktop, process allowing. All this talk about companies resting on their laurels is neither here nor there, unless people want to suggest Intel intentionally poured billions down the drain. Intel did not stop developing architecture. If anything at all, they've learnt how not to make 10nm and sub nodes, a mistake that'll provide valuable lessons for them moving forward. They already seem to be working out the core issues with 10nm and we should soon know if their recent confidence (in 10nm) is justified by the delay and their insistence that they have a working process node.

 

[scannall]

Tech companies can't afford to slow down their pace of innovation. Someone is always waiting in the wings to upend their apple carts. Hector Ruiz made the mistake of letting AMD rest on their laurels, and look where it got them? It didn't even take very long for AMD to go down the toilet. Intel is lucky that they'll have more time to recover.

Hector was more interested in building a golden parachute, and getting as much cash as he could from AMD.

 

[jpiniero]

If anything at all, they've learnt how not to make 10nm and sub nodes, a mistake that'll provide valuable lessons for them moving forward.

Intel literally said the same thing about 14 nm with the year delay that 14 nm had. EUV I imagine is quite different so I don't know how much anything they learned with 10 nm will help.

 

[cellarnoise]

EUV has blended several (3) generations of shrinkage. And very expensive at that. Intel tried to go too small without it on the current formulas. Crazy to watch this miscalculation from a huge Corp.

Leading tech is a nation state / military issue. Will be interesting to see what quietly happens (to start) if nations fall behind.

We are about at Terminator levels of observation anyway... A.I. level that we fear is how far off?

 

[ThatBuzzkiller]

EUV has blended several (3) generations of shrinkage. And very expensive at that. Intel tried to go too small without it on the current formulas. Crazy to watch this miscalculation from a huge Corp.

Leading tech is a nation state / military issue. Will be interesting to see what quietly happens (to start) if nations fall behind.

For the last statement, I suspect that you'll see the deindustrialization of silicon valley unfold before our very eyes ...

Once Intel, Micron, Qualcomm, Apple, Xilinx, and lastly Nvidia fall down or get ousted to foreign competitors (most likely Chinese) there will be massive layoffs initiated overtime across the semiconductor industry ...

The thought of having very talented individuals such as HW engineers out of employment or underemployed and not being able to control a strategically important industry like semiconductors or worse a hostile foreign power like China having near total control over it is entirely nerve-wracking

Massive amounts of economic damage would be unleashed on America and it would have irreversible social consequences with communities being torn apart as seen in the American rust belt. Even highly educated people will be politically self challenged on their progressive beliefs as well once they too find themselves living from paycheck to paycheck which could drastically change the voting demographics ...

No country or nation can realistically maintain their wealth or long-term economic stability without an industrial economy sustaining it ...

 

[Jimzz]

Are there any deep dives / Xrays of the new 10nm CPUs from intel yet? Wonder how the final product compares to the original density intel was saying they would hit.

 

[Nothingness]

Are there any deep dives / Xrays of the new 10nm CPUs from intel yet? Wonder how the final product compares to the original density intel was saying they would hit.

Sunny Cove - Microarchitectures - Intel - WikiChip

Sunny Cove (SNC), the successor to Palm Cove, is a high-performance 10 nm x86-64 core microarchitecture designed by Intel for an array of server and client products, including Ice Lake (Client), Ice Lake (Server), Lakefield, and the Nervana NNP-I. The microarchitecture was developed by Intel's...

en.wikichip.org

Not exactly a deep dive but it's all I found.

 

[DrMrLordX]

Intel still has the superior IPC, it just happens to be on mobile.

And that's where it will stay.

Hector was more interested in building a golden parachute, and getting as much cash as he could from AMD.

Yes indeed, he was quite the charlatan. But he really did think x2 would hold off Netburst for awhile longer, possibly to make his egress all the more lucrative. That sure didn't work out for him.

 

[maddie]

Sunny Cove - Microarchitectures - Intel - WikiChip

Sunny Cove (SNC), the successor to Palm Cove, is a high-performance 10 nm x86-64 core microarchitecture designed by Intel for an array of server and client products, including Ice Lake (Client), Ice Lake (Server), Lakefield, and the Nervana NNP-I. The microarchitecture was developed by Intel's...

en.wikichip.org

Not exactly a deep dive but it's all I found.

Looking at those cores, can we assume that Intel does a lot more manual circuit design vs automated tools compared to AMD?

 

[DrMrLordX]

Looking at those cores, can we assume that Intel does a lot more manual circuit design vs automated tools compared to AMD?

@dmens might be able to tell you more about that (or maybe not). But my guess would by "yes".

 

[dmens]

@dmens might be able to tell you more about that (or maybe not). But my guess would by "yes".

When I left, Intel was woefully behind on the rest of industry on design automation, with an institutional bias towards obsolete in-house tools. Not-invented-here is gospel at Intel.

 

[DrMrLordX]

When I left, Intel was woefully behind on the rest of industry on design automation, with an institutional bias towards obsolete in-house tools. Not-invented-here is gospel at Intel.

Thanks for the input there. I remember the Bulldozer design effort being panned for relying excessively on automated design tools. How times have changed.

 

[maddie]

Really interesting comparing these Intel die shots with the one in this post.

Page 12 - Info - [Toms, Anand] AMD EPYC Benchmarks

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[mtcn77]

And that's where it will stay.



Yes indeed, he was quite the charlatan. But he really did think x2 would hold off Netburst for awhile longer, possibly to make his egress all the more lucrative. That sure didn't work out for him.

I disagree. Many remember him without the 'netbook' escapade.
He was quite the visionary at the dawn of netbooks, he in fact pioneered its beginning with Geode processors. Too bad it was a very lean project without much premium covering its assets which served to Intel's proclivities further down the line to Intel's latent mobile write-off indulgences on ARM opposition which backfired quite spectacularly.

 

[mtcn77]

Plus, I think we need not disassociate the process node from process performance. Intel has been doing great on 14nm, a long in the tooth reason is process performance due to power customization - big gates don't deduce from the performance, nor increase impedance. Best execution at any particular node is still the dominant factor, not power, nor p/w - something derived from both.

 

[dmens]

Thanks for the input there. I remember the Bulldozer design effort being panned for relying excessively on automated design tools. How times have changed.

That is a silly criticism. AMD did not have the manpower or resources of Intel so they could never throw people into the meat grinder the way Intel did.

In any case, the same arrogance and hubris on doing things "the old way" was one of the reasons that Intel's mobile effort ended up as a catastrophic failure. Intel not only ensured its irrelevance in that market, an entire engineering organization was effectively disbanded as a result of that failure. And even after that debacle they still refuse to change anything. Intel is intent on ending up like IBM.

 

[NostaSeronx]

I remember the Bulldozer design effort being panned for relying excessively on automated design tools.

It should be noted however that Bulldozer is hand-drawn/hand-custom/hand-crafted and not synthesized like Bobcat, Jaguar, or Zen. The automation detailed didn't actually start till Steamroller. Even then, it is specifically mentioned that is only semi-hand-custom, not wholly synthesized like BT, JG, ZN.

Bulldozer/Piledriver => Hand-drawn for maximum speed and density in 32nm.
Steamroller => The same blocks again, but rebuilt using the 13-track library in 28nm.
Excavator => The same blocks again, but rebuilt using the 9-track library in 28nm.

 

[Nothingness]

In any case, the same arrogance and hubris on doing things "the old way" was one of the reasons that Intel's mobile effort ended up as a catastrophic failure. Intel not only ensured its irrelevance in that market, an entire engineering organization was effectively disbanded as a result of that failure. And even after that debacle they still refuse to change anything. Intel is intent on ending up like IBM.

It also explains why their foundry initiative crashed.

 

[mtcn77]

It should be noted however that Bulldozer is hand-drawn/hand-custom/hand-crafted and not synthesized like Bobcat, Jaguar, or Zen. The automation detailed didn't actually start till Steamroller. Even then, it is specifically mentioned that is only semi-hand-custom, not wholly synthesized like BT, JG, ZN.

Bulldozer/Piledriver => Hand-drawn for maximum speed and density in 32nm.
Steamroller => The same blocks again, but rebuilt using the 13-track library in 28nm.
Excavator => The same blocks again, but rebuilt using the 9-track library in 28nm.

Yes, at the time Intel didn't miss out its chance to pass on how it was the one, true honest cpu designer. I admit, I was persuaded into thinking Bulldozer was an auto-disaster, but Steamroller Richland & its compact library was the first iteration I suppose.
PS: In previous redaction, I said Richland. Now I remember the reason - Richland is the first iteration of a cpu optimised node AMD built a gpu on. So they have had to migrate their high density library a.k.a. Radeon IP for it to work.

 

[StrangerGuy]

Yes indeed, he was quite the charlatan. But he really did think x2 would hold off Netburst for awhile longer, possibly to make his egress all the more lucrative. That sure didn't work out for him.

That selfish POS caused more damage to the PC industry than anybody did by rejecting NV's merger offer.

 

[NTMBK]

PS: In previous redaction, I said Richland. Now I remember the reason - Richland is the first iteration of a cpu optimised node AMD built a gpu on. So they have had to migrate their high density library a.k.a. Radeon IP for it to work.

Nah, they've been doing it since Llano on 32nm SOI.

 

[mtcn77]

Nah, they've been doing it since Llano on 32nm SOI.

Llano might be it, however through sheer execution alone Richland marks a turning point. I'd say it isn't the landmark, but switching speed alone ensures that it is possible. Auto-designs aren't the fastest because hand traced circuits cut down on excess transistors and interface routing. So, it has to be the opposite of an unoptimised design if it goes fast. I discredit any node progress, it has to be a less dense footprint, better layout.

 

[Panino Manino]

Don't know if 10nm is the cause, but Charlie from Semi Accurate published an apocalyptic piece about new Intel delays.