Discussion Intel’s Unified Core: Hammer Lake

[Quintessa]

P/E are getting replaced by Unified Core based on atom but LBT wants that they keep HT with going unified

Right, "Unified Core" ≠ "Big Core Lite". It's more like Intel fusing P/E DNA into one scalable block (closer to beefy Atom lineage) while keeping HT to avoid the "one-thread-per-core" efficiency ceiling.

also N3B didn't clock much high lol

Yeah, early silicon's not screaming. Apple's M3 barely stretches past 4GHz, and that's with insane power management. If Panther Lake Unified Cores on N3B are hitting mid-4s stable, that's already a win for mobile/laptop.
LBT seems to be betting:

• Simplify scheduler hell (one core type, no E/P juggling)
• Keep HT for bursty desktop loads
• Pray TSMC's N3B/N2 ramps thermals enough to not choke clocks

 

[poke01]

Right, "Unified Core" ≠ "Big Core Lite". It's more like Intel fusing P/E DNA into one scalable block (closer to beefy Atom lineage) while keeping HT to avoid the "one-thread-per-core" efficiency ceiling.

Yeah, early silicon's not screaming. Apple's M3 barely stretches past 4GHz, and that's with insane power management. If Panther Lake Unified Cores on N3B are hitting mid-4s stable, that's already a win for mobile/laptop.
LBT seems to be betting:

• Simplify scheduler hell (one core type, no E/P juggling)
• Keep HT for bursty desktop loads
• Pray TSMC's N3B/N2 ramps thermals enough to not choke clocks

Apple moved on from N3B, M4 is clocking up to 4.5GHz on N3E

I hope 18A clocks higher than N3B Arrow lake-H parts

 

[511]

Apple moved on from N3B, M4 is clocking up to 4.5GHz on N3E

I hope 18A clocks higher than N3B Arrow lake-H parts

I am hoping for 5.5 GHz lol with 0 IPC Improvement

 

[adroc_thurston]

Apple moved on from N3B, M4 is clocking up to 4.5GHz on N3E

It's not just the node there.

 

[Quintessa]

I hope 18A clocks higher than N3B Arrow lake-H parts

Arrow Lake-H stuck on N3B is basically the sacrificial lamb. It's why you see those middling 4.2 - 4.3GHz ceilings, thermals + voltage headroom just aren't there.
18A? Totally different ballgame:

• Gate-all-around + PowerVia should free up ~10 - 15% more usable power budget.
• Intel's own target whispers: ~5GHz+ sustained on mobile SKUs, higher for desktop (5.5 - 6.0GHz flirting territory).

If Panther Lake/18A delivers that, Intel leapfrogs both N3B Arrow Lake and N3E Apple in clocks, at least until N2 hits.

 

[Doug S]

Maybe the question should be asked whether consumers are going to give a sh-- whether the PC they buy has that "Copilot Plus" branding. Microsoft doesn't have the power they used to, it is unlikely they will be able to force all PCs sold by major brands to have that branding they way they might have been able to 10 or 20 years ago.

 

[adroc_thurston]

it is unlikely they will be able to force all PCs sold by major brands to have that branding they way they might have been able to 10 or 20 years ago.

yes they can, though it is more of a carrot than a stick

 

[dangerman1337]

From the RZL, Serpent and Titan Lake thread since all three don't have Unified Core. Wonder what Intel is aiming for with Unified Core in Hammer Lake in terms of ST? Equal or more to Griffin Cove? Also wonder about the core count and cache size for LGA 1954 SKUs.

 

[adroc_thurston]

Equal or more to Griffin Cove

I mean obviously more, flat 1t for a gigatock is not viable.

 

[soresu]

Gate-all-around + PowerVia should free up ~10 - 15% more usable power budget.

It should be noted that "Gate-all-around" or GAA is an umbrella term for multiple different transistor designs.

The "ribbon FET" GAA transistor used by the SOTA Intel nodes is basically little different from MBCFET/Nanosheet used by Samsung 3nm+ and TSMC N2+.

AFAIK the transition design of Forksheet between Nanosheet and CFET is also a GAA design.

 

[Tigerick]

Let's try to put upcoming desktop CPU from Intel and AMD with corresponding process node:

1. H2-2026: Nova Lake (16+32, N2+18A) vs Zen 6 (24c/48T)
2. H2-2027: Razor Lake (16+32, A16?+14A?) vs Zen 6c? (32c/64T, N2/N2P)
3. H2-2028: RIP or Refresh
4. H2-2029: Hammer Lake (Unified cores with SMT?, A14?+14A-E?) vs Zen 7 (32c/64T, A14)

Hammer Lake and Zen7 should be aimed to use TSMC's A14 node. Thus, they should be released at the same time pending TSMC's schedule. Does it make sense for you? Oh Yeah, Silicon Fly's 100 speculated cores are definitely a joke. Threads.... depending on SMT implementation.

 

[Geddagod]

Let's try to put upcoming desktop CPU from Intel and AMD with corresponding process node:

1. H2-2026: Nova Lake (16+32, N2+18A) vs Zen 6 (24c/48T)
2. H2-2027: Razor Lake (16+32, A16?+14A?) vs Zen 6c? (32c/64T, N2/N2P)
3. H2-2028: RIP or Refresh
4. H2-2029: Hammer Lake (Unified cores with SMT?, A14?+14A-E?) vs Zen 7 (32c/64T, A14)

Hammer Lake and Zen7 should be aimed to use TSMC's A14 node. Thus, they should be released at the same time pending TSMC's schedule. Does it make sense for you? Oh Yeah, Silicon Fly's 100 speculated cores are definitely a joke. Threads.... depending on SMT implementation.

What nodes they will use is gonna be extremely interesting.
I would not be surprised in the slightest if Razor lake has compute on an 18AP iteration for even desktop flagship. I think if Intel can manage to compete in Fmax, they will throw away any perf/watt benefits in desktop. And because the core is a tock too, there's less pressure for Fmax to match last gen as well. We will see though.
14A is almost certainly too early for Razor Lake given LBT saying 14A is a 28-29' node, IMO.

 

[adroc_thurston]

I think if Intel can manage to compete in Fmax, they will throw away any perf/watt benefits in desktop

funniest post ever maybe unintentionally so

 

[Geddagod]

funniest post ever maybe unintentionally so

That's what Intel has been doing for desktop with 14nm since forever, with 10nm?
And since when has desktop cared that much about power?

 

[adroc_thurston]

That's what Intel has been doing for desktop with 14nm since forever, with 10nm?

They now do it the normal way, kiddo.

And since when has desktop cared that much about power?

Less power and more "18AP is dung slow".

 

[msj10]

What nodes they will use is gonna be extremely interesting.
I would not be surprised in the slightest if Razor lake has compute on an 18AP iteration for even desktop flagship. I think if Intel can manage to compete in Fmax, they will throw away any perf/watt benefits in desktop. And because the core is a tock too, there's less pressure for Fmax to match last gen as well. We will see though.
14A is almost certainly too early for Razor Lake given LBT saying 14A is a 28-29' node, IMO

still most likely N2P for RZL, maybe N2X if they want to push it.

 

[Geddagod]

Less power and more "18AP is dung slow".

I mean Intel 7 prob looked terrible in comparison to N5 too, but they managed to push voltages high enough in that node to not just compete but outright beat AMD in Fmax.

still most likely N2P for RZL, maybe N2X if they want to push it.

A lil insane that the first company confirmed to use a TSMC -X node is Qcomm.

 

[adroc_thurston]

I mean Intel 7 prob looked terrible in comparison to N5 too, but they managed to push voltages high enough in that node to not just compete but outright beat AMD in Fmax.

They now do it like everyone else.
And they suck at doing stuff like everyone else.

A lil insane that the first company confirmed to use a TSMC -X node is Qcomm.

AMD's been using overdrive Vt's since Raphael.

 

[dangerman1337]

Let's try to put upcoming desktop CPU from Intel and AMD with corresponding process node:

1. H2-2026: Nova Lake (16+32, N2+18A) vs Zen 6 (24c/48T)
2. H2-2027: Razor Lake (16+32, A16?+14A?) vs Zen 6c? (32c/64T, N2/N2P)
3. H2-2028: RIP or Refresh
4. H2-2029: Hammer Lake (Unified cores with SMT?, A14?+14A-E?) vs Zen 7 (32c/64T, A14)

Hammer Lake and Zen7 should be aimed to use TSMC's A14 node. Thus, they should be released at the same time pending TSMC's schedule. Does it make sense for you? Oh Yeah, Silicon Fly's 100 speculated cores are definitely a joke. Threads.... depending on SMT implementation.

Not sure about nodes (think RZL will be N2X) and timeframes but I'd be tad suprised if Hammer Lake isn't Q4 2029 because Intel pretty much always gets something out for Q4 as much as possible (Coffee Lake, Alder Lake, Raptor Lake, Raptor Lake Refresh, Arrow Lake etc). And likely by Q4 2029 we'll see RTX 70/RDNA 6/maybe Xe4 dGPU so Intel will probably want to see Desktop HML launched by then. Ofc HML-H etc for laptop is CES 2030.

still most likely N2P for RZL, maybe N2X if they want to push it.

I think RZL if it's Q4 2027 or Q1 2028 it'll be TSMC N2X.

 

[Geddagod]

They now do it like everyone else.
And they suck at doing stuff like everyone else.

Intel 18A can't be eventually optimized for extremely high voltages to compete in perf just like Intel 7 was?

AMD's been using overdrive Vt's since Raphael.

Seems like there's some other BEOL stuff that determine if a node is an -X node as well.
Either way, AMD isn't willing to admit to using an -X node, even if techinsights claims the are, but Qualcomm is more than happy to report themselves as using an -X node.

still most likely N2P for RZL, maybe N2X if they want to push it.

I agree with you in terms of what I think is most likely going to happen, but I think Intel would want to push a return to Intel as much as possible. It's going to be really interesting to see Intel's node choices over the next couple of years.
I think it would be better if they put server on external once they hopefully get design competitive in Coral Rapids, while pushing as much client stuff internally as possible to compensate for that.

 

[adroc_thurston]

Intel 18A can't be eventually optimized for extremely high voltages to compete in perf just like Intel 7 was?

no.
welcome to the world of backside power.

Seems like there's some other BEOL stuff that determine if a node is an -X node as well.

no it's just xtor tunings.

Either way, AMD isn't willing to admit to using an -X node, even if techinsights claims the are, but Qualcomm is more than happy to report themselves as using an -X node.

AMD likes to ninja like a billion different things.

 

[DavidC1]

Right, "Unified Core" ≠ "Big Core Lite". It's more like Intel fusing P/E DNA into one scalable block (closer to beefy Atom lineage) while keeping HT to avoid the "one-thread-per-core" efficiency ceiling.

Yeah, early silicon's not screaming. Apple's M3 barely stretches past 4GHz, and that's with insane power management. If Panther Lake Unified Cores on N3B are hitting mid-4s stable, that's already a win for mobile/laptop.
LBT seems to be betting:

• Simplify scheduler hell (one core type, no E/P juggling)
• Keep HT for bursty desktop loads
• Pray TSMC's N3B/N2 ramps thermals enough to not choke clocks

Unified Core isn't an "LBT bet". It was done because the P core design and the design team is sucking.