Discussion Intel Nova Lake in H2-2026: Discussion Threads

Page 15 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DavidC1

Golden Member
Dec 29, 2023
1,616
2,671
96
No but getting more ILP is harder without becoming a 13mm^2 Real American Core in the process.
What are you blabbering about? Apple already noticeably outperforms both Arrowlake and Zen 5 at just 4.5GHz. Other ARM vendors significantly outpace them too.
You do understand that everyone in aa64 camp will be shipping 5GHz CPUs next year?
Again, big issue on the x86 camp. Because of their clockspeed ideology, than skill. It worked in 2003, but when your desktop CPU is clocked just 25% higher than a phone, it's a big deal.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,038
8,526
106
What are you blabbering about?
Royal Core.
Apple already noticeably outperforms both Arrowlake and Zen 5 at just 4.5GHz
Axe SME memeboost and Not Really.
Again, big issue on the x86 camp
Nope.
but when your desktop CPU is clocked just 25% higher than a phone, it's a big deal.
Very true, shame that phones now juice 7-10W into 1t turbo making it very Not Usable in a phone.
Outside of fridge, anyway.

Again, everyone and their mom will be shipping 5G client cores.
Juice is good for you. Juice them up. More watts, more volts, more oomph is the meta. Enjoy!
 

DavidC1

Golden Member
Dec 29, 2023
1,616
2,671
96
Royal Core.
The team got fired for it basically.
Axe SME memeboost and Not Really.
X925 is ~30% faster per clock, while achieving fraction of the power.

So you believe it is because of the ISA?
Very true, shame that phones now juice 7-10W into 1t turbo making it very Not Usable in a phone.
Outside of fridge, anyway.
Not even Lunarlake uses that low and they are financially killing themselves to do that.

When there's 4-6x power difference at same performance, then one is making bad decisions.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,038
8,526
106
The team got fired for it basically.
Yeah but that's the point.
ILP ain't free.
X925 is ~30% faster per clock
Yeah it's far more chungus than something like Zen5 outside of SIMD.
while achieving fraction of the power.
Ehhh.
So you believe it is because of the ISA?
It's because you can't ship X5 in server without getting CBT.
Not even Lunarlake uses that low and they are financially killing themselves to do that.

When there's 4-6x power difference at same performance, then one is making bad decisions.
idk that's your point.
 
  • Like
Reactions: Io Magnesso

poke01

Diamond Member
Mar 8, 2022
3,733
5,077
106
I don't think they're really much faster in SIR or w/ever else.
Either way his point is moot since the juice is coming for everyone, all at once.
Apples core and Zen cores currently are simliar in non vector performance and Strix Halo proves this. SH is faster also due to SMT in 3D and cpu encoding etc but what is Apple better at is perf/w and single threaded apps.

Not the end for x86 at all. Zen6 will go even further
 

Doug S

Diamond Member
Feb 8, 2020
3,298
5,734
136
I flat out don't believe we have reached a fundamental perf/clock wall. I think the x86 vendors can do 50-60% better per clock with 2026 technologies. They are still infested with Netburst/Bulldozer ideologies. The reason is because they are both selling CPUs, and clockspeed still sells, even in supposedly all-ascended Anandtech Forum people.

Outside of forums like this nobody gives a sh-- about clock speed anymore. The days of marketing on clock rate when P4 was king are long gone. The average consumer has NO IDEA what clock speed a given CPU is, because it is all model numbers now and even if you dig for specs on a certain model you see a lot of hemming and hawwing about the actual clock speed since there's the default speed, the boost speed and then the extra special boost speed if you have really good cooling and there are asterisks about whether it is power limited or not and you have different boost speeds depending on how many cores are active. You think the average consumer could read that and mentally attach a single number to it?

Heck with the ridiculous profusion of SKUs most consumers couldn't even tell you what CPU is in their current PC beyond "Intel" or "AMD" (and I'm only about 70% confident they could name even that) Ask them if it is Zen 3 or Zen 4, or i5 or i7 and they've got no clue what you're even talking about.

Intel & AMD ship at the clock rates they do because that's what their designers believe deliver them the fastest CPUs. What Apple is doing is irrelevant - and Apple has been slowly catching up with clock rates so if that keeps up who knows they might be shipping at similar clock speeds by the end of the decade.
 

Io Magnesso

Senior member
Jun 12, 2025
502
138
71
Well, the ARM CPU will soon reach 5GHz sooner or later.
It's different if it's said that it can be stable for a long time to some extent.
 

Io Magnesso

Senior member
Jun 12, 2025
502
138
71
Royal Core.

Axe SME memeboost and Not Really.

Nope.

Very true, shame that phones now juice 7-10W into 1t turbo making it very Not Usable in a phone.
Outside of fridge, anyway.

Again, everyone and their mom will be shipping 5G client cores.
Juice is good for you. Juice them up. More watts, more volts, more oomph is the meta. Enjoy!
Certainly, if you think about it, The mobile SoCs released in 2024 are more efficient when they perform the same performance as the previous generation…
but If you apply a load to the maximum, you can easily consume 10W or more…
In particular, it seems that 8 Elite may consume 20W. It's an extreme example, but
 

poke01

Diamond Member
Mar 8, 2022
3,733
5,077
106
Intel & AMD ship at the clock rates they do because that's what their designers believe deliver them the fastest CPUs.
Also they do high clocks cause x86 cores do not go into phones. Imagine if you need 30 watts to reach optimal 1t to deliver similiar performance as a 285K in a phone. So they will aim for uarchs that will keep them <10 watts for maximum 1t in a phone
 

Io Magnesso

Senior member
Jun 12, 2025
502
138
71
The X925 is certainly amazing, but It seems to be quite large compared to the previous X4.
Well, there is a possibility that only the X925 core is made with high-performance cells.
 

Io Magnesso

Senior member
Jun 12, 2025
502
138
71
The team got fired for it basically.

X925 is ~30% faster per clock, while achieving fraction of the power.

So you believe it is because of the ISA?

Not even Lunarlake uses that low and they are financially killing themselves to do that.

When there's 4-6x power difference at same performance, then one is making bad decisions.
LNL is financially Self-destruct...?
 

511

Platinum Member
Jul 12, 2024
2,877
2,888
106
Apples core and Zen cores currently are simliar in non vector performance and Strix Halo proves this. SH is faster also due to SMT in 3D and cpu encoding etc but what is Apple better at is perf/w and single threaded apps.

Not the end for x86 at all. Zen6 will go even further
This is on an inferior N4P node vs N3E
 

Io Magnesso

Senior member
Jun 12, 2025
502
138
71
Compute tiles looks smaller comparatively to SoC but can't say anything until we get die size leak but it guess they put in some work to make the tile smaller i would be happy at 8+16 tile at 80-90mm2.
Certainly, it's closer to a square than the vertically long conventional one, isn't it
 

511

Platinum Member
Jul 12, 2024
2,877
2,888
106
TIL that RPL had a 68 Cycle L3 Cache they should get it under 55 Cycles with NVL.

a4c19a72-21bc-4c43-8631-78e1b312b1ce_1048x396.png
 

poke01

Diamond Member
Mar 8, 2022
3,733
5,077
106
Compute tiles looks smaller comparatively to SoC but can't say anything until we get die size leak but it guess they put in some work to make the tile smaller i would be happy at 8+16 tile at 80-90mm2.
So up to ~180mm2 of N2 die. Not bad.

I’m assuming the the other tiles are 18A?