Discussion Intel Nova Lake in H2-2026: Discussion Threads

[Io Magnesso]

The Chinese forum that leaked the info said that the P-core Haifa team is nearly devoid of personnel. Probably preparing for Unified Core transition. Not happy to have got to know you!

Yea but Novalake's LPE cores are really light load cores like Meteorlake was meant to do but failed. The performance contribution(if any) should be not meaningfully significant.

However, although the possibility is low, the story will change if the IO tile for desktop is used not only for HX but also for normal H and P laptop SKUs…

 

[OneEng2]

Oh no, N2 is a much, much faster node.

Finally something I agree is possible. I think the hot spots may limit the frequency on 18A... but it isn't all bad IMO. It may well be that 18A offers higher density than N2 because of routing. It also may offer lower power ..... which would be quite good in laptop and DC applications.

Without blowing off the BSPDN I doubt 18AP Could be faster but the hotspots will be difficult to cool.

Agree.

Halo chip configuration:

4 Beast Coves
8 Lion Coves
16 Skymonts

I have also wondered about this. Why not have 2 cores that are fat and fast as an example?

I like the idea.

Intel will be 52 core and AMD 24 (48 threads) core, how it can win in multithreading at all??

If the current relative performance per core between Zen 5 and Skymont is maintained, it can't .... at least not in anything that isn't using AVX512.

6.2-6.4 GHz I am expecting similarish clock between Zen6/NVL also finally someone will be dethroning 14900KS for the highest clock.

If it is true that NVL is only going to give 1.1x ST, I can't see AMD exceeding 6 Ghz by much if at all. Why would they?

t's 1t it's never power limited on desktop (anywhere, really).

It is not socket power limited. It might well be locally thermally limited. Power at higher clocks becomes much more super-linear. It most definitely becomes thermal limited.

Or are you suggesting that current processors are pipeline limited?

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.

Yes, but not across the board .... and at a cost. You never get something for nothing.

Yeah but that's the point.
ILP ain't free.

Indeed. Going from 1 to 2 pipes .... big gain. 2-4 .... a little gain. 4-8 .... not much at all.

This is also why almost no processors use SMT4. There just isn't a good PPA case for it.

*So I guess getting 50% margin on a future product means investing zero in uarch and just clocks?

Intel had to start thinking about cost. They should have been doing this years ago, but better late than never.

Will there be a loss of performance in Intel processors because of it? Absolutely. The day's of performance at ANY cost is over at Intel..... and it may have already gone on too long to save the company.

based on the slide the IPC Improvement can be extracted 15% from node and 33% from core count and dividing 1.7/(1.33*1.15) we get is likely 11-12%

Not a bad guess IMO.

^^^

Hopefully AMD is smart and only bothers with VCache models for DIY. Gonna need top tier gaming performance to justify the prices.

AMD can't survive on only the high end. Ask DEC and CRAY.

 

[Io Magnesso]

Yea.

"Slight improvements" are already expected with the very minor changes in Darkmont.

Majority of Intel also believed that they could easily reach 5, 6, 7 even 10GHz. Prescott rumors were mentioning 5.5GHz, then 6GHz, then 6.6GHz every quarter! In the end it whimpered and cratered just under 4GHz.

I think it's ridiculous that the e-core team is nothing special.
Even if it wasn't special, it's much more capable than an incompetent P-core team

 

[Io Magnesso]

Finally something I agree is possible. I think the hot spots may limit the frequency on 18A... but it isn't all bad IMO. It may well be that 18A offers higher density than N2 because of routing. It also may offer lower power ..... which would be quite good in laptop and DC applications.

Agree.

I have also wondered about this. Why not have 2 cores that are fat and fast as an example?

I like the idea.

If the current relative performance per core between Zen 5 and Skymont is maintained, it can't .... at least not in anything that isn't using AVX512.



If it is true that NVL is only going to give 1.1x ST, I can't see AMD exceeding 6 Ghz by much if at all. Why would they?

It is not socket power limited. It might well be locally thermally limited. Power at higher clocks becomes much more super-linear. It most definitely becomes thermal limited.

Or are you suggesting that current processors are pipeline limited?

Yes, but not across the board .... and at a cost. You never get something for nothing.

Indeed. Going from 1 to 2 pipes .... big gain. 2-4 .... a little gain. 4-8 .... not much at all.

This is also why almost no processors use SMT4. There just isn't a good PPA case for it.

Intel had to start thinking about cost. They should have been doing this years ago, but better late than never.

Will there be a loss of performance in Intel processors because of it? Absolutely. The day's of performance at ANY cost is over at Intel..... and it may have already gone on too long to save the company.

Not a bad guess IMO.

AMD can't survive on only the high end. Ask DEC and CRAY.

Agree, other ARMs are excellent and have narrowed the gap, but Apple has top-class performance in IPC as a dependency.

 

[poke01]

Yes, but not across the board .... and at a cost. You never get something for nothing.

What is the cost?

 

[511]

What is the cost?

 

[511]

https://twitter.com/x/status/1945317471197782405

 

[Io Magnesso]

https://www.zhihu.com/question/1920608785766028968/answer/1922041831182569570

Who is David

 

[Io Magnesso]

https://www.zhihu.com/question/1920608785766028968/answer/1922041831182569570

Who is David

Even if incompetence is ridiculed, I don't understand the meaning
There's no evidence that this man can be guaranteed to be an Intel employee

 

[Io Magnesso]

In the first place, Diamond Rapids is not LNC
It can be said that LNC is using the developed version of Cougar Cove...
However, the Cougar Cove is named Panther Cove on the DMR. It seems that some functions are not available in Cougar Cove.
At this point, it is clear that it is not the LNC as it is as it is
If you say that there is a core based on LNC, you can understand it.

 

[poke01]

https://twitter.com/x/status/1945317471197782405

I hope this becomes reality. I would finally get a new laptop if this happens. Since it’s Intel it won’t be stuck on one laptop.

 

[511]

I hope this becomes reality. I would finally get a new laptop if this happens. Since it’s Intel it won’t be stuck on one laptop.

Same I am planning to buy one if this happens with 128GB Ram

 

[MS_AT]

https://twitter.com/x/status/1945317471197782405

Is this Strix Halo equivalent?

 

[511]

Is this Strix Halo equivalent?

Yes with a big fat bus should definitely better based on Xe3P.

 

[MS_AT]

Yes with a big fat bus should definitely better based on Xe3P.

But this would be effectively Medusa Halo competitor? or the timelines are more favorable?

 

[511]

Yeah in the same

But this would be effectively Medusa Halo competitor? or the timelines are more favorable?

same timeline as Medusa Halo I don't know who is going to win lol with these expecting as close competition as LNL/HX 370.

 

[DavidC1]

I think it's ridiculous that the e-core team is nothing special.
Even if it wasn't special, it's much more capable than an incompetent P-core team

Doing nothing but getting the fourth cluster for 12-wide decode alone should bring "slight" improvements, nevermind the host of changes that accompany the width change in every such architecture.

They should pave the way for future of x86.

Back when Pentium M came out, then Core Duo, the performance gap between it and the Pentium 4 mobile parts(there were two variants, a desktop variant called Pentium 4 mobile and laptop optimized part called Pentium 4-M) were confusingly small.

I would not be surprised if that's the case with Novalake, where Arctic Wolf beats the P core companion by 3-5% per clock.

Mind you how Skymont performs is with Arrowlake's crippled uncore. Despite it, it outperforms the predecessor by 32%. With a proper subsystem, the true gains over the predecessor might be close to the 38% as in the early presentations. That means Lion Cove is also about ~15% faster than predecessor, not 10%.

They did hit 6Ghz just a 15-20 years later 🤣

By completely eating up overclocking headroom and with a heatsink that is the size of a brick and weighs much as a ultrabook.

And watercooling is common nowadays. Back then it was pseudo-exotic. When you watercooled your chip, it could clock far higher, in the range of 5GHz, full 30% higher than stock.

Look at how the top exotic overclocks hasn't budged too much in the past 20 years. This is a fundamental issue, one that should have been seen by engineers but they couldn't. It's a case where an outsider can see what experts cannot.

 

[511]

By completely eating up overclocking headroom and with a heatsink that is the size of a brick and weighs much as a ultrabook.

And watercooling is common nowadays. Back then it was pseudo-exotic. When you watercooled your chip, it could clock far higher, in the range of 5GHz, full 30% higher than stock.

Look at how the top exotic overclocks hasn't budged too much in the past 20 years. This is a fundamental issue, one that should have been seen by engineers but they couldn't. It's a case where an outsider can see what experts cannot.

TBH RPL is the one that budged Overclockers after FX Set the record

 

[DavidC1]

Yes with a big fat bus should definitely better based on Xe3P.

They better keep developing the dGPU ARC because I wouldn't buy an Intel GPU for my main system, especially not for a $2000 laptop. AMD options are far better due to compatibility and sustained performance over wide range of games even if Intel had a Halo iGPU part already.

The Iris Pro parts were a joke because of how bad the compatibility and widely varying performance was. They came a long way since then, but they still need long ways to go. If they worked on drivers and better hardware for the past decade, then they wouldn't have had to deal with this now.

It's like how keeping things organized every day lessens the workload versus not doing it and trying to do it 10 years later. The task becomes monumental. Every year they treated GPUs like HD Audio it put them further away from ever having a competitive GPU.

Intel ARC has less than 1% marketshare by the way. It didn't even register in JPR's data. Battlemage hype has done almost nothing. I am playing in my head of upgrading my desktop to play Expedition 33 and ARC isn't an option for me because of resale value and potential issues.

By the way I'm expecting ~15% for Novalake's P core per clock. Unless the >10% means close to 20%, I'm expecting zero clock gains.

 

[511]

They better keep developing the dGPU ARC because I wouldn't buy an Intel GPU for my main system, especially not for a $2000 laptop. AMD options are far better due to compatibility and sustained performance over wide range of games even if Intel had a Halo iGPU part already.

well LNL laptops are pretty decent but at 2000$+ it's not good lol

The Iris Pro parts were a joke because of how bad the compatibility and widely varying performance was. They came a long way since then, but they still need long ways to go. If they worked on drivers and better hardware for the past decade, then they wouldn't have had to deal with this now.

It's like how keeping things organized every day lessens the workload versus not doing it and trying to do it 10 years later. The task becomes monumental. Every year they treated GPUs like HD Audio it put them further away from ever having a competitive GPU.

the only thing that they kept developing was QSV and QSV is certainly the best in terms of support speed it would have been a similar story if they started dGPU 10 years ago.

Intel ARC has less than 1% marketshare by the way. It didn't even register in JPR's data. Battlemage hype has done almost nothing. I am playing in my head of upgrading my desktop to play Expedition 33 and ARC isn't an option for me because of resale value and potential issues.

By the way I'm expecting ~15% for Novalake's P core per clock. Unless the >10% means close to 20%, I'm expecting zero clock gains.

Cause they are afraid of screwing up like they did with alchemist also the wafers are at TSMC and they are not producing tons of these they needed to it on their own I3 process they wouldn't have a supply and cost problem

 

[DavidC1]

Don't take my criticism for Intel as an unsubstantiated one. I do it because I still have love for them. I hope the E core team continues to execute. Love what they've accomplished so far.

well LNL laptops are pretty decent but at 2000$+ it's not good lol

I like LNL other than the die waste/spyware unit. The GPU is not performant enough to be a main factor anyway. iGPUs at that level is always nice-to-have, but nothing more.

For me LNL is really about the battery life.

the only thing that they kept developing was QSV and QSV is certainly the best in terms of support speed it would have been a similar story if they started dGPU 10 years ago.

QSV alone is not enough. Plus, you can get the cheapest Intel iGPU and it's fine. Game support though, oh neglecting that is a big deal.

Say you had 1 million games in 2010 right? And it gets to get 1.5 million in 2015? What's easier work to do, in 2010 or 2015?

Their driver support for ARC has been good but it'll need work for 5-10 years. Meanwhile, AMD/Nvidia has been putting lot of game support for 30 years. The entire DX11 library has to be basically hand tuned for ARC.

 

[Io Magnesso]

Don't take my criticism for Intel as an unsubstantiated one. I do it because I still have love for them. I hope the E core team continues to execute. Love what they've accomplished so far.

I like LNL other than the die waste/spyware unit. The GPU is not performant enough to be a main factor anyway. iGPUs at that level is always nice-to-have, but nothing more.

For me LNL is really about the battery life.

QSV alone is not enough. Plus, you can get the cheapest Intel iGPU and it's fine. Game support though, oh neglecting that is a big deal.

Say you had 1 million games in 2010 right? And it gets to get 1.5 million in 2015? What's easier work to do, in 2010 or 2015?

Their driver support for ARC has been good but it'll need work for 5-10 years. Meanwhile, AMD/Nvidia has been putting lot of game support for 30 years. The entire DX11 library has to be basically hand tuned for ARC.

Well, QSV was good, and it's still good
I know the game support is a problem
However, if there is a problem with the new titles that appear in this world, it may be better to question the developer a little.
As for those who say that, it's acceptable that the optimization is lax, but they don't even check the operation.
Arc's Game support may never improve

 

[Io Magnesso]

Well, personally, it's better than AMD, which doesn't introduce the latest GPU architecture to the general laptop APU.

After all Even if Medusa Halo has the latest architecture, the variant is just a premium

Intel, which is making new initiatives, is more interesting and feels fresh

Even if you have a disability, it's fun to keep saying "Still" and see who moves forward.

 

[511]

Don't take my criticism for Intel as an unsubstantiated one. I do it because I still have love for them. I hope the E core team continues to execute. Love what they've accomplished so far.

also the foundry and other teams as well

I like LNL other than the die waste/spyware unit. The GPU is not performant enough to be a main factor anyway. iGPUs at that level is always nice-to-have, but nothing more.

it's fine for it's level

For me LNL is really about the battery life.

QSV alone is not enough. Plus, you can get the cheapest Intel iGPU and it's fine. Game support though, oh neglecting that is a big deal.

Yup

Say you had 1 million games in 2010 right? And it gets to get 1.5 million in 2015? What's easier work to do, in 2010 or 2015?

Their driver support for ARC has been good but it'll need work for 5-10 years. Meanwhile, AMD/Nvidia has been putting lot of game support for 30 years. The entire DX11 library has to be basically hand tuned for ARC.

don't forget professional support alongside Game support lol people using it for blender Davinci and other stuff as well also AI frameworks

 

[LightningZ71]

Medusa Halo will, unless AMD changes their ways, have to be on a post RDNA4 architecture. They don't have an completed N3 family design for RDNA4 as it's N4/N5 only so far. Same for RDNA3/3.5. It's going to have to be whatever comes next, RDNA5/Next/UDNA whatever. If they do add 20% more CUs (to 48 units) and go for a few speed grades faster RAM, it should be a notable improvement in performance over what they currently have.