Discussion Intel Nova Lake in H2-2026: Discussion Threads

[Io Magnesso]

wow, thats a basically a dead core. 4 years of clock stagnation. If Intel doesn't use Nanoflex to its advantage they have bad designers, it should be easy to surpass 6GHz day 1 of NVL-S release. No need for a special sku, this isn't 2022 anymore.

It's not a bad thing to create a processor with high performance cells
It depends on the purpose

 

[Io Magnesso]

5.8Ghz for "top" SKU followed in 6 months by the 6Ghz edition.

Nova Lake is the top 5.3GHz
Zen6 is the highest in the top 5.1 to 5.2 GHz
As for why the clock has dropped since the previous generation...
In other words, the limit of the tunnel effect has arrived…

 

[csbin]

Zen6: 10%+ IPC
NVL: >10% ST Performance

https://twitter.com/x/status/1940032798422978631

 

[dullard]

No you won't. Power isn't free. Inverse square law says if you straight up increase everything, twice the perf/clock needs four times the power.

Inverse square law only applies to single core. You can also get twice the performance at the same clock with just double the power if you double the cores. This, of course, assumes the application can handle double the threads without Amdahl's law kicking in.

 

[Jan Olšan]

DOA ST Performance

Keep in mind it is not "+10 %" but "greater than +10 %". Same case as the Zen 5/6 IPC leaks.

 

[MS_AT]

Keep in mind it is not "+10 %" but "greater than +10 %". Same case as the Zen 5/6 IPC leaks.

Problem is it is worded as > 10% overall performance what suggests its close to 10 and does not allow for frequency hopium of course this might be marketing plot but given Intel's recent track record it is the less likely possibility

 

[511]

Keep in mind it is not "+10 %" but "greater than +10 %". Same case as the Zen 5/6 IPC leaks.

The problem is 10%+ ipc And 8% more clocks (6.2GHZ) will net you ~20% ST while Intel's is gonna be Greater than 1.1X

 

[dullard]

Problem is it is worded as > 10% overall performance what suggests its close to 10 and does not allow for frequency

I probably missed the juicy details, but the leak that I saw said exactly

Leadership Gaming Performance
>1.1x higher ST and 1.6x MT performance1

So, could you please help me out as to how that suggests there is no frequency change? Also, what is the footnote 1? What is this compared to? And why are people talking about anything but gaming performance on this leak? The >1.1x ST probably also applies to general performance, but that was not explicitly stated.

 

[adroc_thurston]

So, could you please help me out as to how that suggest there is no frequency change

10% is the minimum I'd expect from a tock which leaves about whole 1% left for the frequency gain.

 

[dullard]

10% is the mining I'd expect from a rock which leaves about whole 1% left for the frequency gain.

You could make that claim if it was iso-power. But do we even know that?

 

[adroc_thurston]

You could make that claim if it was iso-power. But do we even know that?

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

 

[Doug S]

Why does it need to clock at a certain frequency?

Since we know ST gains are 1.1x, it would be beneficial for Intel if they can achieve that at noticeably below 5.7GHz.

No one cared about clock stagnation from 2006 to 2011.

You can't predict that exactly. It's beyond silly when Intel/AMD claims exact numbers. 17% really? Why not 16%? Just say 15-20% like in the old days and be done with it. Or, undersell and call it 15%.

Yes, THIS!

Why are people so hung up on frequency? If you make changes to the core that increase IPC like widening it or shortening the pipeline you can gain performance without increasing the clock. That would be PREFERABLE to Intel & AMD having a race to see who's first to 7 GHz if you care at all about power.

If news got out that Intel's top frequency was going to be 4.5 GHz people here would lose their minds and be talking about how Intel is going bankrupt next week. If news got out that Intel was hitting 4000 in GB6 people would be saying "Intel is back, baby!" Those are M4's numbers for those who don't recognize them, proving it is quite possible to have high performance without having to go anywhere near 6 GHz.

 

[dullard]

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

Do we even know the leak is about desktop?

I'm basically trying to see where all the assumptions are before analyzing the comments.

 

[adroc_thurston]

Do we even know the leak is about desktop?

Yeah.

I'm basically trying to see where all the assumptions are before analyzing the comments.

That's not an assumption, that's Intel pre-Si viz for NVL-S 16+32.

 

[dullard]

That's not an assumption, that's Intel pre-Si viz for NVL-S 16+32.

Ah, finally a useful response. The leak that I saw had none of that information.

 

[MS_AT]

So, could you please help me out as to how that suggests there is no frequency change? Also, what is the footnote 1? What is this compared to? And why are people talking about anything but gaming performance on this leak? The >1.1x ST probably also applies to general performance, but that was not explicitly stated.

That might have been lost in the context. My response was, the leak was suggesting overall >1.1 performance uplift, so if all of that came from IPC, there would be no room for frequency uplift. Since it's a new node, there will be some freq update, what makes high ( that is higher than 10%) ipc uplift impossible as most likely some of the >1.1 uplift comes from frequency. Since it's worded as >1.1 it makes it likely that the expected overall improvement is close to 10%, as if it was close to 20% they would cite >1.15 etc. Of course the last part is just me reaching out to a conclusion and can be completely wrong.

On the other hand, not sure why you thought I had anything gaming performance related in mind.

 

[dullard]

That might have been lost in the context. My response was, the leak was suggesting overall >1.1 performance uplift, so if all of that came from IPC, there would be no room for frequency uplift. Since it's a new node, there will be some freq update, what makes high ( that is higher than 10%) ipc uplift impossible as most likely some of the >1.1 uplift comes from frequency. Since it's worded as >1.1 it makes it likely that the expected overall improvement is close to 10%, as if it was close to 20% they would cite >1.15 etc. Of course the last part is just me reaching out to a conclusion and can be completely wrong.

I'm trying to put together a full picture. Assumptions are certainly allowed and may be wrong. But, I want to at least get the assumptions together.

Lets assume:

• Assume Adroc is correct that >1.1x ST and 1.6x MT applies to the Nova Lake-S 16P + 32E chip.
• Assume this is compared to the Arrow Lake 285K.
• Assume PL1 stays the same at 125W and PL2 stays the same at 250 W for both chips.
• Assume this is on TSMC N2 which is about 25% to 30% lower power at the same frequency than N3B in Arrow Lake.

Then each tile in MT gets only 125W. Each core gets half the power in MT. So, your assumption that there is a frequency uplift seems a stretch to me--at least in MT at PL2 power levels. At half power per core, I would expect a frequency decrease even with a new node. Since N2 is up to 30% more power efficient, but there is a 50% power decrease per core, I don't see how frequency can even stay the same. It'll have to drop signficantly.

On the other hand, not sure why you thought I had anything gaming performance related in mind.

The leak specifically mentioned gaming performance. We are assuming that it applies to general performance, and it probably does. But it is still just an assumption.

 

[adroc_thurston]

Then each tile in MT gets only 125W. Each core gets half the power in MT. So, your assumption that there is a frequency uplift seems a stretch to me--at least in MT at PL2 power levels. At half power per core, I would expect a frequency decrease even with a new node.

Oh, no one cares about nT.
We're talking 1t freq (or lack thereof).

 

[Kepler_L2]

Assume PL1 stays the same at 125W and PL2 stays the same at 250 W for both chips.

It's 150W PL1 and 320W PL2

 

[adroc_thurston]

and 320W PL2

what da hell.
If this is what they need for 60% moar nT then Atom guys need to commit sudoku too.

 

[MS_AT]

Then each tile in MT gets only 125W. Each core gets half the power in MT. So, your assumption that there is a frequency uplift seems a stretch to me--at least in MT at PL2 power levels.

I was talking only about >1.1 ST improvement and if we assume that the improvement does not come from single source (either IPC or freq) it has to be a sum of those, what rules out either of those could be increased by 10% at the same time. That's all.

 

[dullard]

I was talking only about >1.1 ST improvement and if we assume that the improvement does not come from single source (either IPC or freq) it has to be a sum of those, what rules out either of those could be increased by 10% at the same time. That's all.

Note: I originally quoted you simply because you were the latest relevant post to frequency. I'm talking about MT where you can get a better estimate of IPC and frequency. Trying to do this with 1T is a fool's errand as Adroc said, you could throw essentially infinite power to it have massive frequencies and get no real useful estimates.

 

[dullard]

It's 150W PL1 and 320W PL2

Do you have a link? 320 W PL2 requires new motherboard design, new heat sinks, etc.

 

[adroc_thurston]

320 W PL2 requires new motherboard design, new heat sinks, etc.

well yeah it is a new socket and a new platform.
a true Intel classic.

I'm talking about MT where you can get a better estimate of IPC and frequency

No you don't.

you could throw essentially infinite power to it have massive frequencies and get no real useful estimates.

You can't, Vmax is one for all.

 

[poke01]

Do you have a link? 320 W PL2 requires new motherboard design, new heat sinks, etc.

It also requires more memory bandwidth