Discussion Intel Nova Lake in H2-2026: Discussion Threads

[branch_suggestion]

https://twitter.com/x/status/1938962228923027497

Intel official NVL performance estimate.

>1.1x ST
1.6x MT


View attachment 126355

Seems about right but nT uplift is very underwhelming for 2x the cores, maybe a bunch of poorly scaling benchmarks or actually keeping power to a similar PL2? Kinda shows that membw is a thing that is important, yes.
1T is DOA, Z6 should be >1.2x vs Z5.
Meanwhile in the land of pure Krzanich brand hopium.

https://twitter.com/x/status/1938941739781558679

 

[511]

Seems about right but nT uplift is very underwhelming for 2x the cores, maybe a bunch of poorly scaling benchmarks or actually keeping power to a similar PL2? Kinda shows that membw is a thing that is important, yes.
1T is DOA, Z6 should be >1.2x vs Z5.
Meanwhile in the land of pure Krzanich brand hopium.

https://twitter.com/x/status/1938941739781558679

This dude is best ignored

 

[branch_suggestion]

This dude is best ignored

Oh I know, all the new wave Intel twitter folk are quite special.

 

[CouncilorIrissa]

https://twitter.com/x/status/1938962228923027497

Intel official NVL performance estimate.

>1.1x ST
1.6x MT


View attachment 126355

Is that even NVL and not, say, PTL?

 

[511]

Is that even NVL and not, say, PTL?

What will they be comparing PTL to to have this much gain?

 

[CouncilorIrissa]

What will they be comparing PTL to to have this much gain?

Comparing to LNL in nT would give this kind of figures methinks.
It's just that it's way too early for this kind of a slide to appear about NVL.

 

[511]

ARL Slide also appeared early and we were doubting those and those came to be true

 

[Kepler_L2]

Is that even NVL and not, say, PTL?

It's Nova Lake

 

[MS_AT]

https://twitter.com/x/status/1938962228923027497

Intel official NVL performance estimate.

>1.1x ST
1.6x MT


View attachment 126355

The bit about leadership gaming performance makes sense, with 1.6 MT performance maybe you will be able to play 2 games at the same time. Or at least ask AI to play them for you.

 

[511]

The bit about leadership gaming performance makes sense, with 1.6 MT performance maybe you will be able to play 2 games at the same time. Or at least ask AI to play them for you.

Ofc the NPU wil bring gaming performance improvement /s.

 

[adroc_thurston]

It's just that it's way too early for this kind of a slide to appear about NVL.

no it's not, it's a q4'26 launch, pre-Si viz is available everywhere.

 

[igor_kavinski]

ARL Slide also appeared early and we were doubting those and those came to be true

Funny because Intel must've known how bad ARL's roughly 5% ST uplift looked but did nothing to improve the situation through any means necessary.

When I first saw those ARL slides, I was like, ok maybe that's projected perf uplift. Most likely they will do something drastic to at least push the halo part to greater than 10% IPC (through frequency gain or special tweaked die). One more shred of evidence that Pat wasn't concerned about consumer CPUs at all and this is where they took the worst reputation hit. When 265K is selling for close to $200, you know that only a miracle will save them from a disastrous quarter.

 

[adroc_thurston]

how bad ARL's roughly 5% ST uplift looked but did nothing to improve the situation through any means necessary.

Well how could they? LNC missed like every target in existence.

 

[511]

Funny because Intel must've known how bad ARL's roughly 5% ST uplift looked but did nothing to improve the situation through any means necessary.

You can't magically improve the situation this late it was projected at 1.05 and it was roughly in the ballpark if you want to improve arrow lake would have come out Q4 25.

 

[511]

Well how could they? LNC missed like every target in existence.

They even removed HT and failed spectacularly with LNC.

 

[igor_kavinski]

if you want to improve arrow lake would have come out Q4 25.

That SHOULD have happened! Better to delay than lose face.

 

[511]

That SHOULD have happened! Better to delay than lose face.

While you think this is right it's not they had commitments to customers it would have caused a lot more if they have delayed it.

 

[adroc_thurston]

They even removed HT and failed spectacularly with LNC.

Kinda the opposite.

Better to delay than lose face.

Nope.

While you think this is right it's not they had commitments to customers it would have caused a lot more if they have delayed it.

ding ding ding

 

[511]

Kinda the opposite.

What do you mean?

 

[igor_kavinski]

ding ding ding

As if their customers didn't have enough of previous gen to sell. Few more months wouldn't have mattered.

 

[adroc_thurston]

As if their customers didn't have enough of previous gen to sell. Few more months wouldn't have mattered.

that's not how product refreshes work my friend.

 

[igor_kavinski]

What do you mean?

I suspect Lion Cove performed pretty good with HT but they wanted to save time on validation. It may also have made it harder for them to hit the TDP targets they decided to go with in the end.

 

[igor_kavinski]

that's not how product refreshes work my friend.

They cancelled MTL-S. Doesn't seem like anyone divorced Intel over that.

 

[511]

They cancelled MTL-S. Doesn't seem like anyone divorced Intel over that.

They sold RPL refresh in exchange for that along with RPL

 

[igor_kavinski]

They sold RPL refresh in exchange for that along with RPL

Seriously, textbooks will be written on how Pat fumbled the whole thing. He had complete control. Kept making grave mistakes. Over and over. His problem is that he didn't take much interest in engineering matters. Left it up to the fools. The same fools who came up with Rocket Lake and then Raptor Lake. I wouldn't be surprised if disabling AVX-512 was Pat's decision and it was made in a desperate attempt to upsell the Xeon W CPUs which are still priced extremely high. Intel had no interest in making money. Just wanted to keep up appearances of being their former self. A technology leader. Without any technical leadership.