Discussion Intel Nova Lake in H2-2026: Discussion Threads

Page 19 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

511

Platinum Member
Jul 12, 2024
2,884
2,894
106
They said Pantherlake iGPU is 40% faster right? That's with a 50% higher compute unit count, so aside from niche scenarios(where Battlemage is weak at) there isn't much per EU/MHz perf gains.
I have heard 80% faster In synthetics also you seem to forget 8 Core MTL-H vs 8 Core LNL was 33% difference for the same cores but in TS and some other stuff it was not observable.
So 24 Xe3P is essentially Battlemage in a mobile form factor. Yes it has 20% more units but mobile is slower plus bandwidth is only about 2/3rd.
It will definitely be faster than B580 though for the main reason not suffering from overhead Xe3 is a big overhaul as big as Xe2 was vs Xe1
 

DavidC1

Golden Member
Dec 29, 2023
1,619
2,672
96
I have heard 80% faster In synthetics also you seem to forget 8 Core MTL-H vs 8 Core LNL was 33% difference for the same cores but in TS and some other stuff it was not observable.
You can't compare -H with Lunarlake. You have to compare with similar power Meteorlake. There's 15-20% gap there. Comparing to Meteorlake it's 15% faster even in synthetics. This has always been true, even in the Sandy Bridge days. The -H parts had ~20% advantage.

Arrowlake-H's higher power envelope catches up to Lunarlake even in lots of games.
It will definitely be faster than B580 though for the main reason not suffering from overhead Xe3 is a big overhaul as big as Xe2 was vs Xe1
There's usually a ~20% penalty for the mobile part and the absolute bandwidth is quite a bit less. It's only at 70%, plus it has to share which will be bulk of the performance difference(less effective bandwidth + penalties). It's not a small gap to make up for.

I have rule of thumbs for mobile vs desktop, iGPU vs dGPU:
-If the architecture is well balanced, iGPU is equal to 25% perf difference or roughly 1/2 the memory bandwidth
-5-7% faster per clock for a desktop CPU
-15-20% faster iGPU for -U vs -H
 

511

Platinum Member
Jul 12, 2024
2,884
2,894
106
You can't compare -H with Lunarlake. You have to compare with similar power Meteorlake. There's 15-20% gap there. Comparing to Meteorlake it's 15% faster even in synthetics. This has always been true, even in the Sandy Bridge days. The -H parts had ~20% advantage.

Arrowlake-H's higher power envelope catches up to Lunarlake even in lots of games.
that can be due to higher CPU perf as well not just GPU also ARL's iGPU is hybrid between Xe2/Xe1
There's usually a ~20% penalty for the mobile part and the absolute bandwidth is quite a bit less. It's only at 70%, plus it has to share which will be bulk of the performance difference(less effective bandwidth + penalties). It's not a small gap to make up for.
Yeah but we don't know the cache changes
I have rule of thumbs for mobile vs desktop, iGPU vs dGPU:
-If the architecture is well balanced, iGPU is equal to 25% perf difference or roughly 1/2 the memory bandwidth
-5-7% faster per clock for a desktop CPU
-15-20% faster iGPU for -U vs -H
we will see
 

DavidC1

Golden Member
Dec 29, 2023
1,619
2,672
96
that can be due to higher CPU perf as well not just GPU also ARL's iGPU is hybrid between Xe2/Xe1
And the better CPU matters in a non-upgradeable device like a laptop. Plus higher TDP means they both have more breathing room. The -U parts have more strict power management which impacts performance a bit.
Yeah but we don't know the cache changes
Those changes will need to be huge to make up for it.
we will see
Fundamentals do not change. Also, being conservative usually is right in this space, since hype is the game in town.
Is it? From open source patches so far it looks like a pretty minor update
FWIW, Exist50 said Xe3 changes make Xe2 changes look minor in comparison.

They missed many potential fixes in BMG such as ReBar requirements, the huge disadvantages in die and power use, or even the idle power management block.

The two GPU competitors have been honing their craft for 2 decades. The problems above are basics that should have been a nothingburger by now.
 
  • Haha
Reactions: Io Magnesso

DavidC1

Golden Member
Dec 29, 2023
1,619
2,672
96

Description​

Image Sharpening, Adaptive Tesselation, and Anisotropic Filtering are not working as expected. Enabling any of these options results in no noticeable change.

Resolution​

Image sharpening, Adaptive Tesselation, and Anisotropic Filtering features were removed from IGS version 25.22.1502.2, included in driver 32.0.101.6913. From this versions onwards, these features won't be available.
Yeaaa, Intel has a LONG way to go. They should focus on making it work. What a poor excuse.

When I had an Nvidia card I appreciated how I could force those features in game if I wanted(Geforce 2), even if the game didn't have support for them. Could not do that on AMD or Intel.
 
  • Like
Reactions: Thibsie

511

Platinum Member
Jul 12, 2024
2,884
2,894
106
They missed many potential fixes in BMG such as ReBar requirements, the huge disadvantages in die and power use, or even the idle power management block.
ReBar is not getting fixed tbh PPA and overhead are getting fixes though
The two GPU competitors have been honing their craft for 2 decades. The problems above are basics that should have been a nothingburger by now.
It's not that easy it seems you always forget some stuff due to deadlines and what not lol
 
  • Like
Reactions: Io Magnesso

Tup3x

Golden Member
Dec 31, 2016
1,241
1,363
136

Yeaaa, Intel has a LONG way to go. They should focus on making it work. What a poor excuse.

When I had an Nvidia card I appreciated how I could force those features in game if I wanted(Geforce 2), even if the game didn't have support for them. Could not do that on AMD or Intel.
NVIDIA currently has issues with AF on 5000 series cards. In some games it causes serious artifacts. I do hope that they fix it instead of removing it. There are so many games with lacking of just broken anisotropic filtering setting. Heck, some even use bilinear filtering... Sadly it never worked in Vulkan apps (which sucks). Negative lod can help especially if game has DLAA/DLSS/good TAA support.
 

ondma

Diamond Member
Mar 18, 2018
3,276
1,679
136
While we are speculating, lets go even further. Rumor: https://www.notebookcheck.net/Leake...-the-last-P-core-E-core-design.1059015.0.html

Predicting 100 unified cores for 2028. I thought this looked really interesting, and Notebook Check IMO isnt a junk site like WCCF Tech, so I thought it was interesting. Then I noticed that the source was an X post by our old friend on these forums, SiliconFly. So I immediately added a ton of salt to the post, but hopefully it might come true. Not sure who would need a hundred cores, but a more efficient unified core would certainly be welcome.
 

511

Platinum Member
Jul 12, 2024
2,884
2,894
106
Not my (our) problem. Intel wants to enter a new market they should be competitive.
Oh yeah definitely i never said they shouldn't be competitive lol but reBar should be the norm tbh.

also based on the leaked Panther Lake iGPU a 24Xe3 Core should be roughly ~106mm2 N3E or slightly smaller on 18AP(~100mm2)

While we are speculating, lets go even further. Rumor: https://www.notebookcheck.net/Leake...-the-last-P-core-E-core-design.1059015.0.html

Predicting 100 unified cores for 2028. I thought this looked really interesting, and Notebook Check IMO isnt a junk site like WCCF Tech, so I thought it was interesting. Then I noticed that the source was an X post by our old friend on these forums, SiliconFly. So I immediately added a ton of salt to the post, but hopefully it might come true. Not sure who would need a hundred cores, but a more efficient unified core would certainly be welcome.
Yeah seems unlikely also wonder what they are targeting for Unified should be between 25-30%+ over Arctic Wolf cause this is a big change
 

ToTTenTranz

Senior member
Feb 4, 2021
456
842
136
I don't know if it's axed, could just be delayed for LPDDR6 or they're skipping it and going straight to next-gen (i.e Zen7+RDNA6 in 2028 instead of Zen6+RDNA5 in 2027).

Called it.

If Strix Halo fails to move the needle in getting high end design wins for AMD, Medusa Halo will be cancelled.



Though to be honest, it doesn't look like Medusa Halo with RDNA3.5 and LP5X has much of a reason to exist. Without a substantial boost in memory bandwidth, couldn't they just pair a Strix Halo IOD with Zen6 CCDs?
 

mikk

Diamond Member
May 15, 2012
4,291
2,381
136
They said Pantherlake iGPU is 40% faster right? That's with a 50% higher compute unit count, so aside from niche scenarios(where Battlemage is weak at) there isn't much per EU/MHz perf gains.

So 24 Xe3P is essentially B580 in a mobile form factor. Yes it has 20% more units but mobile is slower plus bandwidth is only about 2/3rd.

Looks like from eyeballing and rough search it should be fairly competitive performance-wise with AMD.


There is no bandwidth increase for PTL-H over LNL, a linear increase needs a good GPU design improvement in this case. And no, Intel didn't say 40%, they haven't said anything. It is way too early to claim not much EU/MHz gains.
 

511

Platinum Member
Jul 12, 2024
2,884
2,894
106
PTL will be bandwidth starved even more than LNL it won't have enough mem bandwidth to feed the 12Xe3 iGPU.
This is just bruteforcing with EU Count.
 

511

Platinum Member
Jul 12, 2024
2,884
2,894
106
Fortunately they have doubled the L2 iGPU cache size. Better than nothing. Xe2 to Xe3 isn't just EU count bruteforcing.
There are architecture improvement but it will be held back by not enough mem bandwidth
 

OneEng2

Senior member
Sep 19, 2022
673
920
106
While we are speculating, lets go even further. Rumor: https://www.notebookcheck.net/Leake...-the-last-P-core-E-core-design.1059015.0.html

Predicting 100 unified cores for 2028. I thought this looked really interesting, and Notebook Check IMO isnt a junk site like WCCF Tech, so I thought it was interesting. Then I noticed that the source was an X post by our old friend on these forums, SiliconFly. So I immediately added a ton of salt to the post, but hopefully it might come true. Not sure who would need a hundred cores, but a more efficient unified core would certainly be welcome.
Yea, I saw that too.

I am finding it hard to see how people will utilize 52 cores. I keep thinking that all that die space could do so much more and for more people if it were used a different way.

While 2028 is still a few years away, it is STILL hard to imagine that so many cores will be useful in consumer electronics.

Going a bit further, how will you feed such a beast? I suppose you could move the desktop up to higher bandwidth memory than DDR8000 (perhaps RDIMMs at 16000MT?)... or move to a quad channel memory setup?

Still, it's an awful big cost in die size, MB layout, and RAM to cater to cinebench isn't it?
 

DavidC1

Golden Member
Dec 29, 2023
1,619
2,672
96
Predicting 100 unified cores for 2028. I thought this looked really interesting, and Notebook Check IMO isnt a junk site like WCCF Tech, so I thought it was interesting. Then I noticed that the source was an X post by our old friend on these forums, SiliconFly.
Unified Cores will bring efficiency, but the problem is node gains are crashing. 14A is just 30% more density, meaning in terms of density it's equal to 1/3 of a node. Basically you need 6 years to get same density gains as 2 years.

If Arrowlake with 2 process node gains when the gains were big with new uarch in both cores got us what we got now, even if we assume uncore wasn't terrible. This is also why 60% for Novalake isn't that surprising. The bad part of Novalake isn't MT(which is really good), but ST.

What happens when density part crashes to 1/3 with future?
 

Io Magnesso

Senior member
Jun 12, 2025
506
138
71
PTL will be bandwidth starved even more than LNL it won't have enough mem bandwidth to feed the 12Xe3 iGPU.
This is just bruteforcing with EU Count.
No, it's hard to imagine that there will be a big difference or a disadvantage...
 

Io Magnesso

Senior member
Jun 12, 2025
506
138
71

Yeaaa, Intel has a LONG way to go. They should focus on making it work. What a poor excuse.

When I had an Nvidia card I appreciated how I could force those features in game if I wanted(Geforce 2), even if the game didn't have support for them. Could not do that on AMD or Intel.

Well, I don't know what you like to say
GeForce 2 is an antique
Even AMD and NVIDIA are terrible
As long as you don't criticize bad things, it won't get better
 

Io Magnesso

Senior member
Jun 12, 2025
506
138
71
And the better CPU matters in a non-upgradeable device like a laptop. Plus higher TDP means they both have more breathing room. The -U parts have more strict power management which impacts performance a bit.

Those changes will need to be huge to make up for it.

Fundamentals do not change. Also, being conservative usually is right in this space, since hype is the game in town.

FWIW, Exist50 said Xe3 changes make Xe2 changes look minor in comparison.

They missed many potential fixes in BMG such as ReBar requirements, the huge disadvantages in die and power use, or even the idle power management block.

The two GPU competitors have been honing their craft for 2 decades. The problems above are basics that should have been a nothingburger by now.

Is it power consumption?
This problem was also found in RDNA3, which consumes high power when idle. (Improved with rDNA4)
NVIDIA is Improved power management with Blackwell
They're not perfect, they're just moving forward to make it better Don't forget that effort