Discussion Intel Nova Lake in H2-2026: Discussion Threads

[511]

E-cores could never clock high enough regardless of TDP.

We will see maybe they will do 5Ghz E cores with Arctic Wolf

 

[MS_AT]

I will always prefer core vs thread in my workload i never seen more then 5-10% from SMT or HT mostly less, and no i do not run crappy appz that need SMT to make it use core as it should

So which apps are these?

 

[Fjodor2001]

E-cores could never clock high enough regardless of TDP.

What is high enough? Also, note that the whole point is to instead have more cores, but clocked somewhat lower. To squeeze out the very last bit of perf from a core by bumping frequency above the range where it performs best costs a lot of power consumption, and results in poor perf/watt.

 

[Thunder 57]

What is high enough? Also, note that the whole point is to instead have more cores, but clocked somewhat lower. To squeeze out the very last bit of perf from a core by bumping frequency above the range where it performs best costs a lot of power consumption, and results in poor perf/watt.

I was replying to @511 saying NVL could match TR if they bumped it up to the same TDP. I think that was more of a joke and yes of course clocking cores high is less efficient. I'm sure Intel will clock them in their sweet spot whatever that ends up being.

 

[Fjodor2001]

I'm sure Intel will clock them in their sweet spot whatever that ends up being.

I hope so too. But based on how much TDP has increased on DT for both Intel and AMD in the last few CPU generations, I’m not so sure that’ll be the case.

There are some people on this forum (at least on Zen5) who have intentionally limited the TDP in BIOS settings to clock a bit lower, and notice that the performance drop then is not that big. So the stock TDP seems to have been set to squeeze out max perf to win the perf crown.

That said, for the 52C NVL-S SKU specifically I think all TDP will be needed, since there are so many cores. But what about the other SKUs in the lineup?

 

[Hulk]

I hope so too. But based on how much TDP has increased on DT for both Intel and AMD in the last few CPU generations, I’m not so sure that’ll be the case.

There are some people on this forum (at least on Zen5) who have intentionally limited the TDP in BIOS settings to clock a bit lower, and notice that the performance drop then is not that big. So the stock TDP seems to have been set to squeeze out max perf to win the perf crown.

That said, for the 52C NVL-S SKU specifically I think all TDP will be needed, since there are so many cores. But what about the other SKUs in the lineup?

I think the question of TDP is really dependent on workload.
For example, say you have a really well-threaded application. A sh$t ton of cores clocked in the sweet spot of the v/f curve will perform brilliantly. Also, an app that only used a handful of cores should run fine as well because while it might need to clock high, 5 or 6 cores won't present a gigantic power load when the other 40+ cores aren't doing anything.

The issue comes in with multitasking. If you are loading up the E cores and the P's at full clock and bumping up against the TDP then process is going to be very important to provide best performance from the architecture.

 

[coercitiv]

For example, say you have a really well-threaded application.

Let's make a list of these. Which apps scale to 24+ cores?

 

[511]

Let's make a list of these. Which apps scale to 24+ cores?

Blender/FFmpeg anything to do with CPU Based Rendering and Media.

 

[reb0rn]

Let's make a list of these. Which apps scale to 24+ cores?

Why 24 is it because of new zen6, this is pointless to argue..... same as we seen with 4 core and intel decade ago, same mindset

 

[MS_AT]

Let's make a list of these. Which apps scale to 24+ cores?

Every app that does not require active user input I mean you can always run N copies of it

Blender/FFmpeg anything to do with CPU Based Rendering and Media.

Doesn't this vary by codec for encoding? Different formats might have different sweet spots. I have seen many opinions that scaling dies for x265 past 16 cores. Not to mention it's also a function of what is being encoded/transcoded. Still a neat idea I have heard about is to split your original materials into parts and then transcode the parts in parallel what is giving better results timewise, not sure quality wise as I have not been experimenting with this myself.

For code compilation you generally require sufficient code base size and sane dependency graph, then it it might scale quite well.

 

[coercitiv]

Why 24 is it because of new zen6, this is pointless to argue.....

The Intel tile is 8+16.

 

[511]

For code compilation you generally require sufficient code base size and sane dependency graph, then it it might scale quite well.

Don't think that will be problem for many studios and companies with large code bases.

 

[MS_AT]

Don't think that will be problem for many studios and companies with large code bases.

Believe me, sane dependency graphs are a problem. The bigger your code base grows the harder it gets

 

[Fjodor2001]

Let's make a list of these. Which apps scale to 24+ cores?

Every app that does not require active user input I mean you can always run N copies of it

Good point actually.

Also, it doesn't have to be several instances of the same app that you run in parallel. A more realistic use case could perhaps be to for example run video transcoding in the background, and at the same time compiling SW, while also surfing the web or reading email, and whatever else you'd like to do in parallel.

 

[Hulk]

Let's make a list of these. Which apps scale to 24+ cores?

As some have mentioned video encoding and things like blender, but yes, there aren't a lot, which is why I wrote the real issue with TDP is multitaking. No one app will load up all those cores but I can run 3 or 4 at once that will slam the system. That is how I work, bursty, lots of things going on at once.

 

[OneEng2]

It is quite obvious that DC can eat up as many cores as can be applied to the system. If we are talking only about desktop applications, I agree. It's hard to see the need for lots of cores.

 

[deasd]

When comes to multi-core optimization, I completely agree what Raja said:





It's either GPU can complete tasks faster than CPU, or the MT scaling of majority of tasks have stagnated.

 

[LightningZ71]

In essence, there's only so much work to do, and eventually, it's more work to slice it up and administer all the slices than to keep fewer slices with higher locality.

 

[Keysplayr]

It's probably gonna be a clocked boosted Arrow Lake before they say bye-bye and forget it ever happened

200 series got a bad rap. It is perfectly fast I'm sure anyone who is using one can tell you. AMD has the gaming advantage for sure, but productivity with the 200 series is pretty good. IPC is good. I don't see the refresh (hopefully there is at least something) bringing that much improvement. Same die likely just tweaked where it can be tweaked.

 

[511]

200 series got a bad rap. It is perfectly fast I'm sure anyone who is using one can tell you. AMD has the gaming advantage for sure, but productivity with the 200 series is pretty good. IPC is good. I don't see the refresh (hopefully there is at least something) bringing that much improvement. Same die likely just tweaked where it can be tweaked.

Well in tech media if it can't game it's not good as for 48C/T it's not overly complicated to make it. The main Money Maker will be 8+16 and it's bin.