Discussion Intel Nova Lake in H2-2026: Discussion Threads

[Geddagod]

No, because my Turin 64 core ACTUAL cores wild cream it.

Wow! Your HEDT/server product will CREAM an Intel client part? Crazy.

 

[Geddagod]

Intel's eDRAM based i7-5775c was able to do pretty well against even Comet Lake CPUs considering its age.

Was this not because Anandtech testing with JEDEC memory with both chips, making CML look worse than it should have been?

 

[511]

delete

What did I miss

 

[511]

BTW does Markfw means Mark for win?

 

[igor_kavinski]

Was this not because Anandtech testing with JEDEC memory with both chips, making CML look worse than it should have been?

Possible but not easy to see the article now.

 

[Saylick]

Per MLID:


Two compute tiles on TSMC N2P with large amounts of L3 cache to compete against AMD's V-cache. Talk about brute force.

 

[gdansk]

Per MLID:
View attachment 128488

Two compute tiles on TSMC N2P with large amounts of L3 cache to compete against AMD's V-cache. Talk about brute force.

I know this may be the delusions of MLID. But it's so big we all want it to be true, don't we?
20% 1T and 80% MT would be good.

 

[511]

I know this may be the delusions of MLID. But it's so big we all want it to be true, don't we?
20% 1T and 80% MT would be good.

I mean 80% MT is easily possible it's 2 Compute die on improved process so it's very possible also the node stuff he still gets it wrong man
Hub is 18A and
4+8 is 18AP and so is 4+0
I don't know if 512 EU Celestial exists cause it's 384 EU According to Raichu and it's SIMD 8 Count anyway so it will be 24Xe3P cores.
iGPU Tiles are Intel and so it the PCD

 

[511]

The most insane thing is able to overwrite exe

 

[gdansk]

The most insane thing is able to overwrite exe

Insanely stupid. I shan't be using it.
Imagine the poor developers telemetry being littered with impossible stack traces originating with Intel junk they didn't even write.

 

[igor_kavinski]

The most insane thing is able to overwrite exe

Replacing unoptimized code with highly optimized code. They can't keep doing it for more than a hundred games at most. Especially if the games receive regular patches. They would've to test the rewritten EXEs on their own. Additional burden on their software teams.

Probably going to do it in the most popular games to win some benchmarks.

 

[Joe NYC]

The most insane thing is able to overwrite exe

It is almost as if Intel is assuming that all the game development will be done on (and optimized for) AMD processors by the time Nova Lake ships.

 

[igor_kavinski]

Imagine the poor developers telemetry being littered with impossible stack traces originating with Intel junk they didn't even write.

Easy to fix that. Just compare the hash of the EXE and if it doesn't match, discard the telemetry data.

 

[Kepler_L2]

Replacing unoptimized code with highly optimized code. They can't keep doing it for more than a hundred games at most. Especially if the games receive regular patches. They would've to test the rewritten EXEs on their own. Additional burden on their software teams.

Probably going to do it in the most popular games to win some benchmarks.

Regular APO which is just thread affinity is only supported in like 10 games, there's no way "APO+" gets any real support outside of top 5 benchmarked games.

 

[igor_kavinski]

It is almost as if Intel is assuming that all the game development will be done on (and optimized for) AMD processors by the time Nova Lake ships.

The Zen5X3D damage has been astounding. Almost overnight, Intel CPUs were kicked out of GPU testbeds of major review sites.

 

[511]

Or simply accept Intel's help in devolving in the game and let them do this stuff at least Intel will do a better job of optimizing game than devs could do.

 

[Kepler_L2]

It is almost as if Intel is assuming that all the game development will be done on (and optimized for) AMD processors by the time Nova Lake ships.

I don't think it's an AMD vs Intel issue, it's just that most games are still not even using AVX, let alone AVX-512/AVX10, plus Nova Lake has APX which again won't be used in games for probably the next 10 years.

 

[igor_kavinski]

Regular APO which is just thread affinity is only supported in like 10 games, there's no way "APO+" gets any real support outside of top 5 benchmarked games.

If they use AI, maybe they get to 100 games

 

[511]

I don't think it's an AMD vs Intel issue, it's just that most games are still not even using AVX, let alone AVX-512/AVX10, plus Nova Lake has APX which again won't be used in games for probably the next 10 years.

Each of these require more binaries more dev effort we should be in transition though in next 3-4 Years though with game shipping up to date binaries along with comparability ones.

 

[igor_kavinski]

plus Nova Lake has APX which again won't be used in games for probably the next 10 years.

Good point. What if they splice APX instructions into the EXE to get an edge over AMD?

 

[511]

Intel software team is cracked anyway so I don't doubt they can pull that but in how many games though.

 

[511]

Good point. What if they splice APX instructions into the EXE to get an edge over AMD?

APX doesn't have instructions set it just has more physical register.

 

[igor_kavinski]

APX doesn't have instructions set it just has more physical register.

Change the existing code to USE those registers, hmmm???

 

[gdansk]

APX doesn't have instructions set it just has more physical register.

And if you want to use a new register...? Gotta use the REX2 prefix.
Which means devs will need to make binaries just for APX processors (not many will) or let Intel dynamically patch recompile their binaries (insane).

 

[511]

And if you want to use a new register...? Gotta use the REX2 prefix.

Yes it requires recompile either way. 🤣 Modern game devs don't even want to use x86_64_V3 man that is haswell level ISA.