Discussion Intel Nova Lake in H2-2026: Discussion Threads

[Fjodor2001]

I don't think Intel will give NVL-S 48 PCIe5 lanes + additional PCIe lanes via chipset. That would make the mobos too expensive for the consumer line.

Latest I've seen is that NVL-S will have 32 PCIe 5.0 lanes and 16 PCIe 4.0 lanes. Source:

https://videocardz.com/newz/intel-nova-lake-s-reportedly-supports-ddr5-8000-memory-and-offers-36-pcie-5-0-lanes

Are you saying that'll not be sufficient for most MT workloads on ~50C/T? What are you expecting them to be used for?

About memory BW, as I said you need to have 12800MT/s DC to match 6400MT/s QC.

Yes, agreed. But who says you need all that memory bandwidth for most MT use cases on 52C/T?

Also, there are other aspects to consider, like the memory controller, caches, etc.

Intel or you? All I am saying is that Nova Lake will not invalidate TR as a product line.

Intel, since they are the ones releasing it. So it means they think there is a market for NVL-S 52C on DT, with the memory bandwidth and PCIe lane config it has.

It will not invalidate all TR SKUs, but several of them. There'll be many MT workloads which will be handled much cheaper using NVL-S than TR.

Regarding the low core count TR SKUs, it must be a real niche market. E.g. someone that needs a lot of RAM but little CPU perf? But as long as they have the TR range anyway, I guess they might as well include them in it.

For those TR SKUs which still will be relevant after NVL-S (e.g. due to need for 64C+, or more total RAM), could EPYC SKUs not be used instead? Then they'd only need to maintain one combined HEDT/Server CPU range and platform.

 

[LightningZ71]

Keep costs in mind too. Entry level TR workstations are in the $4k and up price range, before you add and RAM or storage and doesn't even include a video output. To get even a display lit up, you blow a whole PCIe slot.

You can probably get a really nice NVL setup before you even approach that price.

 

[reb0rn]

TR is just way overpriced for home user that need more power, NVL will not be, that is fact

 

[Fjodor2001]

TR is just way overpriced for home user that need more power, NVL will not be, that is fact

Agreed. But that was not always the case.

Remember when Intel was doing 4C forever, and you had to pay a lot for their HEDT to go above that core count? Then AMD turned the tables with 8C Zen1 in 2017. In the same year AMD also provided 16C TR, reasonably priced at $999. So TR blew Intel HEDT out of the water.

Now the tables will turn. AMD is stuck at 16C forever (sure 24C soon, but still), and you have to pay a lot for TR to go above that core count. But this time Intel will turn the tables, and provide 52C NVL-S at reasonable price, to blow TR out of the water.

Oh, how the times are changing...

 

[Thunder 57]

Agreed. But that was not always the case.

Remember when Intel was doing 4C forever, and you had to pay a lot for their HEDT to go above that core count? Then AMD turned the tables with 8C Zen1 in 2017. In the same year AMD also provided 16C TR, reasonably priced at $999. So TR blew Intel HEDT out of the water.

Now the tables will turn. AMD is stuck at 16C forever (sure 24C soon, but still), and you have to pay a lot for TR to go above that core count. But this time Intel will turn the tables, and provide 52C NVL-S at reasonable price, to blow TR out of the water.

Oh, how the times are changing...

We've had this discussion before.

 

[511]

NVL is fine as a platform tbh just need to see the leaked benchmarks for bLLC.

 

[Fjodor2001]

We've had this discussion before.

I think it’s more that you don’t like the outcome of that particular discussion, so you want it silenced.

We’ve also discussed caches, pricing, release date, iGPU, bLLC, performance, process tech, and more or less everything else before, but we keep discussing that too.

 

[Thunder 57]

I think it’s more that you don’t like the outcome of that particular discussion, so you want it silenced.

We’ve also discussed caches, pricing, release date, iGPU, bLLC, performance, process tech, and more or less everything else before, but we keep discussing that too.

It's more about it already being discussed thoroughly but there is still new info dripping every day it seems. I don't care what the outcome is I just disagree with your points. More cores were far more useful back then than today. Just my opinion but I think sales will prove it.

 

[Fjodor2001]

It's more about it already being discussed thoroughly but there is still new info dripping every day it seems. I don't care what the outcome is I just disagree with your points. More cores were far more useful back then than today. Just my opinion but I think sales will prove it.

Yes, agreed about that even though some topics have been discussed before, often there are new aspects coming up (mixed with old ones) when they are revisited. Same in this case.

I figured it was mostly that you don’t find more cores useful, so you don’t like to be reminded of that it’s the direction we’re headed in whether you like it or not. Both Intel and AMD are heading in that direction, with NVL-S and Zen6.

Of course it would be better to increase performance an equal amount by bumping IPC or frequency instead if possible, but currently that’s not an option. Both have hit a (soft) wall. So the best option available currently is to increase core count.

 

[511]

8+16 or 12C/24T would be the more mainstream SKU imo

 

[Fjodor2001]

8+16 or 12C/24T would be the more mainstream SKU imo

We already have such SKUs available with current gen, and we’re also going to have similar options available with Zen6 and NVL-S which those that don’t need more performance than what they provide can get.

 

[fastandfurious6]

leaks when!

 

[511]

leaks when!

We are waiting for you to leak stuff 😛

 

[regen1]

So what do we know about P-cores and E-cores apart from shared L2
and support for AVX10.2/AVX512, APX, FRED ?

P-cores go 12Wide and E-cores go wider(guessing 12Wide as well?)

 

[ondma]

Agreed. But that was not always the case.

Remember when Intel was doing 4C forever, and you had to pay a lot for their HEDT to go above that core count? Then AMD turned the tables with 8C Zen1 in 2017. In the same year AMD also provided 16C TR, reasonably priced at $999. So TR blew Intel HEDT out of the water.

Now the tables will turn. AMD is stuck at 16C forever (sure 24C soon, but still), and you have to pay a lot for TR to go above that core count. But this time Intel will turn the tables, and provide 52C NVL-S at reasonable price, to blow TR out of the water.

Oh, how the times are changing...

I hope they are changing. I hope Intel can at least become competitive. I think that is the best one can hope for.

 

[Thunder 57]

So what do we know about P-cores and E-cores apart from shared L2
and support for AVX10.2/AVX512, APX, FRED ?

P-cores go 12Wide and E-cores go wider(guessing 12Wide as well?)

My 8 ball says hell no on 12 wide. Are they bringing back HT?

 

[511]

My 8 ball says hell no on 12 wide. Are they bringing back HT?

No HT on P cores but do you need HT with 24C or 48C not counting the LPE?

 

[Thunder 57]

No HT on P cores but do you need HT with 24C or 48C not counting the LPE?

If you went to a 12 wide decoder than I'd say yes. That's why I don't see it happening.

 

[511]

If you went to a 12 wide decoder than I'd say yes. That's why I don't see it happening.

The 12 wide would like be cluster decoding not full fat 12 Decoders like in LNC.

 

[Josh128]

Its very doubtful that the 52 core non-HT Nova Lake is going to significantly outpace the 24 core HT Zen 6 in MT workloads if both are on the same node.

Why? Because right now, Arrow Lake has 50% more cores than Zen 5 and a full node advantage, but trades blows with it in MT. Nova Lake will have, at best, node parity and 100% more cores, but a greater % of those cores are Atoms vs the Zen 6 P cores. Nova Lake will likely still have a slight advantage, but its not going to be enough to affect 2nm Threadripper whatsoever. Even a 32 core Threadripper would trounce it.

People are forgetting that TR has a significantly higher TDP to work with than desktop SKUs, which gives a significant advantage in MT workloads. Currently a 32 core 4nm Zen 5 is 41% faster than a 3nm 24 core Arrow Lake in arguably Arrow Lakes best case MT scenario, CB R24.

Consider the above and give process parity 3nm to the 9970X. Theres no way Nova Lake is going to take anything away from the user that needs a TR.

 

[511]

Why? Because right now, Arrow Lake has 50% more cores than Zen 5 and a full node advantage, but trades blows with it in MT. Nova Lake will have, at best, node parity and 100% more cores, but a greater % of those cores are Atoms vs the Zen 6 P cores. Nova Lake will likely still have a slight advantage, but its not going to be enough to affect 2nm Threadripper whatsoever. Even a 32 core Threadripper would trounce it.

You are forgetting that they have the worst N3 node in terms of perf/watt by TSMC Numbers its not much better than N4P. As for Atom it's what keeping ARL Competitive in nT workloaad without it ARL would have been embarrassment. I fully expect NVL to win vs Z6 in MT also TR has 100W higher TDP than ARL and 180W higher than 9950X.

 

[reb0rn]

@Josh128
You math does not compute core vs thread is way to go for optimized appz we on desktop mostly use, also do not forget efficiency is way better on lower clock so more cores is always better for MT
Saying 52 core will be same as 24 you need to make intel core be so bad and broken, or you use fake ideology in math as you have done

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

Its just what gain AMD managed in zen6 I defo will not real AMD topic as its hyped and popular and ppl believe all kinds of lies

 

[Josh128]

You are forgetting that they have the worst N3 node in terms of perf/watt by TSMC Numbers its not much better than N4P. As for Atom it's what keeping ARL Competitive in nT workloaad without it ARL would have been embarrassment. I fully expect NVL to win vs Z6 in MT also TR has 100W higher TDP than ARL and 180W higher than 9950X.

I already said I expect NVL to win nT vs Zen 6. I literally said:

Nova Lake will likely still have a slight advantage, but its not going to be enough to affect 2nm Threadripper whatsoever.

I fully expect NVL to win vs Z6 in MT also TR has 100W higher TDP than ARL and 180W higher than 9950X.

Exactly my point. How is NVL supposed to do anything at all to TR given TRs TDP advantage?

 

[511]

Exactly my point. How is NVL supposed to do anything at all to TR given TRs TDP advantage?

Have the Same TDP as TR with some Boards.😛

 

[Thunder 57]

Have the Same TDP as TR with some Boards.😛

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