Discussion Intel Nova Lake in H2-2026: Discussion Threads

[coercitiv]

I am quite interested in how the market will see a 12c part compared to a 52 core part. The optics definitely are in Intel's favor on this one.

It will be the same story as today with 8C part vs 20C part. The superior core count comes into play only after everything else is perceived as equal. Free cores get the sale when they're the cherry on top.

 

[MoogleW]

It will be the same story as today with 8C part vs 20C part. The superior core count comes into play only after everything else is perceived as equal. Free cores get the sale when they're the cherry on top.

Assuming the two are comparable in performance, right now it's rather 16 core (32 threads) Ryzen vs 24 core (24 threads) Intel if we are actually being serious in the comparison

 

[DavidC1]

It will be the same story as today with 8C part vs 20C part. The superior core count comes into play only after everything else is perceived as equal. Free cores get the sale when they're the cherry on top.

It's easy to make people buy the 52C part. You lock the 12C part so you can't clock it higher, have less cache and clockspeed, so the 52C is faster in all areas. Once the market is established, the way to get higher profit is to steer the consumer to the direction you want.

 

[fastandfurious6]

nova lake confirmed they're gonna slap 2 cpu together with glue and gazillion Ecores? 🤣

 

[coercitiv]

Assuming the two are comparable in performance, right now it's rather 16 core (32 threads) Ryzen vs 24 core (24 threads) Intel if we are actually being serious in the comparison

We are very serious in the comparison, though the original post was more about the optics of upselling consumers (not direct comparisons). My comparison however included the 9700X and the 265K, which are both priced around $300, with the Intel CPU being slightly cheaper. Intel is in the awkward position where they have to sell a bigger compute die of ~115mm2 for less money than AMD's ~70mm2 die, and that is assuming the rest of the package costs the same... which it most certainly does not. By way of comparison with the competition, Intel is offering 12 E cores for free.

It's easy to make people buy the 52C part. You lock the 12C part so you can't clock it higher, have less cache and clockspeed, so the 52C is faster in all areas. Once the market is established, the way to get higher profit is to steer the consumer to the direction you want.

We're talking about two things at once:

• the optics of what consumers are presented with when up-selling from 1x compute die to 2x compute in premium products from the same manufacturer
• how these optics change when a competitor offers a different mix of performance & cost for the 1x compute die segment (and presumably loses the 2x die race)

Intel will cover us in cores, but how does that attract sales from a consumer base that is already mostly content with 8P cores and will receive a bump of 4 extra cores and 50% more L3? Their segmentation strategy works well within their lineup, not necessarily against the competition.

 

[DavidC1]

We're talking about two things at once:

• the optics of what consumers are presented with when up-selling from 1x compute die to 2x compute in premium products from the same manufacturer
• how these optics change when a competitor offers a different mix of performance & cost for the 1x compute die segment (and presumably loses the 2x die race)

Intel will cover us in cores, but how does that attract sales from a consumer base that is already mostly content with 8P cores and will receive a bump of 4 extra cores and 50% more L3? Their segmentation strategy works well within their lineup, not necessarily against the competition.

When the TAM is reached, every market has to wrestle with the same question anyway. It's not sustainable in the long term sure, but they can maintain the slow as molasses decline in sales for quite a while longer(or until the debt problems collapse). There's no long term solutions to make people buy more either.

Shouldn't worry about the companies that we aren't directly involved with and be glad that such high performance computers are available to everyone that wants it. We are not responsible for tech market's success in any way shape or form, unless you are working for AMD/Intel/Nvidia/etc. Which is the true spirit of Moore's Law, which disproportionately benefits lower power, lower cost, and smaller computers over the opposite. You could call it true democratization of computing is now.

Market is a constant balance between benefitting the consumers versus the corporation. Within the company it's a fight between product quality/company profitability/employee treatment/management pay. Nvidia being a $4T company and milking every bit of the market they are in are directly connected. So when people complain that their GPUs are ripoffs, well... that's because it benefits them financially. If they were lot less profitable then we'd have cheaper products, simple as that. If you want better products per $ then you should wish for revenue/margins of AMD/Intel/Nvidia to drop.

 

[Fjodor2001]

Lol, no it won't. 12 c will be mainstream next gen because the mainstream is a single CCD with full cores or one cut down. Core count is going up, does that every now and again, but while cores are getting stronger it's not that important for performance.

Yet again the single CCD with Vcache will be the DIY choice.

Apart from a feeling, a personal one (albeit shared by a vocal delusional minority of people on tech forums) do you have any reasoning or data that shows that customers are 1) interested in 16c today or 2) that 16c could become mainstream next gen based on logic (looks to me that this involves two cut down 12c dies to 16c (from 24c) wasting 33% of cores. Can't see that catching on personally.

I guess it also depends on whether we're talking about AMD or Intel SKUs.

For Intel NVL-S this is the expected/rumored lineup:

So all SKUs except the bottom one will have 16C or more (including LP cores), and all except the bottom three will have more than 16C (excluding LP cores). So for Intel 16+C will definitely be mainstream with NVL-S.

 

[fastandfurious6]

wondering about intel performance with 144mb cache... maybe gives crazy boost even higher than amd? who knows

 

[OneEng2]

It will be the same story as today with 8C part vs 20C part. The superior core count comes into play only after everything else is perceived as equal. Free cores get the sale when they're the cherry on top.

Not sure. I can see a big bright "52 CORES!" sign on every premium Intel computer having a top tier NVL. It's marketing. 52 is twice as big as 24.

For people like us, the pitch has to be quite different, but we are not the majority of the market.

Assuming the two are comparable in performance, right now it's rather 16 core (32 threads) Ryzen vs 24 core (24 threads) Intel if we are actually being serious in the comparison

One could argue it is 32t vs 24t vs 16c vs 24c. This is confusing enough and the numbers are close enough that consumers aren't swayed much by the argument. IMO.

It's easy to make people buy the 52C part. You lock the 12C part so you can't clock it higher, have less cache and clockspeed, so the 52C is faster in all areas. Once the market is established, the way to get higher profit is to steer the consumer to the direction you want.

It's debatable that consumers in general ever look at a benchmark to determine what is "faster".

We are very serious in the comparison, though the original post was more about the optics of upselling consumers (not direct comparisons). My comparison however included the 9700X and the 265K, which are both priced around $300, with the Intel CPU being slightly cheaper. Intel is in the awkward position where they have to sell a bigger compute die of ~115mm2 for less money than AMD's ~70mm2 die, and that is assuming the rest of the package costs the same... which it most certainly does not. By way of comparison with the competition, Intel is offering 12 E cores for free.


We're talking about two things at once:

• the optics of what consumers are presented with when up-selling from 1x compute die to 2x compute in premium products from the same manufacturer
• how these optics change when a competitor offers a different mix of performance & cost for the 1x compute die segment (and presumably loses the 2x die race)

Intel will cover us in cores, but how does that attract sales from a consumer base that is already mostly content with 8P cores and will receive a bump of 4 extra cores and 50% more L3? Their segmentation strategy works well within their lineup, not necessarily against the competition.

Great post.

Yes, I was referring to the optics of what consumers are presented with which is what will determine sales ..... along with the deal AMD vs Intel can bring to the table with the OEM's.

I don't believe most people even know how many cores their computer has in it today. I am far from your average consumer, and don't know what this work laptop has in it. Isn't that sad?

Where this changes is in the HEDT and gaming. Both of these consumers are educated (extremely so). Actual performance will drive these buying decisions in the top margin parts in the desktop (and server for that matter).

 

[511]

Not sure. I can see a big bright "52 CORES!" sign on every premium Intel computer having a top tier NVL. It's marketing. 52 is twice as big as 24.

I feel like people are making a big Deal out of 52C they are just simply adding one more compute tile nothing complex everything else is shared between all the NVL-SKU

 

[adroc_thurston]

I feel like people are making a big Deal out of 52C they are just simply adding one more compute tile nothing complex everything else is shared between all the NVL-SKU

People see big number and freak out.
No different from 3950X.

 

[Thunder 57]

It's easy to make people buy the 52C part. You lock the 12C part so you can't clock it higher, have less cache and clockspeed, so the 52C is faster in all areas. Once the market is established, the way to get higher profit is to steer the consumer to the direction you want.

Ah artifical segmentation, Intel's specialty. Might as well cut AVX10 as well just to be thorough!

Not that AMD doesn't, but outside of some early Zens lacking SMT I haven't seen it lately. Maybe they'll do it again and cut L3 on 8C Zen 5.

 

[OneEng2]

I feel like people are making a big Deal out of 52C they are just simply adding one more compute tile nothing complex everything else is shared between all the NVL-SKU

I am guessing that the generic "people" might make a big deal out of 52c. Generic "people" don't know that a "tile" is.

Once-upon-a-time, people ubiquitously equated MHz to performance level. Intel understood this and made P4 clock really high.

AMD countered this with model numbers successfully.

Today, my concern is we may be moving toward the core wars where a customer is trained to equate the number of cores to performance.

It may ring hallow though. Model numbers have made it pretty much a sure thing that you can't compare one OEM to another through the main name of the CPU. If that were the case, then the 14900K would beat the crap out of a 9950 X3D.

So perhaps, the number of cores WONT be used as a marketing tool. I am just speculating that this is all Intel currently has IMO. "Play to your strengths".

 

[ondma]

The market dictates the price of a product. The product DOESN'T dictate the price the market will pay.

Max core count, top bin processor (regardless of how many cores and how fast it goes) sits at around $800 at launch. The 9950 was like this, but now after ramp up is sitting around $450.

A new generation of CPU's will not make a new market .... and therefore will not change the price consumers are willing to pay for a particular market segment.

This is absolutely true.

Both AMD and Intel will drive mainstream pricing at a single CCD product. For Zen 6, that means a 12c/24t part.

I am quite interested in how the market will see a 12c part compared to a 52 core part. The optics definitely are in Intel's favor on this one.

Not necessarily. The 52 core part will be a MT monster, but probably not the best performance for gaming, and overkill for nearly every use case. I mean just look at the market now. Intel has a 24 core part, but the product in demand is the 8 core 9800 x3d because it has the performance.

 

[Thunder 57]

I am guessing that the generic "people" might make a big deal out of 52c. Generic "people" don't know that a "tile" is.

Once-upon-a-time, people ubiquitously equated MHz to performance level. Intel understood this and made P4 clock really high.

AMD countered this with model numbers successfully.

Today, my concern is we may be moving toward the core wars where a customer is trained to equate the number of cores to performance.

It may ring hallow though. Model numbers have made it pretty much a sure thing that you can't compare one OEM to another through the main name of the CPU. If that were the case, then the 14900K would beat the crap out of a 9950 X3D.

So perhaps, the number of cores WONT be used as a marketing tool. I am just speculating that this is all Intel currently has IMO. "Play to your strengths".

That's a popular story but not what happened. Intel didn't just say screw IPC lets clock to the moon (at least not the engineers). They really thought Dennard Scaling would continue and they could keep clocking higher. Along with their process advantage (also helping clocks) they could add more transistors than the competition. Marketing took those GHz numbers and ran with them. As far as today, everybody uses model numbers so I don't see a "core war".

 

[adroc_thurston]

Intel didn't just say screw IPC lets clock to the moon (at least not the engineers). They really thought Dennard Scaling would continue and they could keep clocking higher.

Yup.
Speed demons were the hot stuff in the 90s and it looked like they could last forever.

 

[igor_kavinski]

Speed demons were the hot stuff in the 90s and it looked like they could last forever.

We are back at that stage again with 14900KS and Zen 6 if rumors are true.

 

[511]

We are back at that stage again with 14900KS and Zen 6 if rumors are true.

It took a lot of time from going 4 -> 5 and than 5 -> 6Ghz and I doubt we will be breaching 7 GHz any time soon.

 

[Thunder 57]

We are back at that stage again with 14900KS and Zen 6 if rumors are true.

Nah. Zen 6 isn't touching 7GHz. Base frequencies could be higher and boost might hit 6.5GHz but 7GHz is MLID BS nonsense.

Here's an edit just for fun. Who remembers the French magazine that had Zen "info" in binary? Anyone remember what it claimed?

 

[MoogleW]

wondering about intel performance with 144mb cache... maybe gives crazy boost even higher than amd? who knows

More than AMD? Unlikely. But the entirety of 9800X3D hype is from cache and not Zen5% architecture so gains may have big outliers too. Unless their cache somehow brings regressions

 

[Fjodor2001]

More than AMD? Unlikely.

How do we know which one will perform best, and based on what? The Intel BLLC is new and not yet tested so hard to know exactly how it will perform.

 

[OneEng2]

How do we know which one will perform best, and based on what? The Intel BLLC is new and not yet tested so hard to know exactly how it will perform.

We don't.

The reason I am leaning toward Intel not performing as well as AMD in L3 cache is that the ARL generation has insane latency issues and I am guessing that some of it is architectural in nature and will require a major revision to fix.

This is 100% pure speculation though.

 

[Fjodor2001]

The reason I am leaning toward Intel not performing as well as AMD in L3 cache is that the ARL generation has insane latency issues and I am guessing that some of it is architectural in nature and will require a major revision to fix.

So maybe BLLC will solve that? Also, you mentioned ARL but this is for NVL, so Intel may have included related architectural changes in the latter.

 

[adroc_thurston]

We don't.

The reason I am leaning toward Intel not performing as well as AMD in L3 cache is that the ARL generation has insane latency issues and I am guessing that some of it is architectural in nature and will require a major revision to fix.

This is 100% pure speculation though.

Intel had pretty dog L3 since TGL.

 

[511]

Intel had pretty dog L3 since TGL.

In servers as well