News Intel's Nova Lake mobile specs leaks out, top end HX variant to feature 28 CPU cores and 4 Xe cores — Entire lineup is reportedly limited to a sing...

[Admin]

Core configs for Intel's Nova Lake mobile lineup has reportedly leaked, with the top HX variant featuring up to 28 CPU cores and 4 GPU cores. Nova Lake-H and Nova Lake-U families also detailed.

Intel's Nova Lake mobile specs leaks out, top end HX variant to feature 28 CPU cores and 4 Xe cores — Entire lineup is reportedly limited to a sing... : Read more

 

[usertests]

What stood out to me. First, Panther Lake with 10 Xe cores. I thought 12 was planned but maybe I wasn't paying attention.

Panther Lake is practically a subset of Nova Lake, with the bottom feeder configuration reappearing and the U/H-alike getting a graphics focused version. Putting the strongest graphics in a separate SKU is easy to do with tiles, and probably a good idea to avoid wasting silicon that will sit unused in laptops with discrete graphics.

Intel is not being generous with P-cores here. Nothing except the HX gets more than 4, and that's basically a BGA desktop chip. I don't think it matters that much for gaming but someone will complain.

2+0+4+2 corresponds exactly to Wildcat Lake. So is that Wildcat Lake as we know it, or three different products that are almost the same? It doesn't make sense. I see the article addresses that:

On the lowest end, a 2P+4LP model is also expected, carrying only two Xe3 cores, positioned as the successor to Intel’s Wildcat Lake platform

It seems too soon for that, but maybe it's coming much later than the others.

I don't think it matters that 52 cores isn't coming to laptops. But a bLLC competitor to AMD's Dragon/Fire Range X3D would be appreciated.

 

[thestryker]

What stood out to me. First, Panther Lake with 10 Xe cores. I thought 12 was planned but maybe I wasn't paying attention.

There is this article just didn't show all of the leaked information:

P/E/LPE/Xe3:
PTL-H:
4/8/4/12
4/8/4/4
4/0/4/4

WCL:
2/0/4/2

in addition to what they did show:
PTL-H:
4/4/4/10
4/4/4/4
2/0/4/2 (this is apparently not WCL)

 

[Alvar "Miles" Udell]

The headline: top end HX variant to feature 28 CPU cores
The reality: top end HX variant to feature 8 P-Cores, 16 E-Cores, and 4 LP-E Cores

So will it be beaten by a 16 core AMD CPU?

 

[usertests]

2/0/4/2 (this is apparently not WCL)

Very strange. I think some of this is probably wrong, or Intel is learning they can rapidly iterate after splitting everything into chiplets (I have no idea if Wildcat Lake is monolithic or chiplet-based).

So will it be beaten by a 16 core AMD CPU?

AMD is going to be doing the same stuff, especially in mobile. Strix Point is already a 4+8, Medusa APUs look even more complicated and are including 2 LP cores.

However, a sequel to Fire Range with 2 chiplets should be 24 classic cores (maybe 24+0+2), easily beating 8+16+4 from Intel.

 

[YSCCC]

It looks to me like a Raptor Lake config (8p 16e) or maybe arrow lake config without hyper threading, but now put another 4 even slower LP E cores inside (Xe core looks to me just iGPU core?) and this don't sounded very competitive to the upcoming AMD...

 

[usertests]

It looks to me like a Raptor Lake config (8p 16e) or maybe arrow lake config without hyper threading, but now put another 4 even slower LP E cores inside (Xe core looks to me just iGPU core?) and this don't sounded very competitive to the upcoming AMD...

Basically it's half the potential of desktop since they can use two of the 8+16 tile but they choose not to according to the leak. But it's not unchanged from Raptor/Arrow since the P/E cores are updated. Coyote Cove and Arctic Wolf or whatever they are calling them. IPC should be up, and likely clocks, with better performance per watt.

HX is a desktop part put into laptops. You don't need more than 8 P-cores and 16 E-cores for gaming and it should be fine for productivity. What would sweeten the deal: Intel brings bLLC (big L3 cache) to laptops to compete with 9955HX3D.

LP E-cores are good for idling or light tasks if they work properly.

4 Xe3 cores isn't even bad. If it's just a little faster than the Arrow Lake-S iGPU, it will be faster than an 8500G and maybe a GTX 1630.

I don't think this 28 core will beat AMD's Zen 6 Fire Range successor, but it will be a good chip.

 

[thestryker]

Very strange. I think some of this is probably wrong, or Intel is learning they can rapidly iterate after splitting everything into chiplets (I have no idea if Wildcat Lake is monolithic or chiplet-based).

Yeah I have no idea what exactly would make up the difference between them. It is possible there will be a CPU Tile without E-cores though given that there is a 4/0/4 PTL configuration. Perhaps it's something different with IO or SOC Tiles, but I'd be surprised if WCL was monolithic unless maybe they're using a different manufacturing process.

 

[TerryLaze]

It looks to me like a Raptor Lake config (8p 16e) or maybe arrow lake config without hyper threading, but now put another 4 even slower LP E cores inside (Xe core looks to me just iGPU core?) and this don't sounded very competitive to the upcoming AMD...

Only because everybody is over-hyping zen 6.
Also if amd releases a 24core CPU it will be either super expensive or it will ruin amd financially.

 

[usertests]

Only because everybody is over-hyping zen 6.
Also if amd releases a 24core CPU it will be either super expensive or it will ruin amd financially.

What exactly do you think is overhyped about Zen 6? AMD have been making tiny 8-core chiplets for years, moving that up to 12-core after a couple of node shrinks (from N4X to N2P/N2X) won't be a problem for them. The 24-core flagship desktop CPU could start out at $800 (9950X was $649 MSRP), but it would be profitable for them even at $500, because ~75mm^2 CCDs cost so little.

The IPC isn't overhyped, it will probably be in the neighborhood of +10% again. The only thing overhyped about Zen 6 is the notion of the turbo clock speed hitting 7.0-7.5 GHz territory, from the current 5.7 GHz.

 

[TerryLaze]

The 24-core flagship desktop CPU could start out at $800 (9950X was $649 MSRP),but it would be profitable for them even at $500, because ~75mm^2 CCDs cost so little.

Depends on what you consider profitable I guess, if they barely make any money then they won't have enough to design and produce the next gen.

The only thing overhyped about Zen 6 is the notion of the turbo clock speed hitting 7.0-7.5 GHz territory, from the current 5.7 GHz.

That's not enough?!
Also people think that the turbo clocks will be the all core clocks...................................................................................................................................................................
A 24 core CPU using the same amount of power as the 16 core one would run 30% slower at all core, maybe 20% with the power efficiency gains, and people think it will be running 20% faster than the 16 core one.
We are talking about a 40% + difference here, that's over hyped into the stratosphere.

 

[usertests]

Depends on what you consider profitable I guess, if they barely make any money then they won't have enough to design and produce the next gen.

My point is that AMD will be making good margins on the 24-core, easily. If the demand is low, prices will drop, but they can still make a profit. The kind of CPU holding its value right now is the 9800X3D, so I expect 10/12-core Zen 6 X3D to do really well.

Zen 6 is the time for AMD to move past the 16-core to counter Intel high core count parts. Boosting core counts by 50% is not going to ruin them financially, and it's going to help them compete effectively with a "52 core".

We are talking about a 40% + difference here, that's over hyped into the stratosphere.

Zen 6 has to respect the same AM5 socket TDP/PPTs, so that could be a limiting factor for a 24-core.

But it's skipping the TSMC N3 nodes, using either N2P or N2X (competing leaks). TSMC says:
* N3E uses -34% power at the same perf, OR +18% perf vs. N5
* N2P uses -36% power at the same perf, OR +18% perf vs N3E
* N2X is another 10% perf above N2P

0.66 * 0.64 = ~0.42 or -58% power at the same perf. The Zen 5 CCD is on N4X, not N5, but it's clear that there is going to be a huge efficiency improvement moving from N4X down to a 2nm-class GAAFET node. 24 cores should be doing fine with the same socket power limits and coolers. And when most of those cores are unused, maybe one of them will boost to 7 GHz which is what our world of inefficient code really craves...

 

[TerryLaze]

Zen 6 has to respect the same AM5 socket TDP/PPTs, so that could be a limiting factor for a 24-core.

But it's skipping the TSMC N3 nodes, using either N2P or N2X (competing leaks). TSMC says:
* N3E uses -34% power at the same perf, OR +18% perf vs. N5
* N2P uses -36% power at the same perf, OR +18% perf vs N3E
* N2X is another 10% perf above N2P

0.66 * 0.64 = ~0.42 or -58% power at the same perf. The Zen 5 CCD is on N4X, not N5, but it's clear that there is going to be a huge efficiency improvement moving from N4X down to a 2nm-class GAAFET node. 24 cores should be doing fine with the same socket power limits and coolers. And when most of those cores are unused, maybe one of them will boost to 7 GHz which is what our world of inefficient code really craves...

Yeah so if they copied zen 5 1:1 then the 50% more cores will be running at the same clocks and maybe a little bit higher....if it has more transistors then clocks will drop accordingly.

Maximum clocks do not depend solely on the margin of available energy you have. If this will be like previous zen releases where they would only get the max clocks during idle time...nah, not even then, you don't reach 7Ghz without exotic cooling.