Discussion Intel current and future Lakes & Rapids thread

[dullard]

Well, guess we don't know what the compute tile will be - how many cores, cache size, etc. Intel seems to be going larger on P Cores.

The P cores are getting more transistors, but they are sticking with 8 P cores for many generations. So as the node shrinks, it might end up with smaller and smaller areas. Of course, that requires a lot of hand-waving regarding information we don't know such as how much more will cache sizes grow. As far as I know, the number of E cores on Meteor Lake hasn't leaked yet. If it is just a node shrink of Raptor Lake (similar to Intel's old "Tick"), 8 P and 16 E cores, then Meteor Lake could be a small chip. A small chip helps meet production limitations.

What Intel seems to be planning on is a very short Meteor Lake production. It is sandwiched between Raptor Lake (4Q 2022) and Arrow Lake (2H 2023). If Meteor Lake is late or Arrow Lake is early, then Meteor Lake might not really have much time in their production schedule.

 

[nicalandia]

The P cores are getting more transistors, but they are sticking with 8 P cores for many generations.

Do intel still limited to 10 cores before Switching to Mesh?

 

[jpiniero]

Do intel still limited to 10 cores before Switching to Mesh?

They might go to "12" but not much more than that.

 

[nicalandia]

They might go to "12" but not much more than that.

I think they have found that 8 cores is the sweet spot(at least for them) and adding more would hinder efficiency... But a four chiplet(3 CCD and one IO) Zen4/5 will put a hurt on them at least on SMT

 

[IntelUser2000]

Do intel still limited to 10 cores before Switching to Mesh?

So if Raptor Lake is getting 16 E cores, and we have 24 with Meteor and 32 with Lunar, there's a way of getting there without increasing ring stops much.

Raptor Lake - 2x 8 core E tile, for two ring stops, same as Alder Lake.
Meteor Lake - 3x 8 core tiles
Lunar Lake - 4x 8 core tiles

Then you get into the absolute limit of ring stops.

 

[eek2121]

The P cores are getting more transistors, but they are sticking with 8 P cores for many generations. So as the node shrinks, it might end up with smaller and smaller areas. Of course, that requires a lot of hand-waving regarding information we don't know such as how much more will cache sizes grow. As far as I know, the number of E cores on Meteor Lake hasn't leaked yet. If it is just a node shrink of Raptor Lake (similar to Intel's old "Tick"), 8 P and 16 E cores, then Meteor Lake could be a small chip. A small chip helps meet production limitations.

What Intel seems to be planning on is a very short Meteor Lake production. It is sandwiched between Raptor Lake (4Q 2022) and Arrow Lake (2H 2023). If Meteor Lake is late or Arrow Lake is early, then Meteor Lake might not really have much time in their production schedule.

Intel really doesn’t need 8 P cores. If they could give Gracemont a 15-20% IPC bump, triple down on the number of cores, improve latency, and possibly give frequencies a bump, they could get away with 4 P-cores.

What I could see them doing is using TSMC 4nm/3nm alongside Intel 4 for the compute tile, and TSMC 6nm/Intel 7nm for the GPU tile.

I don’t buy that they will have any serious shortages. Everyone I spoke with thought Alder Lake would sell out, however so far stock looks better than AMD did at launch. Intel could easily only launch enthusiast parts similarly to ADL-S if they have a shortage.

Note that they have stated there will be a client release.

However, we don’t know the die sizes for future products. They could be smaller.

 

[TESKATLIPOKA]

Twitter

PL: 150W is pretty good. PL: 241W is only 8% faster, but consumes 45% more power.
BTW the performance difference between 125W -> 241W PL is 28%.

 

[Exist50]

The 12700k scores 22,689 points in CBr23 @160W. I found that interesting.



Meteor Lake would be Core 14xxx series. I doubt we will see further delays. Intel has already produced and powered on the early silicon with no issues.

Do not assume that process is the only reason for delays.

 

[TESKATLIPOKA]

Post moved to the other thread.

 

[KOMACHI_ENSAKA]

https://twitter.com/x/status/1449292869353705474

Might be Gracemont-only SKUs coming.

I'm sorry. I was a little late. It's just 0+8 is reserved. It doesn't indicate a SKU exists.

 

[mikk]

Interesting:

https://twitter.com/x/status/1456966531460501507

 

[pakotlar]

Interesting:

https://twitter.com/x/status/1456966531460501507

Thats pretty remarkable. If you watch the video CPU package power is 45W however. Still excellent

 

[mikk]

Thats pretty remarkable. If you watch the video CPU package power is 45W however. Still excellent

9W iGPU usage for whatever reason.

 

[pakotlar]

9W iGPU usage for whatever reason.

CPU GT is gpu cores? Cool

 

[TESKATLIPOKA]

Interesting:

https://twitter.com/x/status/1456966531460501507

Actually, M1 Max is a bit better, but ADL was only 6C+8c and speed was 3+2.4GHz.
i9 12900K 35W - 14288
M1 Max 30W - 12326
12326/30*35 -> 14380
In other words, It has 0.6% better efficiency. 😉

 

[Accord99]

Thats pretty remarkable. If you watch the video CPU package power is 45W however. Still excellent

Also impressive is the 25244 score with package power of 117W, which is better than the 5950x.

Given the high performance/watt of Alder Lake at reasonable clock speeds, we may see AMD "recommend" 200+W power settings in reviews for its 16-core CPUs going forward.

 

[jpiniero]

I took a fresh look at the feasibility of Meteor Lake, and it does seem feasible but the CPU chiplet would need to be dual sourced I think despite being tiny. Without the dual source they would most likely have to do something like they did with Comet/Ice where they do Raptor Lake mobile alongside Meteor Lake products.

Either way I think it's likely to be only mobile and any OEM desktops willing to use the mobile processors.

 

[Hitman928]

Also impressive is the 25244 score with package power of 117W, which is better than the 5950x.

Given the high performance/watt of Alder Lake at reasonable clock speeds, we may see AMD "recommend" 200+W power settings in reviews for its 16-core CPUs going forward.

Is he manually undervolting?

 

[Accord99]

Is he manually undervolting?

I think so, but I don't think it's too unfair. The 5950x is already pretty well-tuned by AMD, you can't use a very high negative offset (compared to the lower tier 5000 models) with PBO curve optimizer before a 5950x becomes unstable.

 

[Hitman928]

I think so, but I don't think it's too unfair. The 5950x is already pretty well-tuned by AMD, you can't use a very high negative offset (compared to the lower tier 5000 models) with PBO curve optimizer before a 5950x becomes unstable.

Eh, depends on the sample and how you undervolt. You can get a pretty strong undervolt if you're willing to shave off the top range of frequencies. Also not sure if it is truly stable at these voltages, hard to follow when I don't speak a word of the language. If it is manual undervolt results, it's still interesting, but makes the comparison to the competition far less valid.

 

[Markfw]

Also impressive is the 25244 score with package power of 117W, which is better than the 5950x.

Given the high performance/watt of Alder Lake at reasonable clock speeds, we may see AMD "recommend" 200+W power settings in reviews for its 16-core CPUs going forward.

Its better in single core, but not multi core or 28,641

AMD Ryzen 9 5950X Benchmark, Test and specs

AMD Ryzen 9 5950X benchmark results and review of this cpu with specs including the number of cores, threads, memory bandwidth, pcie lanes and power consumption. Benchmarks in Cinebench R23 and Geekbench 5

www.cpu-monkey.com

And cpu-mokey gave the 12900k 27,198 stock, so by the same reviewer, it looses also.

 

[eek2121]

Its better in single core, but not multi core or 28,641

AMD Ryzen 9 5950X Benchmark, Test and specs

AMD Ryzen 9 5950X benchmark results and review of this cpu with specs including the number of cores, threads, memory bandwidth, pcie lanes and power consumption. Benchmarks in Cinebench R23 and Geekbench 5

www.cpu-monkey.com

I have to wonder how much higher it would go moving from 117W -> 143W

 

[Markfw]

I have to wonder how much higher it would go moving from 117W -> 143W

See above, it still looses by cpu-monkey.

Edit: I also trust a known site vs some unknown chinese site.

 

[Accord99]

28K score requires an overclock on the 5950X (and all-core clock speeds around 4.4 GHz). At stock, its score is in the 25K range

CPU Benchmarks and Hierarchy 2025: CPU Rankings

We've run thousands of CPU benchmarks on all new and older Intel and AMD CPUs and ranked them.

www.anandtech.com

 

[Hitman928]

28K score requires an overclock on the 5950X (and all-core clock speeds around 4.4 GHz). At stock, its score is in the 25K range

CPU Benchmarks and Hierarchy 2025: CPU Rankings

We've run thousands of CPU benchmarks on all new and older Intel and AMD CPUs and ranked them.

www.anandtech.com

That's true, CB and especially r23 seems to be a strong spot for ADL. The 5950x at 4.4GHz should still require a bit less power than the 12900k took to achieve 28k and the 5950x will score closer to 29k. One would be officially overclocked and the other unofficially . I think ADL-P has a good chance at beating Cezanne though once you get down to the 45W range or so and lower.