Raptor Lake Leaks + Intel 4 developments

[Bencher]

Interesting, that there are no applications that can take advantage of more than 8 full cores as that would exceed the power budget of the current technology. May as well stick with 8 cores then

I think you - once again - misunderstood what I'm saying. At this point I begin to think you are doing it on purpose...

I said that generally speaking, applications that take advantage of more than 8 cores, take advantage of n cores. In which case, adding E cores is the better option for performance increase

 

[LePhuronn]

I think you - once again - misunderstood what I'm saying. At this point I begin to think you are doing it on purpose...

No, everybody's understanding what you're saying. The trouble is what you're saying is utter, utter tripe.

 

[humbug]

Eh? So what is stopping them from making "real" 16/24/32/48/64 core parts?

Also if E cores are "just as good", then why aren't there any E-core only parts? You can get 4 E cores in the same space as a P Core, so surely we could have had a 0P 40E core part by now?
Instead of 38 or 40 Core Xeon parts, we could have had 152 or 160 E Core Xeon's which the server market would be lapping up?

Beat me to it ^^^^ the P cores are 4X the size and 4X the power.

 

[Journey]

Can you name an application?

Perhaps look at Sapphire Rapids, and ask why they are using full sized cores for that. If they could use the same amount of space on the dies and get 112 cores in not 56 etc. then why use them at all?

 

[Bencher]

Eh? So what is stopping them from making "real" 16/24/32/48/64 core parts?

Also if E cores are "just as good", then why aren't there any E-core only parts? You can get 4 E cores in the same space as a P Core, so surely we could have had a 0P 40E core part by now?
Instead of 38 Core Xeon W-33xx parts, we could have had 152 E Core Xeon's which the server market would be lapping up?

First of all, I don't understand what you mean "real". Secondly, doesn't your argument absolutely disproves your point? How can we have 38 core xeons if power consumption is stopping them from going to 16??? LOL

I never said E cores are just as good. I said that for applications that scale beyond 8 cores up to an infinite amount of cores, a CPU with e cores IS actually faster. Take CBR, as an easy example. With the amount of die space of a 12900k you could have 10p cores instead of the current 8 + 8 configuration. The latter is absolutely faster than 10p cores would be

 

[Bencher]

Perhaps look at Sapphire Rapids, and ask why they are using full sized cores for that. If they could use the same amount of space on the dies and get 112 cores in not 56 etc. then why use them at all?

I don't know what workloads they are supposed to perform but im assuming it has nothing to do with workloads that the end consumer is, so that's kinda irrelevant, no?

 

[Dg834man]

16p core part would rape a 8p+8e core part, but would be pulling like 400w, bencher absolute fail comment

 

[Bencher]

Beat me to it ^^^^ the P cores are 4X the size and 4X the power.

Same clockspeeds? I really doubt it. In fact I think P cores are more efficient at same clockspeeds than E cores.

 

[Troezar]

I think you - once again - misunderstood what I'm saying. At this point I begin to think you are doing it on purpose...

I said that generally speaking, applications that take advantage of more than 8 cores, take advantage of n cores. In which case, adding E cores is the better option for performance increase

No I'm just trying to understand how it works. You're obviously more well versed in all this than I am. So once you get to using 8P cores you get diminishing returns and the E cores are more efficient? Could they not just make 12P cores that clocked a little lower and used less energy? As I understand it, as you push the clocks the work done for any given amount of power goes down, it doesn't scale linearly?

 

[Bencher]

16p core part would rape a 8p+8e core part, but would be pulling like 400w, bencher absolute fail comment

So you are saying that a 16 P core part would ignore Intel's build in 240w TDP? Uhm...okay I guess

 

[humbug]

First of all, I don't understand what you mean "real". Secondly, doesn't your argument absolutely disproves your point? How can we have 38 core xeons if power consumption is stopping them from going to 16??? LOL

I never said E cores are just as good. I said that for applications that scale beyond 8 cores up to an infinite amount of cores, a CPU with e cores IS actually faster. Take CBR, as an easy example. With the amount of die space of a 12900k you could have 10p cores instead of the current 8 + 8 configuration. The latter is absolutely faster than 10p cores would be

Firstly, no they wouldn't, 1 P core is not faster than 4 E cores.
Secondly, the power consumption would be higher.

 

[Bencher]

No I'm just trying to understand how it works. You're obviously more well versed in all this than I am. So once you get to using 8P cores you get diminishing returns and the E cores are more efficient? Could they not just make 12P cores that clocked a little lower and used less energy? As I understand it, as you push the clocks the work done for any given amount of power goes down, it doesn't scale linearly?

Yes, the more you increase clockspeeds the less efficient you are. That's why the 12900k is drawing a buckload of power, cause it's running at 4.9 ghz.

They could make a 12P core part, but with the same die size they could make an 8P + 16E core part. Both these CPUS would have the exact same Single threaded performance, but the 2nd one would have vastly better multithreading performance. So...why would you - if you were Intel - build the first one?

 

[Bencher]

Firstly, no they wouldn't, 1 P core is not faster than 4 E cores.
Secondly, the power consumption would be higher.

Im not sure where you disagree, i never said 1p core is faster than 4 ecores, im saying the exact opposite.


Also, why would the power consumption be higher? I don't get the argument you people keep bringing up, are you saying that CPU's would suddenly decide to ignore the build in TDP? Wth are you saying?

 

[Troezar]

Yes, the more you increase clockspeeds the less efficient you are. That's why the 12900k is drawing a buckload of power, cause it's running at 4.9 ghz.

They could make a 12P core part, but with the same die size they could make an 8P + 16E core part. Both these CPUS would have the exact same Single threaded performance, but the 2nd one would have vastly better multithreading performance. So...why would you - if you were Intel - build the first one?

Ok I see so you're saying Intel need to make more efficient cores, do more work for a given clockspeed/power consumption? In the meantime they can use these E cores to work around the deficiency?

 

[humbug]

Yes, the more you increase clockspeeds the less efficient you are. That's why the 12900k is drawing a buckload of power, cause it's running at 4.9 ghz.

They could make a 12P core part, but with the same die size they could make an 8P + 16E core part. Both these CPUS would have the exact same Single threaded performance, but the 2nd one would have vastly better multithreading performance. So...why would you - if you were Intel - build the first one?

This is what people are saying, it scales better using the die space for E cores rather than P cores.

Its why they do it.

 

[Bencher]

Ok I see so you're saying Intel need to make more efficient cores, do more work for a given clockspeed/power consumption? In the meantime they can use these E cores to work around the deficiency

No, e cores are LESS efficient than P cores. They are not used for efficiency.

A cpu can either be 16P cores or 8P cores + 32E cores, right? Okay, both of these CPUs have the same single threaded performance, but the 2nd one has way better multithreaded performance. So why would choose the 1st one over the 2nd one?

 

[Bencher]

This is what people are saying, it scales better using the die space for E cores rather than P cores.

Its why they do it.

Uhm, this is what im saying, I don't know why you are arguing. That's my point all along, that it is about die size, not power consumption. The power is irrelevant.

 

[humbug]

Good Grief....

 

[Armageus]

First of all, I don't understand what you mean "real". Secondly, doesn't your argument absolutely disproves your point? How can we have 38 core xeons if power consumption is stopping them from going to 16??? LOL

You know exactly what I mean by real - "P" Cores. It's nice that you still didn't answer the question as to what is stopping them from making "real" 16/24/32/48/64 core parts?

We have 38 Core Xeon's as they are 300W parts, that cannot be sensibly cooled in desktop PCs.

I never said E cores are just as good. I said that for applications that scale beyond 8 cores up to an infinite amount of cores, a CPU with e cores IS actually faster. Take CBR, as an easy example. With the amount of die space of a 12900k you could have 10p cores instead of the current 8 + 8 configuration. The latter is absolutely faster than 10p cores would be

So why don't "E Cores" exist in the server product lines then (where software actually scales beyond 8 cores)?

No, e cores are LESS efficient than P cores. They are not used for efficiency.

Eh? "E"fficiency cores are less efficient than "P" Cores? The clue is literally in the name?

 

[KompuKare]

Ok I see so you're saying Intel need to make more efficient cores, do more work for a given clockspeed/power consumption? In the meantime they can use these E cores to work around the deficiency?

But really, that the P cores currently (in Alder Lake haven't looked at Raptor Lake) still have the AVX512 are but aren't using, tells us how badly put together Alder Lake is. Maybe they are using it as dark silicon but it must be one of the strangest design choices in recent decades, and is frankly embarrassing for such a large corporation.