Discussion Intel current and future Lakes & Rapids thread

[H433x0n]

At 2GHz 288 cores would be within 300-350W, but then you have to look at the perf, that s a low throughput overall.


Have you numbers that suggest that Intel 3 is significantly better than say TSMC s N5.?.

If anything Intel say that it will be about 42% better than Intel 7, that would barely close the gap if we are to look at current number between N5 and Intel 7.

Where do you see N5 provides 42% better performance per watt than Intel 7? There are no apples/apples comparisons. The closest we can get is here but it's unclear whether or not they're comparing the same arm a7 core.

 

[SiliconFly]

Where do you see N5 provides 42% better performance per watt than Intel 7? There are no apples/apples comparisons. The closest we can get is here but it's unclear whether or not they're comparing the same arm a7 core.

True there are no apples/apples comparisons, but a ballpark estimate should be around 15% to 20% max.

 

[Khato]

Where do you see N5 provides 42% better performance per watt than Intel 7? There are no apples/apples comparisons. The closest we can get is here but it's unclear whether or not they're comparing the same arm a7 core.

While it's obviously not N5/Intel 7, the possibility of a direct comparison is what I find most appealing about the prospect of an ARL CPU tile being made on both 20A and N3B.

 

[Abwx]

True there are no apples/apples comparisons, but a ballpark estimate should be around 15% to 20% max.

Read right, i said that Intel claim about 42% better perf watt from 7 to 3 process.

From the comparisons we have between N5 fabbed AMD CPUs and 7 Intel fabbed equivalentCPUs it look like that N5 is more than 42% more efficient than Intel 7.

 

[Geddagod]

From the comparisons we have between N5 fabbed AMD CPUs and 7 Intel fabbed equivalentCPUs it look like that N5 is more than 42% more efficient than Intel 7

Link?

 

[adroc_thurston]

Sierra Glen != Gracemont

It quite literally is Gracemont with server RAS.

 

[dullard]

Read right, i said that Intel claim about 42% better perf watt from 7 to 3 process.

Eh, if you are going to complain about people reading right, then you should type right. I hate grammar posts in CPU threads, but you brought the reading right up.

Here is your post. Look at what "it" refers to. Heck the whole paragraph with the 42% number doesn't even include Intel 3. That, said, I got your gist.

 

[SiliconFly]

Read right, i said that Intel claim about 42% better perf watt from 7 to 3 process.

From the comparisons we have between N5 fabbed AMD CPUs and 7 Intel fabbed equivalentCPUs it look like that N5 is more than 42% more efficient than Intel 7.

We can't compare efficiencies of an AMD CPU fabbed in N5 using HP/HD cell library with an Intel CPU fabbed in Intel 7 UHD library.

Intel 7's UHD cells are very power hungry compared to its own HP/HD cells.

 

[Abwx]

Eh, if you are going to complain about people reading right, then you should type right. I hate grammar posts in CPU threads, but you brought the reading right up.

Here is your post. Look at what "it" refers to. Heck the whole paragraph with the 42% number doesn't even include Intel 3. That, said, I got your gist.

View attachment 85981

The second paragraph is the continuation of the first, i started by mentioning Intel 3, from here the "it" aplly to it since it s the process that i question, not to be pedantic...

 

[dullard]

The second paragraph is the continuation of the first, i started by mentioning Intel 3, from here the "it" aplly to it since it s the process that i question, not to be pedantic...

That just isn't how English works, but okay.

"It" decided that already.

 

[Abwx]

We can't compare efficiencies of an AMD CPU fabbed in N5 using HP/HD cell library with an Intel CPU fabbed in Intel 7 UHD library.

Intel 7's UHD cells are very power hungry compared to its HP/HD cells.

We can more or less compare a intel 8 P core to an AMD 8C, there s numbers at Computerbase, they benched a simulated 8 P core where they disabled all e cores, that s an ADL 12900K though, but we can see on other benches the improvement of Intel 7 with RPL by comparing a 12900K to a 13700K.

 

[H433x0n]

Read right, i said that Intel claim about 42% better perf watt from 7 to 3 process.

From the comparisons we have between N5 fabbed AMD CPUs and 7 Intel fabbed equivalentCPUs it look like that N5 is more than 42% more efficient than Intel 7.

No - it doesn't. You can only get those types of numbers if you measure power at iso perf and cherry pick comparisons on the edge of the VF curve. That’s not how the industry measures performance per watt.

When comparing both 13900K / 7950X at 125W Zen 4 has a 10-12% performance advantage at iso power.

 

[Abwx]

No - it doesn't. You can only get those types of numbers if you measure power at iso perf and cherry pick comparisons on the edge of the VF curve. That’s not how the industry measures performance per watt.

When comparing both 13900K / 7950X at 125W Zen 4 has a 10-12% performance advantage at iso power.

i compare a 7700K to a 8 P cores 12900K with disabled e cores, then i correct the numbers by comparing a 12900K to a 13700K tested at same power to take account of Intel 7 improvement from ADL to RPL.

You understand the logic..?...

 

[Khato]

It quite literally is Gracemont with server RAS.

You might want to think about that claim for a moment. Specifically with respect to the process used for the Gracemont core.

 

[H433x0n]

i compare a 7700K to a 8 P cores 12900K with disabled e cores, then i correct the numbers by comparing a 12900K to a 13700K tested at same power to take account of Intel 7 improvement from ADL to RPL.

You understand the logic..?...

No, I don’t. It’s twisted logic where you’ve got to jump through a bunch of hoops to get your desired outcome.

Just compare the 2 processors as-is at a fixed power target. It’s simple and a realistic scenario.

 

[SiliconFly]

We can more or less compare a intel 8 P core to an AMD 8C, there s numbers at Computerbase, they benched a simulated 8 P core where they disabled all e cores, that s an ADL 12900K though, but we can see on other benches the improvement of Intel 7 with RPL by comparing a 12900K to a 13700K.

We can see improvements, variations, etc. But we can't compare nodes based on their products.

Actually, we can't even compare 2 different processors from the same node if they're using different design rules, pdks, cell libraries, etc. Their V/F curves will vary.

 

[SiliconFly]

No, I don’t. It’s twisted logic where you’ve got to jump through a bunch of hoops to get your desired outcome.

You nailed it.

 

[Joe NYC]

Eh, if you are going to complain about people reading right, then you should type right. I hate grammar posts in CPU threads, but you brought the reading right up.

Here is your post. Look at what "it" refers to. Heck the whole paragraph with the 42% number doesn't even include Intel 3. That, said, I got your gist.

View attachment 85981

I think you got the grammar wrong. "It" would be referring to the subject of the previous sentence.

 

[eek2121]

At 2GHz 288 cores would be within 300-350W, but then you have to look at the perf, that s a low throughput overall.


Have you numbers that suggest that Intel 3 is significantly better than say TSMC s N5.?.

If anything Intel say that it will be about 42% better than Intel 7, that would barely close the gap if we are to look at current number between N5 and Intel 7.

Intel’s node names are named based on the competition. Intel 20A and 18A should edge out TSMC N3*. However it is going to depend on whether optimizing for performance, power, density, or some combination. Same with TSMC.

I have zero doubt about the performance of the nodes themselves. My concerns are currently about the chip design.

 

[Abwx]

No, I don’t. It’s twisted logic where you’ve got to jump through a bunch of hoops to get your desired outcome.

Just compare the 2 processors as-is at a fixed power target. It’s simple and a realistic scenario.

The only things that get their desired outcomes are the laws of phycics, once your understand how mosfets works you can draw conclusions on whatever available numbers provided they are significant.

It s not that difficult for instance to extract a process power/frequency curve shape from a few datas that are available in benches or from datas provided by foundries, as i told you things do not work by chance or mysteriously, there are precise laws at work.

 

[adroc_thurston]

Specifically with respect to the process used for the Gracemont core.

Atoms are node-independent for a looooong while, ironically for SoFIA purposes.

 

[H433x0n]

It quite literally is Gracemont with server RAS.

It’s not. Crestmont has a 6-8% IPC increase per Ian Cuttress (numbers provided by Intel). I’m not sure what that means for Sierra Glen but I doubt it’s worse than that.

 

[Khato]

Atoms are node-independent for a looooong while, ironically for SoFIA purposes.

Heh, it's no fun providing hints when they're deflected in order to avoid admitting errors. Feel free to continue believing that Sierra Glen is based on Gracemont. For everyone else, it's Crestmont based.

 

[adroc_thurston]

Feel free to continue believing that Sierra Glen is based on Gracemont.

Well I don't have to believe, that's the literal answer Intel presenters gave during their SRF session Q&A.

 

[Khato]

Well I don't have to believe, that's the literal answer Intel presenters gave during their SRF session Q&A.

Oh? What "SRF session Q&A" would that be?