Core Size Info from CPU Cores from Intel/AMD/Apple/MTK/QCOM

[siliconbruh999]

A sheet with core sizes of CPU Cores

Core Size

docs.google.com

 

[bearmoo]

Outstanding work! But just to nit-pick a little, it's really hard to compare apple to apple with these numbers if some include the L2 while others don't.

 

[siliconbruh999]

Outstanding work! But just to nit-pick a little, it's really hard to compare apple to apple with these numbers if some include the L2 while others don't.

well this is a real problem and is part of how the cores are designed in the first place like how do you want your memory hierarchy you want private L2 with a bigger L3 or you want Shared L2.
I think you can simply take the Core+shared size and divide by the number of cores to get effective area as well for a single core from cluster

 

[bearmoo]

The simplest comparison would to exclude all L2s but I understand there might not be such measurement available for all the cores. I am no cpu expert but I believe the shared L2 vs L2+L3 has something to do with what the cpus intended end market. For mobile client, the shared arrangement is probably more efficient where as you probably can't use this for the server market. So we will never have a truely apple to apple comparison.

 

[KevinK]

A sheet with core sizes of CPU Cores

Great list - would be interesting to look at number of transistors and rough power / performance curve of each core.

 

[siliconbruh999]

Great list - would be interesting to look at number of transistors and rough power / performance curve of each core.

first thing requires reverse engineering and second thing i have seen a source for it but it's only for fewer comparison

 

[Xebec]

This is a pretty awesome list -- siliconbruh -- do you have offline data for older CPUs?

It would be fun to graph a 'downward trend' on CPU core sizes once they started shrinking.

The ramp up looks something like this, very roughly (need to subtract FSB part of CPUs):

1993 - P5 Pentium, 294mm2 (seems like peak single core size for Intel)

1996 - P55C Pentium MMX, 128mm2-141mm2

1997 - Pentium II Klamath - >200mm2
1998 - Pentium II Deschutes - 113mm2

2000 - Pentium 4 Willliamette - 217mm2
2004 - Pentium 4 Prescott - 112mm2

2006 - Core 2 Duo (Conroe) - 143mm2 (for both cores + FSB), (Geek Note: <30% per core size vs. Williamette, but ~>2.5X the IPC, 1.6X+ higher clocks, at < 1/2 power.. just 6 years later)

(After this real work is required to subtract memory controllers, PCIe, etc.)

I can do further research and provide estimates/references separately if you want to expand this analysis.

 

[siliconbruh999]

This is a pretty awesome list -- siliconbruh -- do you have offline data for older CPUs?

It would be fun to graph a 'downward trend' on CPU core sizes once they started shrinking.

The ramp up looks something like this, very roughly (need to subtract FSB part of CPUs):

1993 - P5 Pentium, 294mm2 (seems like peak single core size for Intel)

1996 - P55C Pentium MMX, 128mm2-141mm2

1997 - Pentium II Klamath - >200mm2
1998 - Pentium II Deschutes - 113mm2

2000 - Pentium 4 Willliamette - 217mm2
2004 - Pentium 4 Prescott - 112mm2

2006 - Core 2 Duo (Conroe) - 143mm2 (for both cores + FSB), (Geek Note: <30% per core size vs. Williamette, but ~>2.5X the IPC, 1.6X+ higher clocks, at < 1/2 power.. just 6 years later)

(After this real work is required to subtract memory controllers, PCIe, etc.)

I can do further research and provide estimates/references separately if you want to expand this analysis.

You need to ask this guy on Twitter lol this is his sheet

https://x.com/G_melo_ding

He said he was open to feedback