Core size estimates of various processors

[IntelUser2000]

Some comparisons between modern AMD and Intel processors core size which include the L1. The numbers are approximate, except for Llano.

Nehalem 45nm: 24.4mm2
Shanghai 45nm: 15.3mm2
Atom 45nm: 9.7mm2
Bobcat 40nm: 4.6mm2
Westmere 32nm: 17.2mm2
Sandy Bridge 32nm: 18.4mm2
Llano 32nm: 9.69mm2
Bulldozer 32nm: 14.6mm2 + 3.6mm2 FP per module

Most of the numbers are estimates from Hans De Vries. Sandy Bridge might be a bit off as we're still not fully sure on the actual die size.

 

[Idontcare]

Bobcat is a real the outlier there.

Also, bulldozer being 9.1mm^2/core versus Llano at 9.7mm^2, interesting.

 

[cbn]

http://www.arm.com/products/processors/cortex-a/cortex-a9.php

I found this information under "performance" tab. (Both are TSMC 40nm)

Dual core @ 2GHz :6.7 mm2 (including L1 parity and all DFT/DFM)

Dual core @ 800 Mhz: 4.6 mm2 (including all DFT/DFM)

Unfortunately I could not find information for the Cortex A15.

P.S. What does "DFT/DFM" stand for?

 

[Idontcare]

P.S. What does "DFT/DFM" stand for?

Means the core includes the bloat added with the circuits needed for Design For Test and the core layout itself is optimized for yields (design for manufacturing)...both of which tend to decrease the core density and make the core consume more silicon.

 

[aigomorla]

i think AMD holds the biggest die currently in production with magony..

unless someone want to correct me?

 

[cbn]

Means the core includes the bloat added with the circuits needed for Design For Test and the core layout itself is optimized for yields (design for manufacturing)...both of which tend to decrease the core density and make the core consume more silicon.

Thank you for the explanation.

 

[busydude]

Also, bulldozer being 9.1mm^2/core versus Llano at 9.7mm^2, interesting.

According to the roadmap slides, Llano seems like a stop gap product.. it may be the last one based on stars core.

 

[KingstonU]

How useful is a stop-gap product that is so delayed that it's replacement(s?) arrives before it does?

 

[busydude]

How useful is a stop-gap product that is so delayed that it's replacement(s?) arrives before it does?

I guess its more of an experiment for both GF and AMD.. just as ATI 4770 was a test for 40nm fab process for AMD and TSMC.

 

[Khato]

i think AMD holds the biggest die currently in production with magony..

unless someone want to correct me?

The largest x86 die purrhaps, but certainly one of the largest current processors in production would have to be the Itanium 9300 series! Sure it cheats a bit to get that way by use of the 65nm process, but coming in at just shy of 700mm^2 makes it quite the single-die monstrosity! Sure something around 3/4 of it is L3 cache, but it's still simply massive.

 

[IntelUser2000]

Actually, less than 50% is cache: http://www.tcmagazine.com/tcm/news/hardware/11059/intel-tukwila-itanium-cpus-finally-shipping'

The largest die x86 CPU is Nehalem-EX with 684mm2. The Magny Cours is two 346mm2 dies so its not exactly comparable and should be better in terms of manufacturing.

Llano does have a reason for coming. There are no high end Fusion CPUs from AMD before Llano.

 

[ydnas7]

was Bobcat developed using the same team members as the power core in Xbox360/PS3? I have a feeling there is a passing similarity between the 2. (not withstanding differing instruction sets)

 

[Idontcare]

was Bobcat developed using the same team members as the power core in Xbox360/PS3? I have a feeling there is a passing similarity between the 2. (not withstanding differing instruction sets)

Are you asking if AMD's bobcat design was handled by the IBM team that designed CELL?

If that is the question (I'm not sure) then the answer would be "no" to the best of my knowledge.

 

[Idontcare]

How useful is a stop-gap product that is so delayed that it's replacement(s?) arrives before it does?

If you are struggling to get to a break-even point accounting-wise then I suppose you hope for the best but plan for the worse.

Today's stop-gap can also be tomorrow's fall-back should the next-gen stuff face delay. Even of today's stop-gap is delayed, the fall-back option is still in play. Risk management.

 

[KingstonU]

If you are struggling to get to a break-even point accounting-wise then I suppose you hope for the best but plan for the worse.

Today's stop-gap can also be tomorrow's fall-back should the next-gen stuff face delay. Even of today's stop-gap is delayed, the fall-back option is still in play. Risk management.

I suppose I was just venting to be hearing bad news regarding the new product, as a consumer I hope it does come and performs as expected or better.

 

[ilkhan]

Llano does have a reason for coming. There are no high end Fusion CPUs from AMD before Llano.

Isn't BD a fusion CPU?
I guess I'm getting confused as to whats where in the line-up.
Scorpius is BD and is the top end.
"Dorado" and Lynx (llano) are the mainstream lineup. Dorado is what? Llano minus the GPU is just phenom II. Right?
Source

 

[IntelUser2000]

You can look at the roadmap busydude posted. Bulldozer based APUs will arrive but that's the generation after Llano.

And yes, Dorado is codename for the platform based on Phenom II.

 

[ilkhan]

ok, got it. So initial BD will be competing with s1366 and then s2011. No GPU.

 

[Triskain]

BD will compete with Socket 1155 SB, Socket 2011 isn't even on the table for BD considering the realistically expectable performance of a BD module.

 

[Rezist]

BD will compete with Socket 1155 SB, Socket 2011 isn't even on the table for BD considering the realistically expectable performance of a BD module.

We don't know how BD will perform I think it will fall short of the quad core 1155 SB processors.

 

[Arg Clin]

AFAIK performance of the BD have hardly even been hinted just yet, so I see no reason to claim that BD will not compete til S2011. (or the 1355 for that matter)

It seems like guesswork to me, unless there's some yet undisclosed insider info to back it up.

For all we know BD could blow S2011 out of the water. (I don't think it will - I would expect that AMD aims at it being roughly equal, but that is like everything else - gueswork and speculation)

 

[ydnas7]

Are you asking if AMD's bobcat design was handled by the IBM team that designed CELL?

If that is the question (I'm not sure) then the answer would be "no" to the best of my knowledge.

I'm asking if there were key members on the Bobcat team as on the PowerPC team. Ie Chuck Moore (not the FORTH creator either)


http://zmoore.net/CRM%20Resume 070904.pdf

I wasn't thinking of Cell DSPs, but the powerpc cpu common to both ps3 and Xbo 360

 

[Idontcare]

^ ah, understood now, sorry I have no answer on that. The way people move around in this industry I'd be very surprised if there weren't some number of ex-powerpc folks on the bobcat team.