AMD Athlon II X4 620 & 630: The First $99 Quad Core CPU

[Idontcare]

So I was just getting into Anand's latest article, AMD Athlon II X4 620 & 630: The First $99 Quad Core CPU, and the first thing that struck me when I saw the die shot is that the two cores on the bottom (in Anand's dieshot orientation) are smaller that the two on the top, but about 3% linearly from my crude pixelated guestimates.

http://i272.photobucket.com/al...I_X4_Die_annotated.jpg

You can really see the delineation well along the horizontal axis of the two cache blocks for the left-most cores.

Assuming this isn't some weird trickery of the camera angle or some such, doesn't this seem kinda odd to have the cores physically scaled to occupy differing amounts of silicon real-estate? That has got to make verification all the more challenging, as well as interesting to handle the inter-core clock-skew and so on.

If it was posted by anyone other than Anand we'd all cry photoshop, but I doubt Anand's sources here are anyone other than AMD so I'm pretty sure we can rule out this being a faked dieshot photo. Are there other examples in recent times of asymmetric core size scaling in MPU's?

I can fathom some of the benefits to doing so (maximize use of available real-estate budget), but all of those benefits come with serious trade-offs that I would have assumed until now had a foregone conclusion that the downside of the trade-offs outweighed the upside of the benefits. Apparently that is not the case for the Athlon X4 as we are looking at the proof with our own eyes.

 

[Martimus]

IDC,

I can only see it on one core, (the bottom left). TBH, I am not sure it really is smaller though, and i think it is mostly the light playing tricks on your eyes. If you look at the core in the upper right, you can see clearly where the core circuitry ends, while the bottom left core seems to have the same dimmensions. The upper left core appears to be extended further at first glance, until you notice that the extensions are not part of the core itself.

The Cache does appear to be offset between the two cores, and you point that out in your picture. The lower core's cache is definitely smaller in this picture; perhaps to fit in the HTT along that side. They likely have fewer transistors in that cache, and so are more likely to lose that core to defects (Maybe the reason that Tri-Cores are so prevalent, and why converting them to quads generally works well - since the only real issue with most of the defective cores is that the L2 cache was too small).

I am glad you brought this up, as I can now see why activating the disabled core works so well for X3's. It appears to be a cache issue, that came on due to a space constraint.

-Martimus

 

[heyheybooboo]

It's an AMD optical delusion

I'll guess it's a shot that has been slightly 'shopped for 'illustrative purposes'.

Other Propus die shot here.

What is interesting is to compare back to the original 65nm Phenom. Lop off the 2Mb L3 and it is essentially the same.

 

[Idontcare]

Originally posted by: heyheybooboo
It's an AMD optical delusion

I'll guess it's a shot that has been slightly 'shopped for 'illustrative purposes'.

Other Propus die shot here.

What is interesting is to compare back to the original 65nm Phenom. Lop off the 2Mb L3 and it is essentially the same.

Ah, very good then, yes you can clearly see the cache and core size in that propus dieshot are indeed the exact same dimenions.

Just a funky dieshot in the AT article I guess then. Sanity check complete, thanks :beer:

 

[Ben90]

Why do these things look so damn trippy, i dont do shrooms... but if i did i would call AMD up for a 20'x20' poster of a die shot to stare at for a few hours

 

[Idontcare]

Originally posted by: Ben90
Why do these things look so damn trippy, i dont do shrooms... but if i did i would call AMD up for a 20'x20' poster of a die shot to stare at for a few hours

The underlying optical physics at play in those photographs of the dieshots are the same that give rise to the phenomenon of Newton rings and is based on the interference patterns generated by light reflecting off of semi-transparent stack of various film layers present in the IC.

more than you ever wanted to know about microscopy