GK210 vs GM200

[Lepton87]

I know that there aren't really any official die size published but can someone calculate how much space all of that additional caches take and how big it would be? Can it be approaching GM200 in size?

 

[96Firebird]

According to TPU, it is 561mm². So same size as GK110.

 

[Lepton87]

According to TPU, it is 561mm². So same size as GK110.

Did they actually had the card and measured the die with calipers? If so, it seems every GK110 chip has quite a lot of cache disabled.

 

[ocre]

What do you mean, cache disabled?

How do you conclude that?

 

[JDG1980]

Did they actually had the card and measured the die with calipers? If so, it seems every GK110 chip has quite a lot of cache disabled.

No, the reason that GK110 is the same size despite having less cache than Maxwell is that GK110 has far more double-precision computing ability. This chip was designed for Tesla cards first, and only brought into the GPU marketplace later on.

 

[notty22]

GM200 is the biggest.

it features a jaw-dropping 8 billion transistors in a square 601 mm² die, the biggest ever on the 28 nm node.

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_Titan_X/

 

[Lepton87]

No, the reason that GK110 is the same size despite having less cache than Maxwell is that GK110 has far more double-precision computing ability. This chip was designed for Tesla cards first, and only brought into the GPU marketplace later on.

GK210 and GK110 are Kepler. GM200 is Maxwell You got it all wrong but I admit that GK210 was a very silent launch.

What do you mean, cache disabled?

How do you conclude that?

GK210 has 3.5MB(if my quick mental calculation is correct) more cache than GK110 how can they be the same size?

Whereas a GK110(B) SMX has a 256KB register file and 64KB of shared memory, GK210 doubles that to a 512KB register file and 128KB of shared memory.

Can someone calculate how much die space is needed on a 28nm for that additional cache? I don't believe someone has that tesla and can measure the die. NV is silent about it. GK110 and GM200 are already close in size that cache can easily make the GK210 bigger.

 

[NTMBK]

I doubt anyone has measured GK210- it's only available on a super expensive Tesla card, none of the usual suspects will have spent £4000 just to rip the cooler off and break out the calipers.

I would definitely expect it to be larger than GK110.

 

[Lepton87]

I doubt anyone has measured GK210- it's only available on a super expensive Tesla card, none of the usual suspects will have spent £4000 just to rip the cooler off and break out the calipers.

I would definitely expect it to be larger than GK110.

It has to be, otherwise GK110 would have a lot of unused cache. My guess is that it is even larger then GM200.

 

[ocre]

okay, i didnt realize we were talking about the gk210.

So the gk210 should be bigger than the gk110 unless the gk110 has had cache disabled all along.

I get it

 

[Lepton87]

GM200 is the biggest.

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_Titan_X/

it features a jaw-dropping 8 billion transistors in a square 601 mm² die, the biggest ever on the 28 nm node.

I think this statement maybe by factually incorrect I'm looking for information how much more transistors doubling the size of SMX requires to make an educated guess guess.
There's also that Shared Memory/L1 Cache which again double the size from 64KB to 128KB.

GK110 is 561mm2 large and has 7.1B transistors. GM200 is what? 601mm2 and 8 million.

There are 15 SMXes in the GK210/GK110.

So 15x256KB + 64L1 Cache

How large can that be? I think it might just edge out GM200 ever so slightly in size by a few mm2 a be like 603mm2 large. Why NV hasn't published this information?
Because the biggest and largest GPU in the Titan X marketing material and such description in its reviews sounds good to its buyers.
Tesla buyers? They don't care in the slightest.

I might be wrong of course and GK210 may be under 600mm2.