TSMC Set to Produce Nvidia Maxwell Graphics Chips Using 20nm Process Technology.

[blastingcap]

Very cool. :thumbsup:

Reading the article, the main reason seems to be because the US government wants their chips made in the USA. Guess that is one way to push for new fabs in the US...

There is something to be said for the US military to be able to source more of its parts domestically rather than from overseas.

 

[Lepton87]

Why so certain?


I... don't remember the specifics; I remember that I looked over the important aspects as compared by the various processes and noticed the intel generational lead.
So, based on that recollection it would probably be power consumption, and/or thermals, and/or maybe density as well.

Idontcare (the user) can best explain it.

I'm certain because we already have products made on 32nm from a few foundries and those have much better transistor density then Intel's 45nm.

AMD trinity 32nm:
Transistors: 1.303 billion
Die size: 246 mm²
Intel dunnington:
1.9 billion
Die size: 503 mm²

Clearly GF's 32nm process can pack more transistors per mm2 then Intel's process at 45nm.

 

[taltamir]

Very cool. :thumbsup:

Reading the article, the main reason seems to be because the US government wants their chips made in the USA. Guess that is one way to push for new fabs in the US...

I was wondering why ANYONE would be crazy enough to open a factory in the USA (high taxes, high labor costs, high environmental regulation costs)... Everyone is outsourcing production away from USA for a reason. I am shocked that this government is actually wising up and insisting/promoting the building of such a critical piece of infrastructure locally.

The very foundation of modern societies is computer technology. All our science, knowledge, manufacturing, transportation, public records, education, and entertainment are based on it.

Yet china is the only country who seems to care (they are investing in developing their own national CPU) and no country considers the construction of electronics fabs locally to be part of national defense.

Then there is hardening of such facilities... they should be built under mountains and EMP shielded (even if not a hostile act, a major solar flare can occur)

mmm... or maybe just store a large supply of replacements underground rather then the production facilities... that also works.

 

[taltamir]

I'm certain because we already have products made on 32nm from a few foundries and those have much better transistor density then Intel's 45nm.

AMD trinity 32nm:
Transistors: 1.303 billion
Die size: 246 mm²
Intel dunnington:
1.9 billion
Die size: 503 mm²

Clearly GF's 32nm process can pack more transistors per mm2 then Intel's process at 45nm.

Did you make this comparison (of 1 out of the 3 aspects which I described with "typically", "generation", and "competition"?) before or after you said with certainty that intel can't be ahead?

Also, please correct me if I am wrong but the "certainly" statement you made sounded like you were saying X nm is X nm in terms of transistor density no matter who makes them. Rather then "they vary in transistor density, but a full node ahead has always had better density despite said variance"

Lets look at the numbers you posted though...

AMD 32nm
1303 million transistors / 246 mm² = 5.297 mt/mm²
Intel 45nm
1900 million transistors / 503 mm² = 3.777 mt/mm²

That looks more like a half node rather then full node difference (only 1.4x the transistors on the 32nm process).

Now lets look at intel's 32nm process with ivy bridge:
1400 million transistors / 160 mm² = 8.75 mt/mm²
2.3x the xtors density going from 45 to 32nm for intel.

8.75 mt/mm² for intel 32nm process compared to 5.297 mt/mm² on GF's 32nm process.... so what are those nm measuring exactly since the density varies so much?

http://en.wikipedia.org/wiki/Intel_Ivy_Bridge#Models_and_steppings
http://www.intel.com/content/dam/ww...rd-gen-core-family-mobile-vol-1-datasheet.pdf
http://www.intel.com/content/dam/ww...rd-gen-core-family-mobile-vol-1-datasheet.pdf

EDIT: Doing a quick google search I misread a chart and came out from it with Ivy bridge being 32nm. Its not, its 22nm. See correction below

http://www.anandtech.com/show/2930
westmere is intel 32nm and: "240mm^2 chip made up of 1.17B"
So that is
1170 million transistors / 240 mm² = 4.875 mt/mm²

Which is 29% better transistor density then GF's identically named process.

 

[Lepton87]

Did you make this comparison (of 1 out of the 3 aspects which I described with "typically", "generation", and "competition"?) before or after you said with certainty that intel can't be ahead?

Also, please correct me if I am wrong but the "certainly" statement you made sounded like you were saying X nm is X nm in terms of transistor density no matter who makes them. Rather then "they vary in transistor density, but a full node ahead has always had better density despite said variance"

Lets look at the numbers you posted though...

AMD 32nm
1303 million transistors / 246 mm² = 5.297 mt/mm²
Intel 45nm
1900 million transistors / 503 mm² = 3.777 mt/mm²

That looks more like a half node rather then full node difference (only 1.4x the transistors on the 32nm process).

Now lets look at intel's 32nm process with ivy bridge:
1400 million transistors / 160 mm² = 8.75 mt/mm²
2.3x the xtors density going from 45 to 32nm for intel.

8.75 mt/mm² for intel 32nm process compared to 5.297 mt/mm² on GF's 32nm process.... so what are those nm measuring exactly since the density varies so much?

http://en.wikipedia.org/wiki/Intel_Ivy_Bridge#Models_and_steppings
http://www.intel.com/content/dam/ww...rd-gen-core-family-mobile-vol-1-datasheet.pdf
http://www.intel.com/content/dam/ww...rd-gen-core-family-mobile-vol-1-datasheet.pdf

Ivy bridge is at 22nm.

The other thing to add is that some elements in ICs shrink well and some don't, some can be made denser then others. So you mistook 22nm process for 32nm. GF 32nm process should be denser then Intel's because it's gate first and Intel uses gate last in their processes. About the only advantage of gate first is transistor density, everything else is much worse like drive current. That is one of the reason BD was such a failure and PD is nothing to write home about, although it's faster then it's price-wise competition in very well threaded workloads. OC is another matter, PD has a healthy clock advantage at stock that usually reverses when compared to SB and clocks about the same as IVY.
Update: well to be fair to AMD FX-8320 is one of a few AMD processor's I would pick over Intel's price equivalent. At 169$ it's only competition is very OC limited. 3470 3.2 GHz base/3.6GHz turbo would be its competition, AFAIR intel's non-k model can be overclocked 4 bins over turbo clock, so it would be up against Ivy at 4GHz. That still does not sound good. After some thinking I revise my statement and would still pick non-k intel processor. Ivy without moding don't like to be clocked higher then 4.5GHz so that's no much performance lost. Cheapening out on solder and not even properly aligning IHS so it made actual contact with the die is so sad turn of events. That's where no competition leads.

 

[taltamir]

Ivy bridge is at 22nm.

So it is... My apologies...
Lets see then:
http://www.anandtech.com/show/2930
westmere is intel 32nm and:

240mm^2 chip made up of 1.17B

So that is
1170 million transistors / 240 mm² = 4.875 mt/mm²
Which is 29% better then GF's identically named process.

The other thing to add is that some elements in ICs shrink well and some don't, some can be made denser then others.

So you mistook 22nm process for 32nm.

I carelessly misread a chart. Has nothing to do with any preconceptions or what have you.

GF 32nm process should be denser then Intel's because it's gate first and Intel uses gate last in their processes.

So... you totally agree with me then that intel has a superior process which produces smaller chips for "identical" names?

Which brings the question of what 32nm measures exactly. Idontcare explained it quite well some time ago, the nm figure is fairly arbitrary nowadays, it used to have meaning and not anymore.

 

[Lepton87]

Did you make this comparison (of 1 out of the 3 aspects which I described with "typically", "generation", and "competition"?) before or after you said with certainty that intel can't be ahead?

Also, please correct me if I am wrong but the "certainly" statement you made sounded like you were saying X nm is X nm in terms of transistor density no matter who makes them. Rather then "they vary in transistor density, but a full node ahead has always had better density despite said variance"

Lets look at the numbers you posted though...

AMD 32nm
1303 million transistors / 246 mm² = 5.297 mt/mm²
Intel 45nm
1900 million transistors / 503 mm² = 3.777 mt/mm²

That looks more like a half node rather then full node difference (only 1.4x the transistors on the 32nm process).

Now lets look at intel's 32nm process with ivy bridge:
1400 million transistors / 160 mm² = 8.75 mt/mm²
2.3x the xtors density going from 45 to 32nm for intel.

8.75 mt/mm² for intel 32nm process compared to 5.297 mt/mm² on GF's 32nm process.... so what are those nm measuring exactly since the density varies so much?

http://en.wikipedia.org/wiki/Intel_Ivy_Bridge#Models_and_steppings
http://www.intel.com/content/dam/ww...rd-gen-core-family-mobile-vol-1-datasheet.pdf
http://www.intel.com/content/dam/ww...rd-gen-core-family-mobile-vol-1-datasheet.pdf

EDIT: Doing a quick google search I misread a chart and came out from it with Ivy bridge being 32nm. Its not, its 22nm. See correction below

http://www.anandtech.com/show/2930
westmere is intel 32nm and: "240mm^2 chip made up of 1.17B"
So that is
1170 million transistors / 240 mm² = 4.875 mt/mm²

Which is 29% better transistor density then GF's identically named process.

By your own calculations:
1303 million transistors / 246 mm² = 5.297 mt/mm² AMD

1170 million transistors / 240 mm² = 4.875 mt/mm² Intel you still messed up numbers.

So what process is more dense? As I said gate first has the advantage of density at the expense of everything else. It was really a terrible idea. I think if PD was made on Intel's 32nm process it should easily has stock speed over 4,5GHz and and OC well past 5GHz. Maybe even 5GHz stock speed would be possible. 32nm stars cores clocked WORSE then the ones made on 45nm process which easily reached past 4GHz. So AMD made a high clocking CPU on a process that just could't deliver high clocks, that is part of what made it such an unmitigated disaster.

 

[taltamir]

By your own calculations:
1303 million transistors / 246 mm² = 5.297 mt/mm² AMD

1170 million transistors / 240 mm² = 4.875 mt/mm² Intel you still messed up numbers.

So what process is more dense?

Yes, you are right. My mistake

As I said gate first has the advantage of density at the expense of everything else. It was really a terrible idea. I think if PD was made on Intel's 32nm process it should easily has stock speed over 4,5GHz and and OC well past 5GHz. Maybe even 5GHz stock speed would be possible. 32nm stars cores clocked WORSE then the ones made on 45nm process which easily reached past 4GHz. So AMD made a high clocking CPU on a process that just could't deliver high clocks, that is part of what made it such an unmitigated disaster.

Good points all around.

 

[thilanliyan]

that is part of what made it such an unmitigated disaster.

*My 6100 goes and cries in the corner*



I really wish they would release something more power efficient...close to stock this 6100 is not so bad but try to OC it with voltage increases and power consumption goes through the roof. I was hoping for much better thermals with Piledriver but that didn't really pan out I guess. And I'm trying to prolong this motherboard purchase...otherwise I would switch to Intel in a heart beat.