Intel Golftown CPU to have 6 cores, 3GHz, cost $1999!

[Idontcare]

Originally posted by: thilanliyan

Originally posted by: LoneNinja
12core server processor coming out 1H of 2010 called mangy cores(my spelling may be wrong on that one).

Lol at "Mangee" cores ...I believe it's Magny Cours named after the F1 track.

wE aLL kNOW iTS cODE fOR Many Cores...I rEAD iT oN mACRUMORS.com

 

[manimal]

hmm golftown I have heard senior citizens call myrtle beach golftown so this must be code for some new race track they are building in myrtle beach where they will someday race cars and have a chip named after it..

 

[aigomorla]

Originally posted by: LCD123
http://forums.macrumors.com/showthread.php?t=804189

Here's some facts I found.

1. 6 cores is very difficult to make and AMD can't compete so it shall cost $1999!
Oh really i find it funny how people know when there still on NDA

2. Engineering samples are 2.4GHz so final clocks will be up to 3GHz and overclock up to
3.6GHz but will do 4GHz on exotic cooling. Those cpus are heat limited and stock cooling is high end!
Ummm no... cant tell you what they can do.. but no... this statement almost made me spit coffee at my monitor

3. There won't be quad core on 32nm but will for 22nm. Wait for 22nm if you want higher clocked hexcore cpus.
Once again... Hexcore is still on NDA. They dont even come out until Apirl. Do you think the average consumer will get intel's product map when lately intel has been paranoid with AMD?

4. Golftown(i9?) is 20% faster clock for clock than i7
No comment.. maybe.. maybe not....

5. Golftown is server only. May work in x58 mobos however with a special modified bios.
BWAHAHAHAHAHHAHAHAHAHA OK bud... Tell my Computer called Nedeshiko that. Her i9 has a black logo (hint hint).

6. We may see dual cores on 32nm, they will be similar to e8600s and clock the same and be 20% faster clock for clock and cost $500+
Sigh I love bashing gulftown threads because all i see is BS from people who spew them without owning one.

See comments.. with a big Sigh.. double Sigh.. and Triple Sigh...

Im sure all the mods here agree with me on that statement....

I ask you guys wait a few more weeks while my red tape gets lifted...
Once that gets lifted.. i'll spank you guys with all the answers you are looking for.

 

[CoreEye7]

Originally posted by: LCD123
http://forums.macrumors.com/showthread.php?t=804189

Here's some facts I found.

1. 6 cores is very difficult to make and AMD can't compete so it shall cost $1999!

2. Engineering samples are 2.4GHz so final clocks will be up to 3GHz and overclock up to 3.6GHz but will do 4GHz on exotic cooling. Those cpus are heat limited and stock cooling is high end!

3. There won't be quad core on 32nm but will for 22nm. Wait for 22nm if you want higher clocked hexcore cpus.

4. Golftown(i9?) is 20% faster clock for clock than i7

5. Golftown is server only. May work in x58 mobos however with a special modified bios.

6. We may see dual cores on 32nm, they will be similar to e8600s and clock the same and be 20% faster clock for clock and cost $500+

Wrong. Just wrong. On nearly every "fact" posted above. Not to mention where does the information come from- MAC FORUMS! hahahaha now that's funny. O and 4ghz on exotic cooling? Click on the link buddy, that's 5ghz AND an early engineering sample. Please stop posting when you don't have a CLUE what you're talking about. Thanks!

http://www.xtremesystems.org/f...d.php?t=232109&page=11

 

[BD231]

$1999 bucks for Intel??

I'd rather buy a cheap car and drive it into a wall.

 

[aigomorla]

Originally posted by: BD231
$1999 bucks for Intel??

I'd rather buy a cheap car and drive it into a wall.

that will probably be the cost on the highest end Xeon side with 2xqpi.

 

[yh125d]

I can't figure out whether the OP is a serious thread or not


Obviously the responses are silly but OP is confusing

 

[ghost recon88]

If you guys really want to see how Gulftown performs, see the following threads. This kind of info isn't that hard to find

http://kingpincooling.com/forum/showthread.php?t=131

http://www.xtremesystems.org/f...howthread.php?t=237257

http://www.xtremesystems.org/f...howthread.php?t=235767

http://www.xtremesystems.org/f...howthread.php?t=232109


And yes, extreme overclocking is the only kind of overclocking that matters with these chips. If anyone buys one to run on air cooling, they'll wake up the next morning missing their prunes.

 

[Idontcare]

I can't figure out whether the OP is a serious thread or not


Obviously the responses are silly but OP is confusing

I think it is safe to say the OP is taking themselves quite seriously. I don't think this was intended to be a joke-thread.

 

[aigomorla]

And yes, extreme overclocking is the only kind of overclocking that matters with these chips. If anyone buys one to run on air cooling, they'll wake up the next morning missing their prunes.

Ummmmm... no...
They do pretty well on air...

Maybe you missed how coretemps is reading tjunct. and not actualy C.
So the number you see, subtract 105 from it.

Gulftown is meant for both the extreme and non extreme.
Most of the benchmarks you see sub 5ghz is all done on air.

 

[justinm]

lol, I needed a good laugh.

 

[aigomorla]

Who can guess what that is?
(sorry its not mine.. i dont have the balls to cut open a unlreased processor like that)

LOL..

 

[Idontcare]

Who can guess what that is?
(sorry its not mine.. i dont have the balls to cut open a unlreased processor like that)

LOL..

The aspect ratio of nearly 2:1 suggests that is a 4C Sandy Bridge.

 

[aigomorla]

The aspect ratio of nearly 2:1 suggests that is a 4C Sandy Bridge.

its gulftown.. and its a big die...

 

[Idontcare]

its gulftown.. and its a big die...

Shouldn't a 6core 12MB Gulftown on 32nm be smaller than a 4core 8MB 45nm Bloomfield?

Usually node entitlement areal scaling is targeted to come in around 50-60%...so I'd expect a 32nm chip that has 50% more circuitry of similar architecture to be about 20-25% smaller in diesize than a 45nm chip.

 

[OCGuy]

$1999 bucks for Intel??

I'd rather buy a cheap car and drive it into a wall.

Why not buy AMD and hit the OC wall....it isnt that hard to do, and would be a lot safer.

 

[aigomorla]

Shouldn't a 6core 12MB Gulftown on 32nm be smaller than a 4core 8MB 45nm Bloomfield?

Usually node entitlement areal scaling is targeted to come in around 50-60%...so I'd expect a 32nm chip that has 50% more circuitry of similar architecture to be about 20-25% smaller in diesize than a 45nm chip.

Thats what i thought too.

But when we analyzed it, its thinner, but longer... (umm that doesnt sound right).

 

[Idontcare]

What is the diesize then? Serious fail by Intel's 32nm process tech team if the areal scaling for 32nm over 45nm is a meager 25% as needed to result in a 6C gulftown to be larger than a 4C bloomfield. Big opportunity for GF's 32nm design team to step ahead on areal scaling if this is true...

 

[aigomorla]

well i know one thing..

its hard as hell to get even temps on it.

Once i get an okey, i'll show u guys what i mean.

 

[nyker96]

only cost $2k? alright get me a doz. :}

 

[ilkhan]

Actually, scaling the existing die shots of bloomfields makes it a little wider than the sandy bridge die shot, when adjusted to be the same height. So that die does make sense as a gulftown.

Keeping all the cores evenly cooled (as alluded to earlier) will be a chore, as it IS so wide. I hope the hottest cores (inside or outside) get turned off first under turbo mode.

 

[IntelUser2000]

What is the diesize then? Serious fail by Intel's 32nm process tech team if the areal scaling for 32nm over 45nm is a meager 25% as needed to result in a 6C gulftown to be larger than a 4C bloomfield. Big opportunity for GF's 32nm design team to step ahead on areal scaling if this is true...

It's that time of the year(month maybe...) again! Analyze die sizes! XD

Rough pixel scaling with paint have resulted in...

most optimistic die size of Gulftown at 255mm2.

The scaling isn't too bad. It results in areal scaling of 65%(0.97 / 1.5). Remember when we were discussing Clarkdale die sizes and I estimated 90-100 due to historical scaling trend of 70% logic and 50% SRAM on Intel processors based on last few years? Well that kinda seems to hold here. Well, debated topic among some people is that I/O blocks like QPI/memory controller/bus logic scales worse than CPU core.


Except....

That's not Gulftown...

This is...

http://www.bit-tech.net/news/hardware/2009/10/30/six-core-core-i7-photographed/1

See the back side? It's not really wide.

The longer side of the die(Aigomorla) is approximately 50% of the side of the whole CPU package and the shorter side is 30% of the total package size.

S1366 package size:
42.5mm2 x 45mm2

S1156 package size:
37.5mm2 x 37.5mm2

You'll see if you do the calculations on the 37.5mm2 x 37.5mm2 it comes out to be around 210mm2(Similar to rumored Sandy Bridge die, coincidence?) for the link from this thread. If we assume its a S1366 pic and the package size is 42.5mm2 x 45mm2, the die size is rather large 255mm2(very optimistic projection).

Short story: The long and wide die in this thread is Sandy Bridge. The one from Bit-Tech is Gulftown.

So what's the real size of Gulftown? It doesn't seem to be that small, but it is smaller. It should be 235-240mm2.

 

[aigomorla]

i could of sworn the guy who i stole that picture off from said it was a Q3FE which means gulftown.

 

[Idontcare]

Actually, scaling the existing die shots of bloomfields makes it a little wider than the sandy bridge die shot, when adjusted to be the same height. So that die does make sense as a gulftown.

Keeping all the cores evenly cooled (as alluded to earlier) will be a chore, as it IS so wide. I hope the hottest cores (inside or outside) get turned off first under turbo mode.

If it is Sandy, the 2:1 aspect ratio is caused by the addition of the on-die IGP.

Who cares about how even the temps are across the die? All you care about is the max temp at any given location within the die as it is those transistors which are operating in that max temp environment who are dealing with the worst signal/noise ratios and the most challenging reliability scenario versus the rest of the transistors.

But Turbo isn't about managing temps, its about managing power-consumption. Personally I'm more interested in seeing Sandy come with individual Vcc lines for each core so that turbo can maximize the GHz/Vcc (shmoo) plot on a per-core basis while operating within a required power envelope at any given temperature (minimum necessary Vcc to stably operate at any given GHz is temperature dependent, the higher the temp the higher the needed Vcc).

Sandy Bridge Die-shot thread

 

[ilkhan]

even die temps would be required to have all the cores max out at the same time under load. If one core is inadequately cooled, it'll hold the rest of the cores back for overclocking. Right?

You think that sandy has individual Vcc lines? Or are you hoping for that?