brisbane dual core 65nm

Quote:

---

Originally Posted by metro.cl

report:

AM2 65nm is to hot to OC :s

so bad clockers

---

And wheres the logic in saying its hotter when. 1.There are no parts to give Temp results of to compare. 2.Its 65nm at the same volts as the 90nm and 65w. So does that make sense a smaller process on SOI-3 with the same watts and volts as 90nm ones only on a smaller die. Should run cooler at the same speed since its K8. Where does the enormous heat come from? It doesn't have the size to generate more heat unless more volts and watts are used. But it only runs at 1.20v to 1.30v same as 90nm can and would run cooler then the 89w vers but on a smaller die it would be expected to run much cooler. It would have to be a total arc change for it to run hotter. Are we missing something here? There is no way to know if they are or not unless you can show us your magical part that no body else has but you to prove your clam because its just BS right now with out PROOF ofcorse! :rolleyes:

Ill believe it when I see it.

If 65nm is too hot to overclock, what is the point in it?

In all due respect to metro.cl, his word could mean anything it doesnt mean he is right.

Like I said, wait and see.

Quote:

---

Originally Posted by metro.cl

report:

AM2 65nm is to hot to OC :s

so bad clockers

---

damn that is rather unexpected.
Should hopefully be fixed shortly

Quote:

---

Originally Posted by brentpresley

The POINT w/ 65nm is that AMD can make almost TWICE the CPUs from one wafer as they can from 90nm. This is EXTREMELY important when you only have 2 fabs and can't outsource more than 20% of your production to a 3rd party (a la the 1995 AMD/Intel settlement terms).

With almost ALL of their CPUs going to DELL and HP, they really don't give a RAT'S A$$ if they OC well or not. Us overclockers are not even a residual thought to them.

---

umm because Dell Computers never should be overclocked anyways. Wait a bit and I am sure they will come out with some wonderfully overclockable chips

One thing that needs to be addressed to is C2D has 4mb cache on die, these X2's have 1mb. That is a LOT of mm2 of die space less with which the X2's have to dissipate heat. Anyone know the die size of X2's on 65nm? This might not be a big difference but when you throw some voltage and clocks together into that smaller die it could be making them appear hotter than they really are.

Quote:

---

Originally Posted by metro.cl

report:

AM2 65nm is to hot to OC :s

so bad clockers

---

Thanks fella . I suspected this might be the case with SOI @ 65. See past threads. Its to bad maybe AMD can fix this but Intel couldn't fix @ 90nm.

Quote:

---

Originally Posted by Serge84

And wheres the logic in saying its hotter when. 1.There are no parts to give Temp results of to compare. 2.Its 65nm at the same volts as the 90nm and 65w. So does that make sense a smaller process on SOI-3 with the same watts and volts as 90nm ones only on a smaller die. Should run cooler at the same speed since its K8. Where does the enormous heat come from? It doesn't have the size to generate more heat unless more volts and watts are used. But it only runs at 1.20v to 1.30v same as 90nm can and would run cooler then the 89w vers but on a smaller die it would be expected to run much cooler. It would have to be a total arc change for it to run hotter. Are we missing something here? There is no way to know if they are or not unless you can show us your magical part that no body else has but you to prove your clam because its just BS right now with out PROOF ofcorse! :rolleyes:

---

Relax Serge . It is widely known that the smaller the process on SOI the more leakage. Relax its not the end of the world. You can wait until the release and we see reviews . Befor you call metro out. He seems to have provided good info in the past.

Quote:

---

Originally Posted by metro.cl

AM2 65nm is to hot to OC :s

---

where is the proof?
65nm X2 are 65Watt cpu (and later even 35Watt!!!) like C2D
it cannot be hotter - maybe early ES were

though OC-ability can be restricted by its IMC

Quote:

---

Originally Posted by brentpresley

I remember something about 115 - 130mm2, but that is just off the top of my head and may be wrong until I get some time to dig up the reference.

EDIT:
Here it is for 90nm process:
http://www.techwarelabs.com/reviews/.../index_2.shtml

199mm2 for 90nm

Extrapolation of APPROXIMATE size on 65nm:
((65^2)/(90^2))*199mm2 = 104mm2 give or take a bit

---

It's a speculation based on 0.6-0.7 scaling factor. It will be closer to 0.7 due to fact that this shrink is using old transistor design which can't be fully scaled to 65nm.

Metro is telling truth :( . Smaller die @same voltage has greater thermal density.
In bigger die even if you're using more power it's transferred to radiator on bigger area, that is why monsters like Itanium or Power5 can consume up to 200W and still can be cooled efficiently.

Now some math:

89W-->182mm2 (windsor)
'x'W-->125mm2 (where x is desired TDP to maintain same thermal density)
x=89*125/182=61W

Basically thermal density is higher on 65nm than on 90nm process. This is small difference so I conclude that 65nm CPU+Zalman 9500 should go up to 3GHz but over this value it will be very hard.

This is very simplistic approach to complex problem. Other factors are how well new process is implemented, how much 'long lines' was shortened in this process, how this affected electron movement, etc.

AMD certainly will improve over time, for now we even can't buy those chips :slapass: .

It's annoying to NOT KNOW in real terms how performing the newer AMD chips will be (K8L, 65nm chips) or how they overclock, specially when we knew last march / april that the C2D would whipe the floor (4 months before they came available). And, when you put your hopes on a flawed concept like the yet unreleased 4X4 and the reviews confirm something that you hope wouldn't happen, which would be a very consuming platform that delivers not so good performance, and that it is directly competing against a very efficient quad core from INTEL.

Since AMD wants us to wait, let's wait. Maybe there's a conspiracy going on that we aren't aware of...

From a cost standpoint you would think AMD would be all about getting 65nm out the door even if they run a little hot. They guarantee a chip to run at X speed, not 1ghz over it. You can't tell me 65nm is SO broke they cannot make 2ghz-2.4ghz cores (majority of official sales) now to cut their costs by 1/2. I thought they had production worthy wafers coming out of Fab36 since October?

Quote:

---

Originally Posted by metro.cl

report:

AM2 65nm is to hot to OC :s

so bad clockers

---

Too hot to be true?

Quote:

---

Originally Posted by ***Deimos***

Speaking purely from a hobby/enthusiast point of view:
1. Dual/quad core is nearly absolutely useless for games. Heck, it doesn't do anything for SuperPI or 3DMark 99-05 either.
2. All this C2D vs K8L is making me nosious. You could have put together a sweet E6400 rig in July 2006. K8L.. will C2D even stay around in stores that long for a comparison?
3. CPU... I dont know. It all seems so 2001. Truthfully more of a 20th century kind of thing. You know, like back when running functions in a spreadsheet actually took seconds, you left the system running overnight to encode something, fired up your old trusty 56K in search of optimized codecs to play those super demanding DIVX movies, you went out of your way to get a MPEG2 hardware decoder card just to play DVD's or those agrevating waits to play shockwave games.

And all the last 4-5 years that WinXP has been around, the best thing they could come up with as a killer app WAS Dragon Naturally Speaking. Where has all that CPU horsepower gone to in the last couple years?

Microsoft and company, just keep making more and more enormous programs with gigantic memory footprints, flashy special effects everywhere marketed to "improve" productivity, and neverending stream of intelligent "auto-do" (re: aggrevate) features. Pardon me, but at the crossroads when a word processor, an applications which by its very definition is meant to do the most mundane trivial task, becomes so demanding it requires you to purchase a new system, we're all in deep deep doo-doo.

FYI: Word95 ran perfectly well on a 8MB 386DX33 hotrod. Fit quite well on 50MB hard drive too. As for games, I hope PC market lives on to see another decade - everybody for years has known the key, above all else, is getting newer faster video cards. Sure a $999 processor upgrade can make the game run as much as 20% faster... but a $200 video card upgrade can easily boost frame rates 200% or more.

I hope I got my points across...

---

I have to disagree with duo and quad core being useless, i am in the Supreme Commander open beta. The difference between single core and dual core gaming in this beta is like night and day. You dont want to be playing the next generation games without at least dual core, trust me on that :eek: . Not to mention the half life 2 source engine will soon support multiple cores.

Quote:

---

Originally Posted by Lightman

Now some math:

89W-->182mm2 (windsor)
'x'W-->125mm2 (where x is desired TDP to maintain same thermal density)
x=89*125/182=61W

---

The formula is right, but the Watts are not. AMD TDP works like different Power Consumptions categories that usually shares a single model of Heatsinks bewthem them, but it says nothing about real power consumption. That is what we are needing to make a more accurate comparision bewthem them, because on the last shrink, power consumption and temperature got lowered substancially.
Besides, Voltage plays a role in this too. Maybe 65nm K8s can run at a much lower Voltage than the nominal value for archieving that Frequency that a comparable 90nm K8.

Quote:

---

Originally Posted by zir_blazer

The formula is right, but the Watts are not. AMD TDP works like different Power Consumptions categories that usually shares a single model of Heatsinks bewthem them, but it says nothing about real power consumption. That is what we are needing to make a more accurate comparision bewthem them, because on the last shrink, power consumption and temperature got lowered substancially.
Besides, Voltage plays a role in this too. Maybe 65nm K8s can run at a much lower Voltage than the nominal value for archieving that Frequency that a comparable 90nm K8.

---

That is right.
My point was that power density on new 65nm process at least on beginning is higher than on latest 90nm. That's why I used TDP instead of actual power consumption numbers. I'm hoping that you can run 65nm K8 at 1.0V at least up to 2.4GHz, but we need silicon on hand to test it :dammit: .

If AMD could produce quality on par with what a healthy die shrink should be producing, we would be seeing chips by now. I do think that part of the delays and speculation is due to the fact that they did not convert their existing fab into 65nm but built an entire new facility to do it. anything think that Fab36 as a site could be causing the delays vs purely from a problem with the shrink?

Are some IBM silicon currently using SiGe in the wafers? I don't follow their mfg processes.

Quote:

---

Originally Posted by ***Deimos***

1. Dual/quad core is nearly absolutely useless for games. Heck, it doesn't do anything for SuperPI or 3DMark 99-05 either.

---

All these Apps were written before multiple cores existed...New applications will do a much better job of utilising multiple cores...

YA IBM used strained silcon on SOI. but it won't go on 45nm process to much leakage.

The IBM-led, collaborative effort will deploy strained silicon for the 45-nm node, but not silicon-on-insulator (SOI) technology. "Advanced strain engineering techniques are the cornerstone for performance," according to a spokesman for IBM. "SOI is a separate effort and is available through IBM, but is not part of the IBM-Chartered-Infineon-Samsung common platform."

Quote:

---

Originally Posted by brentpresley

I can EASILY be hotter when going to a smaller process.

Remember Prescott? Wattage went through the roof on that one going from 130nm to 90nm. It was because leakage went up by SEVERAL orders of magnitude.

And you should be REALLY careful about calling out a senior member like metro.cl.

He's been known to come back w/ some stuff that makes people just look BAD when they call him out. :D


EDIT - my GUT tells me that if this is true and 65nm has some heat issues (or not necessarily issues, just something that prevents it from OCing), that AMD is doing the RIGHT thing by putting chips out NOW. This gives them time to prefect the 65nm "sauce" before K8L gets cranking next summer.

---

lol seems that you have way to much faith in my contacts

Prescott is not hotter than Northwood due to the manufacturing process, but because it has twice as much transistors.

Northwood = 55M
Prescott = 125M
http://www.aceshardware.com/read.jsp?id=60000315

Initial AMD 90nm process Winchester core is less power hungry than latter Venice, but the Dual Stress Liner process (in Rev E Venice) helps it to work at higher clock frequencies.
http://www.xbitlabs.com/articles/cpu...-venice_5.html

It's unclear if AMD is already using all 3rd gen SOI features in Rev G Brisbane, but even working with almost the same vcore, Brisbane's power consumption is about 30% lower than Rev F Windsor (65w vs 89w at 2.6GHz).

Quote:

---

Originally Posted by brentpresley

The POINT w/ 65nm is that AMD can make almost TWICE the CPUs from one wafer as they can from 90nm. This is EXTREMELY important when you only have 2 fabs and can't outsource more than 20% of your production to a 3rd party (a la the 1995 AMD/Intel settlement terms).

With almost ALL of their CPUs going to DELL and HP, they really don't give a RAT'S A$$ if they OC well or not. Us overclockers are not even a residual thought to them.

---

Wow, so there is intelligence on this planet... perhaps there's hope for womankind afterall (there never was any hope for man to begin with ;)

Quote:

---

Originally Posted by doompc

It's unclear if AMD is already using all 3rd gen SOI features in Rev G Brisbane, but even working with almost the same vcore, Brisbane's power consumption is about 30% lower than Rev F Windsor (65w vs 89w at 2.6GHz).

---

Read my last Post. We can make no assumptions solely based on the TDP because that is not what the Processor truely consumes. Besides, I don't recall that any Rev. F K8s Processor only power consumption Benchmarks have been done to do an accurate comparision (The DDR-II Memory Controller should consume more).

Quote:

---

Originally Posted by brentpresley

The POINT w/ 65nm is that AMD can make almost TWICE the CPUs from one wafer as they can from 90nm. This is EXTREMELY important when you only have 2 fabs and can't outsource more than 20% of your production to a 3rd party (a la the 1995 AMD/Intel settlement terms).

With almost ALL of their CPUs going to DELL and HP, they really don't give a RAT'S A$$ if they OC well or not. Us overclockers are not even a residual thought to them.

---

Unfortunately, the shrink is lame, and roughly 68% the size of 90nm, so even with the same yield %, they would only get 1.47 times as many parts. But even worse, AMD is only claiming to get at least as many good 65nm parts from a wafer as they were 90nm, according to the Inquirer, which said AMD defined "mature yields" to mean just that.

Not surprising that these overclock poorly, given the launch bins compared to 90nm.

Quote:

---

Originally Posted by Serge84

And wheres the logic in saying its hotter when. 1.There are no parts to give Temp results of to compare. 2.Its 65nm at the same volts as the 90nm and 65w. So does that make sense a smaller process on SOI-3 with the same watts and volts as 90nm ones only on a smaller die. Should run cooler at the same speed since its K8. Where does the enormous heat come from? It doesn't have the size to generate more heat unless more volts and watts are used. But it only runs at 1.20v to 1.30v same as 90nm can and would run cooler then the 89w vers but on a smaller die it would be expected to run much cooler. It would have to be a total arc change for it to run hotter. Are we missing something here? There is no way to know if they are or not unless you can show us your magical part that no body else has but you to prove your clam because its just BS right now with out PROOF ofcorse! :rolleyes:

---

OMG (*smack self in face and tries to regain composure*)
I'm not sure what exactly we're debating here, but I'll assume you're questioning how the same 65nm processor can be hotter than a 90nm.
1. Typically, in a perfect world, linearly scaled down transistors would run faster and cooler. However, when gate oxides shrank down to around num of atoms you can count on your toes, we ran into trouble.. big BIG trouble.
2. Have you heard of static leakage current. When just recently Intel had such huge problems with Prescots, why would you think AMD would be immune? Static power is growing exponentially with each die shrink!
3. The smaller the die size, the greater the power/heat concentration. We're approaching levels in a nuclear reactor. Those bygone days of dinky aluminum heatsink are long over. And perhaps soon we might reach limits of copper's heat conductivity.
4. This is AMD's first attempt at 65nm. Ofcourse it will take some time to fine tune the technology. Especially if new materials and techniques are being used. Intel wisely had already gone through the gruesome trials nearly a year earlier before Core2Duo.

Quote:

---

Originally Posted by brentpresley

I remember something about 115 - 130mm2, but that is just off the top of my head and may be wrong until I get some time to dig up the reference.

EDIT:
Here it is for 90nm process:
http://www.techwarelabs.com/reviews/.../index_2.shtml

199mm2 for 90nm

Extrapolation of APPROXIMATE size on 65nm:
((65^2)/(90^2))*199mm2 = 104mm2 give or take a bit

---

Wrong starting data.

90nm Rev F is 183mm^2 for the 512k L2 x 2 part (The 1MB x 2 is 230mm^2)

http://www.anandtech.com/cpuchipsets...oc.aspx?i=2795
http://www.extremetech.com/article2/...1966029,00.asp
etc.


Unfortunately, shrink sucked (.68 or so factor overall), and 65nm Brisbane is ~125mm^2.