A64 939 130nm vs A64 939 90nm

[dennisjai215]

so remember the first 1-2 reviews of the 90nm chips? reviewers were saying 90nm ran about 5c hotter in their benchmarks...

well reading some reviews today.. it said that the 90nm processors runs cooler.. and a member says

Originally posted by: cyberserf
I got the 3200+ from Monarch, seems cooler by 10 degrees compared to my 754 3200+ which was hitting 50c on average
also added the XP-120 thermalright heatsink from the 948U so I don't know if that is what is making the big difference or the 90nm process.
Regardless I am happy.

well we all assume a 948U and a XP120 couldnt be a difference of -15C can it?(since the 948U is a pretty high quality product iteself)

(review said +5c on the 90nm) and then member saying (10c cooler with 90nm) = 15C different

Any AT'ers have both chips to compare?

 

[dennisjai215]

 

[Concillian]

They're also 200 MHz different, with the S939 being slower. That will account for some, but not all of the difference. Good to see more support that at stock voltage these things appear to run pretty cool.

I'm more interested in overclocked temps, or more accurately over-volted temps, and if there's any difference in how the temps scale with voltage, but those will come with time, I'm sure.

 

[miketheidiot]

both run at 2.0 ghz. It's the two 3000's that run at diferent core speeds. (754 @2.0, 939 @1.8)

 

[DrMrLordX]

Er, actually mike

the Socket 754 Athlon 64 3200+ Newcastle runs at 2.2 ghz, not 2.0 ghz.

I sort of assumed that cyberser1 was comparing a 754 Newcastle 3200+ to a 939 Winchester 3200+. If he was using 754 Clawhammer 3200+, it would be running at 2.0 ghz.

 

[sonac]

I heard, that 90nm AMDs actualy make less heat than 130nm, but they run hotter because of smaller core --> smaller surface touching heatspredaer, so heat doesn't "run" so efficently.

 

[dennisjai215]

you guys are going off topic.. who cares about the mhz? im talking about comparing the rated speed and the heat diff

stay on topic plz

 

[BW86]

we need some more reviews on these chips

 

[dennisjai215]

bump

 

[LifeStealer]

It depends on what they are using to measure temps with. If they aren't both using the same external temp probe (Bios temp don't cut it) then you could easily get a +/- 10c difference. So there could be NO difference.

 

[Slaimus]

All the 90nm chips get larger temperature drops if you lower their voltage than 130nm. AMD only dropped the voltage by .1v to 1.4v. If they lowered the default voltage to 1.3v or 1.35v the temperatures would probably be lower than the 130nm parts.

 

[Avalon]

Originally posted by: dennisjai215
you guys are going off topic.. who cares about the mhz? im talking about comparing the rated speed and the heat diff

stay on topic plz

if (Mhz1 > Mhz2) {
heat++;
}

 

[jm0ris0n]

lol.

 

[dennisjai215]

Originally posted by: Avalon

Originally posted by: dennisjai215
you guys are going off topic.. who cares about the mhz? im talking about comparing the rated speed and the heat diff

stay on topic plz

if (Mhz1 > Mhz2) {
heat++;
}

but a 2.8e prescott runs hotter than a 3.2c northwood

HELLO THEY ARE DIFFERENT CORES so the god damn mhz thing doesnt matter all that much it just matters in the same performance/heat

 

[Ronin]

Actually, as the die gets small, the heat DOES go down. The 90nm chips are cooler than their 130nm counterparts.

 

[Concillian]

Higher MHz = more heat. That was my point.

The heat comparisons have not been xxxx+ rating to xxxx+ rating to say 90nm is lower, they have been xxxxMHz to xxxxMHz 90nm is lower.

The fact that you have a 90nm chip running 200 MHz slower than a 130nm chip means you're inherently building in an advantage. Don't believe that the "plus points" numbers mean identical performance because we've seen that dual channel is not anywhere near equal to 200 MHz in most circumstances.

I was just saying that that could account for some of the difference in heat. If you're going to compare the thermal performance of different gate sizes, you should really do it at the same speed, otherwise you aren't testing both on equal footing, one is given an inherent advantage.

You cannot compare the difference between Prescott and Northwood. Prescott is a completely different architecture, with millions more transistors than Northwood. The A64 move to 90nm is a simple die shrink and not a whole new chip.

 

[Ronin]

Speed doesn't factor into the die size, bud. That's the point here. The 90nm will run cooler period, than the same speed 130nm counterparts. It's basic math.

 

[Concillian]

Dude, speed and thermal power dissipated by a CPU are directly related. You can easily make a 90nm run hotter than a 130 by running the 130nm at 1000 MHz and the 90nm at 2400 MHz

If you check out the AMD docs for the last core they listed thermal power dissipation for every CPU they offered in a particular core type, you'll find this data on the Thunderbird core.

MHz / Therm. power
750 / 43.8
800 / 45.5
850 / 47.92
900 / 50.7
950 / 52.5
1000 / 54.3
1400 / 73.5

Now if you plot this out and do a curve fit you get a straight line with an R-squared of 0.9987. This means that thermal power dissipated by the CPU is DIRECTLY PROPORTIONAL TO SPEED.

There is nothing that has changed in more modern CPUs taht will change this. There is going to be a direct correlation to MHz and thermal power dissipated by the CPU.

The only case where I can think you could increase speed and power could be the same or go down is if you are opeating at the lower MHz and the voltage were well above what is needed to run at that speed, then you were to decrease the voltage when increasing core speed. This would result in a flat or negatively sloped line for thermal power dissipation. But it's not something ou typically see people doing.

 

[Ronin]

Did you honestly just compare a 2.4GHz CPU to a 1GHz one? Let's use some common sense here before posting something that's obvious to everyone. Also, note:

The 90nm will run cooler period, than the same speed 130nm counterparts. It's basic math.

 

[Concillian]

I'll admit to missing the 'same speed' part, which kinda makes my last post look ridiculous.

My original point was that the original poster was NOT comparing at the same speed, which would account for a certain amount of his difference.

Also, a die shrink will not necessarily mean lower heat if you don't take extra measures to control tunneling leakage from the gates. This is why there is so much talk about 90nm and 130nm parts. I would disagree that it's basic math that a die shrink = cooler. I think when we were in Pentium days when simple oxides were enough to control gate leakage, it may have been more basic math, but I think now things are quite a bit more complicated.

 

[jm0ris0n]

Yeah, but the reason everyon one is all up in arms is that while it may be a 200mhz difference, that speed increase is more or less insignificant (1-2C) when talking about the shift from 130nm to 90nm.

The original point of this thread was does the 90nm run cooler than the 130nm. Just add 2C to your 2ghz 939 score and then it should be comparable to a 2.2ghz 754.


So, anyone got any real data as the poster asked for

 

[dennisjai215]

Originally posted by: jm0ris0n
Yeah, but the reason everyon one is all up in arms is that while it may be a 200mhz difference, that speed increase is more or less insignificant (1-2C) when talking about the shift from 130nm to 90nm.

The original point of this thread was does the 90nm run cooler than the 130nm. Just add 2C to your 2ghz 939 score and then it should be comparable to a 2.2ghz 754.


So, anyone got any real data as the poster asked for

exactly, why are these fvckers who aren't answering my question posting here?

 

[Glavinsolo]

dennisjai215:

my post with a few temp readings

http://forums.anandtech.com/me...1736&enterthread=y

 

[Kocur]

Hey, just check this link:

http://xtremesystems.org/forums/showthread.php?t=43308



It seems to be quite interesting. Look at the voltage :Q!

There are also some reports on 3500+ running on air at 2.75 a little bit further in this thread.

:thumbsup: