Ideal Hi-end CPU to cool down - Winchester Athlons

Oliver · Post by **Oliver** » Thu Mar 10, 2005 9:31 pm

MikeC wrote:Here are some prelim results from SPCR's lab. Sean & I experimented with two A64s: a 3800+ (2.4G 130nm core) and a 3500+ (2.2Ghz 90nm core). We kept things as consistent as we could, and used minimal power draw components. The clock speed of the 2 CPUs was controlled by changing the multiplier in the BIOS and keeping CnQ turned off.

The system:
Soltek K8TPro VIA chipset board (??20W??)
Matrox G550 VGA (??W??)
IBM 5400/8mb cache 80G 2.5" drive (1W idle, 5W peak)
512mb DDR400 memory (??W??)
Silverstone Fanless PSU (~78% efficiency)

DC system power draw
@ 2.2G (200x12 -- max speed of the Winchester core CPU with bus at 200MHz)
130nm core: 115W
90nm core: 81W
difference: 34W!

@1.8G (200x9 -- The 3000+ A64s run at this speed)
130nm core: 100W
90nm core: 73W
difference: 27W!

With CnQ enabled, the system with Winchester at idle @ 1.8G (3000+) drew only 49W AC!

There isn't much more to say except that even at 1.8G, this is one fast cool system, esp w/ dual ch mem. The temp monitoring system in these boards was not working properly but at no point during the testing did the Winchester HS ever get too hot to touch. It barely seemed to get warm. This is with a pretty low airflow very stock type of HS.

Sean is now hatching all manner of plots to dump his current system onto his poor unsuspecting brother so that he can get his hands on a cool Winchester system.

What would the result have been had you used an actual socket 939 90nm 3000+ 1.8Ghz cpu rather than the 3500+ clocked down to 1800Mhz?

I ask because the Thermal Design Power (TDP) of the actual 3000+ chip
is 67W; but the TDP for the 3500+ chip in Intermediate Phase State 2 is 46W, which is by definition that processor running at 1.8Ghz. So does that mean the actual 3000+ chip really is more like 46W or is it really not as efficient as the 3500+ chip is at 1.8Ghz? Can I take this to loosely mean the 3000+ chip uses 30% more energy or creates 30% more heat at its stock 1.8Ghz speed than the 3500+ chip does at that 1.8Ghz speed.
Would this mean there is a tremendous advantage to using the 3500+ chip at 1.8Ghz?

Thanks

yeha · Post by **yeha** » Thu Mar 10, 2005 9:44 pm

What would the result have been had you used an actual socket 939 90nm 3000+ 1.8Ghz cpu rather than the 3500+ clocked down to 1800Mhz?

it would have been the same, plus or minus say 2% (due to 'hot' transistors possibly used in the 3500+ due to speed binning). more likely it'd be exactly the same.

you have to start ignoring the tdp values!

it really doesn't have any relation to how much power these chips actually draw. amd said 67 watts because they thought that's the most a 90nm cpu they produced could draw - present or future. the present chips don't draw anywhere near that much. it's just how it is

Oliver · Post by **Oliver** » Thu Mar 10, 2005 9:53 pm

Well, then if I decide to go the AMD64 route, I will buy a 3000+ processor
since you think it will not be less efficient (heat producing/energy consuming) than the 3500+ clocked down to 1.8Mhz.

jmatucd · Post by **jmatucd** » Thu Mar 10, 2005 10:18 pm

I'm a bit confused after reading all this... So how much power peak does the Athlon 64 3000+ 90nm 939 consume? (or should I be asking for the realistic 'peak')

More importantly, would it be wise to buy a 90nm 939 3000+ instead of waiting for the 25/35 TDP Turion 754 series @ 3000+

yeha · Post by **yeha** » Thu Mar 10, 2005 10:24 pm

the 3000+ 90nm d-stepping (winchester) athlon64 cpu at 1.8ghz and 1.4 volts will draw, at its absolute maximum, between 30 and 35 watts. that is what every accurate experiment i have seen has told us.

the e-stepping is arriving soon, and amd is trying to phase all its lines over to it to make production cheaper - the turion, socket-939 athlon 64's and opterons will all eventually be e-stepping chips. they have several improvements over the d-stepping winchester chips, such as an improved memory controller, support for sse3 instructions and other low-level changes. whether you want to wait for it to arrive (april is the earliest you'll be able to purchase one most likely) is a matter of personal preference.

i'm waiting for e-stepping chips to show up, but that's just me

jmatucd · Post by **jmatucd** » Thu Mar 10, 2005 10:39 pm

thank you for the complete response

I understand there will be a performance bump with the e-stepping chips (sse3 and the memory controller), but will the power consumption go down as a result or (as the sse3 leads me to believe) the increased performance will be coupled with an increase in power consumption (that isn't marginal).

Currently I'm torn between a 939 90nm winnie 3000+ (current), waiting for the turion on the 754 (which will limit the upgrade path), and waiting for the e-stepping versions of the 939. It comes down to cost--will the e version cost 10 dollars more or 50? heh, details, details, they always get you in the end

*edit*:: I'll wait for the e-stepping series. Thanks for the response and info

Mats · Post by **Mats** » Thu Mar 10, 2005 11:22 pm

jmatucd wrote:It comes down to cost--will the e version cost 10 dollars more or 50?

You won't see any big differences in price. Stores will probably sell them for a little bit higher ($10-20) but AMD will list them at the same prices. They put different prices for speed and type(mobile/desktop/server...) but not for stepping or socket (exept 462 because it's fundamentally different). Look at their prices. They've got the same prices for 130 nm S754 and 90 nm S939.

yeha · Post by **yeha** » Sat Mar 12, 2005 10:13 am

well over on xtremesystems drcrawn says his 252 opteron (e-stepping) runs much cooler than his 130nm fx-53. according to amd's tdp ratings, the 252 should be putting out more power (92 watts for 252 versus 89? watts for fx-53).

as the fx-53 is a 130nm chip and is most likely chewing up 80 watts or so, the fact that the 252 runs cooler at a higher speed (+200mhz) is encouraging! not to mention very confusing given amd's tdp figures. how annoying.

halcyon · Post by **halcyon** » Sun Mar 13, 2005 11:33 am

AMD's TDP figures have always been a MAX for that whole range of CPUs using that codename in their roadmap.

This included higher clocked (and perhaps higher core voltage) as of yet announced CPUs.

So, not all of the CPUs draw that much power or give out that much heat. It is the maximum TDP for the whole RANGE of CPUs (the fastest one being nearest to that max tdp).

Mats · Post by **Mats** » Sun Mar 13, 2005 5:16 pm

yeha wrote:well over on xtremesystems drcrawn says his 252 opteron (e-stepping) runs much cooler than his 130nm fx-53. according to amd's tdp ratings, the 252 should be putting out more power (92 watts for 252 versus 89? watts for fx-53).

as the fx-53 is a 130nm chip and is most likely chewing up 80 watts or so, the fact that the 252 runs cooler at a higher speed (+200mhz) is encouraging! not to mention very confusing given amd's tdp figures. how annoying.

TDP for FX-55 is 104 W.

yeha · Post by **yeha** » Sun Mar 13, 2005 7:01 pm

halcyon wrote:AMD's TDP figures have always been a MAX for that whole range of CPUs using that codename in their roadmap.

yes, however amd did a strange move and specified a separate tdp for the e-stepping 252 opteron (92 watts or so) and a lower tdp for the 250 and under e-stepping opterons (85 watts). this broke with tradition - they are all running the same core, but the most powerful chip in the line was for some reason given a higer tdp, which led many to believe that the 252 e-stepping chip actually would be drawing 92 watts at stock. otherwise, why wouldn't have amd kept up with their 'one tdp value per chip family' practice?

luckily, the 252 doesn't seem to eat up that much power. so it was in the end just another extremely confusing move by amd.