GamePC:s article on A64 vs PM has been around for some days, but it didn't caught my eye until this morning and it is bugging me a bit. I don't know how to interpret the numbers they present as powerdraw. That they find a powerdelta of 24-34W, depending on clockspeed, between max CPU utilisation and idle seems about right. When I compare this with my own Newcastlecore A64 the numbers seem to match rather good.
Of course my own CPU is a 130nm Newcastle and they have tested a 90nm Winchester, so the numbers should not be very comparable. I got curious though, to see how large deltas I had when comparing the same "powerstates" (I lent the term, I know, but I couldn't find any better and I think you understand what I mean).
What got to me however was the large differences in delta between idle at stock and idle at [email protected]. They repot 7-10W, I get a powerdelta of 40W! This is such a large difference that I have to ask. Are these numbers something that can be reproduced on these boards where I know there are people who have a kill-a-watt? What is the powerdraw at 100% load at [email protected]?
So, CnQ has lost much of it's effect, if the numbers may be recreated. For users where every watt counts, like laptop users and powerjunkies, CnQ will still have some effect, but it is not such a "must-have" anymore....
Well, I guess that with such low powerconsumption and the Turion arriving soon, both we and AMD will have some interesting times ahead...
Has CnQ played out its role on Winchestercore CPU:s?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Has CnQ played out its role on Winchestercore CPU:s?
Last edited by Tobias on Mon Feb 14, 2005 5:42 pm, edited 1 time in total.
From the idle power consumption figures I'd say that Cool 'n' Quiet simply wasn't working. In otherwise comparable systems, I can't imagine any other reason for the large 40W disparity when the system was idle.
Perhaps they thought it would just work automatically if supported by the motherboard? The motherboard used, the Asus A8N-SLI Deluxe should in theory support CnQ but, as you mention, I wonder does it work with the bios used?
Anyone here have any experience of running this motherboard, I wonder?
Perhaps they thought it would just work automatically if supported by the motherboard? The motherboard used, the Asus A8N-SLI Deluxe should in theory support CnQ but, as you mention, I wonder does it work with the bios used?
Anyone here have any experience of running this motherboard, I wonder?
Let's do some calculation.
power drop due to voltage drop: 1.1^2 / 1.4^2 = 38%
power drop due to speed reduction: 1.0 / 1.8 = 44%
In total the power drop is 82%.
Assuming a 80% load being the max utilization (I didn't look at the review, just guessing) :
If 82% power drop is just 10W, then full power CPU would be drawing 15W and could be cooled with your regular Zalman NB47J.
A powerdelta of 40W would suggest a TDP of 61W which is more logical
Obviously something is wrong. (I hope it's not my calculations
)
power drop due to voltage drop: 1.1^2 / 1.4^2 = 38%
power drop due to speed reduction: 1.0 / 1.8 = 44%
In total the power drop is 82%.
Assuming a 80% load being the max utilization (I didn't look at the review, just guessing) :
If 82% power drop is just 10W, then full power CPU would be drawing 15W and could be cooled with your regular Zalman NB47J.
A powerdelta of 40W would suggest a TDP of 61W which is more logical
Obviously something is wrong. (I hope it's not my calculations
)Ok, went ahead and read the article. They've used Prime95 so that would invalidate my %80 TDP assumption.
However, it's obvious from their data that CnQ isn't enabled. In fact they explicitly state that clock throttling is disabled.
Plus I didn't see any mention of PSU efficiencies anywhere?
Interesting but flawed article IMO.
However, it's obvious from their data that CnQ isn't enabled. In fact they explicitly state that clock throttling is disabled.
Plus I didn't see any mention of PSU efficiencies anywhere?
Interesting but flawed article IMO.
Unfortunately you can't add the "power drops".power drop due to voltage drop: 1.1^2 / 1.4^2 = 38%
power drop due to speed reduction: 1.0 / 1.8 = 44%
In total the power drop is 82%.
Flip around your first calculations
1.4V -> 1.1V = 62% of original usage
1.8 Ghz -> 1.0 Ghz = 56% of original power usage
To combine these you must multiply them together.
1.8Ghz @ 1.4V -> 1.0Ghz @ 1.1V = .62 * .56 = 34.7% or original power usage
So there is only a 65% drop in power consumption.
According to AMD's A64 Power and Thermal Data Sheet (PDF), the 3000+ Winchester only has those two steps, min (1000 @ 1.1v) and max (1800 @ 1.4v). The 3200+ Winchester has one intermediate step (1800 @ 1.35v) between min and max, and the 3500+ has two intermediate steps (1800 @ 1.3v, 2000 @ 1.35v).as530 wrote:i have an asus a8n-sli deluxe and amd 3000+ winchester. the CnQ seems quite crude, idling at 1MHz and 1.075v, and under any stress, stepping up to the full 1.8MHz and 1.375v. i.e. it doesn't seem to be a smooth gradation.
Oh, I was obviously not clear enough. First some clarifications.
I have messured a 63W powerdelta with my Newcastle and compared to its TDP of 87W, the difference is 25W. Not anything conclusive, but in the same ballpark.
We all know that the TDP:s for A64 is inflated, the question is how much? If the difference is bigger on the 90nm-part (which it very well might) then the span of the powerusage shortens.
Further, it is possible that Winchesters idle at a lower usage compared to newcastles relative their powerdraw at full load.
I guess my question could be refrased "How large effects have other people seen with CnQ on winchester CPU:s?" I am wondering since what I have seen of my own A64 differs alot.
btw, I was wondering
Those numbers only apply to full load, right? We still do not know how much less power the CPU would use at idle [email protected] compared to full load at the same settings. The difference on my system is 5-6W when running Prime95.
I'm sorry but I really thought I was clear on this...Mariner wrote: In otherwise comparable systems, I can't imagine any other reason for the large 40W disparity when the system was idle.
...but I am not sure I was clear enough on the point that it is my 130nm CPU that has a delta of 40W between idle stock and idle CnQ.Tobias wrote: Of course my own CPU is a 130nm Newcastle and they have tested a 90nm Winchester, so the numbers should not be very comparable.
Not necessarily. If we first just compare the numbers at hand. GamePC report a total powerdelta between full load@stock and idle@CnQ of 37W. Compared to its TDP the difference is 30Wbucakb wrote: ... as I suspected. Still the review process there is flawed
I have messured a 63W powerdelta with my Newcastle and compared to its TDP of 87W, the difference is 25W. Not anything conclusive, but in the same ballpark.
We all know that the TDP:s for A64 is inflated, the question is how much? If the difference is bigger on the 90nm-part (which it very well might) then the span of the powerusage shortens.
Further, it is possible that Winchesters idle at a lower usage compared to newcastles relative their powerdraw at full load.
I guess my question could be refrased "How large effects have other people seen with CnQ on winchester CPU:s?" I am wondering since what I have seen of my own A64 differs alot.
btw, I was wondering
Blappo wrote: So there is only a 65% drop in power consumption
Those numbers only apply to full load, right? We still do not know how much less power the CPU would use at idle [email protected] compared to full load at the same settings. The difference on my system is 5-6W when running Prime95.
thanks Wraith! had no idea about this.According to AMD's A64 Power and Thermal Data Sheet (PDF), the 3000+ Winchester only has those two steps, min (1000 @ 1.1v) and max (1800 @ 1.4v). The 3200+ Winchester has one intermediate step (1800 @ 1.35v) between min and max, and the 3500+ has two intermediate steps (1800 @ 1.3v, 2000 @ 1.35v).