Passive cooling for undervolted/clocked AXP-M and 90 nm A64
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Passive cooling for undervolted/clocked AXP-M and 90 nm A64
Unfortunately I couldn't find many recent topics in this forum about passive cooling solutions... So let's start here. I need two passive cooling solutions, one for a heavily undervolted and underclocked socket A Athlon XP-M 2500+ (target: 500 MHz @ 1.1V) and another one for an Athlon 64 3000+ (Winchester, socket 939) which its target is 800 Mhz @ 0.8V.
I'm curious what kind of results people have with different heatsinks.
Thermalright has some nice heatsinks (also confirmed by a SPCR review) for socket 939; the XP-120 and XP-90. But I don't have a clue if these are suitable for passive cooling. Same goes for socket A; the Thermalright SP-97 looks nice... but...
http://thermalright.com/
Any experiences, suggestions, comments? They'll be very welcome! Thanks in advance!
I'm curious what kind of results people have with different heatsinks.
Thermalright has some nice heatsinks (also confirmed by a SPCR review) for socket 939; the XP-120 and XP-90. But I don't have a clue if these are suitable for passive cooling. Same goes for socket A; the Thermalright SP-97 looks nice... but...
http://thermalright.com/
Any experiences, suggestions, comments? They'll be very welcome! Thanks in advance!
I have a XP-M 2600 running at 5x200=1GHz and the temps while folding are:
VCore 1: +1.14 V
temp1: +31°C (case)
temp2: +40.5°C (CPU)
This one is cooled with the Zalman 6000-Cu in the Abit NF7-S motherboard. Motherboard underreports the temps by about 7 degrees or so.
The case is Antec SLK-3700AMB, great airflow that allows me to run the CPU fanless.
VCore 1: +1.14 V
temp1: +31°C (case)
temp2: +40.5°C (CPU)
This one is cooled with the Zalman 6000-Cu in the Abit NF7-S motherboard. Motherboard underreports the temps by about 7 degrees or so.
The case is Antec SLK-3700AMB, great airflow that allows me to run the CPU fanless.
Well, that Thermalright ALX-800 heatsink used fanlessly on an Athlon XP-M 600 MHz @ 1.6V gets too hot... Couldn't test anything else because the motherboard (ASUS A7N8X-E) I used to test doesn't support lower clockspeed and voltage. I can imagine that going from 600 MHz @ 1.6V to 500 MHz @ 1.1V makes a huge difference though...
-
- SPCR Reviewer
- Posts: 2696
- Joined: Thu Dec 11, 2003 9:53 pm
- Location: Scarsdale, NY
- Contact:
The difference is MASSIVE.Prahella wrote:Well, that Thermalright ALX-800 heatsink used fanlessly on an Athlon XP-M 600 MHz @ 1.6V gets too hot... Couldn't test anything else because the motherboard (ASUS A7N8X-E) I used to test doesn't support lower clockspeed and voltage. I can imagine that going from 600 MHz @ 1.6V to 500 MHz @ 1.1V makes a huge difference though...
Would be great! In that case I can reuse that heatsink...Edward Ng wrote:The difference is MASSIVE.Prahella wrote:Well, that Thermalright ALX-800 heatsink used fanlessly on an Athlon XP-M 600 MHz @ 1.6V gets too hot... Couldn't test anything else because the motherboard (ASUS A7N8X-E) I used to test doesn't support lower clockspeed and voltage. I can imagine that going from 600 MHz @ 1.6V to 500 MHz @ 1.1V makes a huge difference though...
Hi, if you're going for passive CPU cooling I would suggest considering the ones designed for it, like Scythe NCU-2000 Fanless CPU Cooler and there is another but I can't remember what it was called It's P4/Athlon 64 fitting. People have used the Zalman 6000 series sinks on Athlons passively.
Hope this helps, Seb
Hope this helps, Seb
Hey thanks for the suggestions!
Anyway, I'm afraid that the Scythe NCU-2000 is not the heatsink for me. For now the heatsink will sit in a big case, but in the future I might want to change the case, and that one will be (a lot?) smaller... Don't know about the Zalman 6000 series, but would it beat the Thermalright XP-90?
Anyway, I'm afraid that the Scythe NCU-2000 is not the heatsink for me. For now the heatsink will sit in a big case, but in the future I might want to change the case, and that one will be (a lot?) smaller... Don't know about the Zalman 6000 series, but would it beat the Thermalright XP-90?
Strange results... Cutting the frequency in half should pretty much make it output half as much heat. Reducing the voltage to half of the original voltage will make for 1/4 of the heat ouput before the voltage reduction, since heat ouput depends on the square of the voltage increase/decrease. I'm quite sure you know this already. Even if it's in theory, it should work pretty well in practice too... Well, let's just take my own XP-M 2600+ as an example:Edward Ng wrote:600 at 1.6 is 35.7 MDP while 500 at 1.1 is 15.9 MDP, according to CPU Heat.
Stock: 1933 MHz and 1.45V, producing 47W of heat according to specsheets.
Underclocked to 500MHz and 1.1V:
500/1933.333... ~= 0.259
This tells us that the CPU theoretically outputs only about 26% of the original heat. Now, let's see what the voltage reduction does:
(1.1/1.45)^2 ~= 0.576
And this tells us that the voltage reduction cuts power output down to just 58% of what it was before the reduction.
Now, all we have to do is multiply the original power output with those two figures:
47W * 0.259 * 0.576 ~= 7W
This sounds more reasonable!
I don't really intend this to look like a lecture, since I'm well aware that most of you already know this. Also, maybe I'm not 100% correct either. There may be something that makes reality differ quite a bit from theory that I'm not aware of.
Anyway, I might add that I wasn't able to run my mobile passively at 1GHz and 1.1V with an SLK-900A and pretty much no airflow. It would run pretty fine, but the heat would just keep building up and after about 30 minutes of Prime95 the it had slowly caused the CPU to reach over 60C. I suspect that the CPU would be much better cooled by a large aluminium heatsink (instead of copper like the SLK-900A) with more space between the fins.
I don't know. Isn't your calculation based on linear formulas? The reality may be different... exponential?Mikael wrote:I don't really intend this to look like a lecture, since I'm well aware that most of you already know this. Also, maybe I'm not 100% correct either. There may be something that makes reality differ quite a bit from theory that I'm not aware of.
Looking at Tom's Hardware Guides latest CPU test, one can see that an 800MHz A64 would at least match (and often be faster than) a 1200MHz Tualatin and lay waste to a PIII-733. I know, not very scientific, but it's true.Prahella wrote:No problem, I'm interested as well.VERiON wrote:sorry fo hijacking your thread, but i think it fits here
Ed - What about a preformance of a 90 nm Athlon 64 3000+ 800 MHz @ 0.8V. Can you compare it to PIII/733@133FSB or Tualatin1.2@100FSB?
The theory for transistors tells us that increasing frequency gives linear increase in power usage. It also tells us that increased voltage affects power ouput by the square of the voltage increase (= non-linear). In other words:Prahella wrote:I don't know. Isn't your calculation based on linear formulas? The reality may be different... exponential?
Twice the frequency = Twice the thermal power
Twice the voltage = Four times the thermal power
As said, this won't hold 100% in reality. There are many factors that affect these things. Thermals should atleast resemble theory somewhat good, though. Anyway, 15W for an AthlonXP @ 500MHz sounds way too much. An underclocked PIII Coppermine could probably come close to or match that figure (which shouldn't really be possible)...
Sounds good...Mikael wrote:Looking at Tom's Hardware Guides latest CPU test, one can see that an 800MHz A64 would at least match (and often be faster than) a 1200MHz Tualatin and lay waste to a PIII-733. I know, not very scientific, but it's true.Prahella wrote:No problem, I'm interested as well.VERiON wrote:sorry fo hijacking your thread, but i think it fits here
Ed - What about a preformance of a 90 nm Athlon 64 3000+ 800 MHz @ 0.8V. Can you compare it to PIII/733@133FSB or Tualatin1.2@100FSB?
I don't see a reason why I shouldn't believe you.. Anyway those other factors include things like current leakage, etc.?Mikael wrote: The theory for transistors tells us that increasing frequency gives linear increase in power usage. It also tells us that increased voltage affects power ouput by the square of the voltage increase (= non-linear). In other words:
Twice the frequency = Twice the thermal power
Twice the voltage = Four times the thermal power
As said, this won't hold 100% in reality. There are many factors that affect these things. Thermals should atleast resemble theory somewhat good, though. Anyway, 15W for an AthlonXP @ 500MHz sounds way too much. An underclocked PIII Coppermine could probably come close to or match that figure (which shouldn't really be possible)...
Two questions:Mikael wrote: Anyway, I might add that I wasn't able to run my mobile passively at 1GHz and 1.1V with an SLK-900A and pretty much no airflow. It would run pretty fine, but the heat would just keep building up and after about 30 minutes of Prime95 the it had slowly caused the CPU to reach over 60C. I suspect that the CPU would be much better cooled by a large aluminium heatsink (instead of copper like the SLK-900A) with more space between the fins.
1. What kind of board did you use?
2. Did you ever tried 500/600 MHz?
You shouldn't believe everything you see on the net! On a more serious note, yeah, with "other factors" I thought about things like current leakage. This would however just make the calculations look even better, since the reduced current leakage at lower frequencies would make the CPU output even less heat than what my numbers show at the moment!Prahella wrote: I don't see a reason why I shouldn't believe you.. Anyway those other factors include things like current leakage, etc.?
1. I used an Abit NF7-S 2.0.Prahella wrote: Two questions:
1. What kind of board did you use?
2. Did you ever tried 500/600 MHz?
2. Yes, I actually did, but with somewhat different hardware. This was back when I had my low-voltage (1.5V) 1700+ and an Alpha PAL8045T heatsink (aluminium/copper combination). I tried 600MHz at 1.1V and it worked out pretty well. I don't think that I had my low speed case fans connected, but I'm not exactly sure about that... Anyway, the CPU reached a maximum of 45C during full load, if I remember correctly. The Alpha heatink seemed to do a pretty good job at passive cooling. I didn't run the computer very long at these settings, since it was only just a quick test to see if it was possible. And it was.
Where did You found A64@800Mhz test?Mikael wrote:Looking at Tom's Hardware Guides latest CPU test, one can see that an 800MHz A64 would at least match (and often be faster than) a 1200MHz Tualatin and lay waste to a PIII-733. I know, not very scientific, but it's true.
I've serched THG.com and found nothing in:
All CPU Charts Part 1:
http://www.tomshardware.com/cpu/20041220/index.html
and the latest
All CPU Charts Part 2:
http://www.tomshardware.com/cpu/20041221/index.html
Oooops! I wasn't very clear on that... They didn't test an 800MHz Athlon64. What I did was to extract the results from the 1.8GHz 3000+ by multiplying the results from THG with 0.4444..... Since scaling isn't completely linear, the real results from an 800MHz A64 would likely be somewhat better than what you get with this method. It should still be quite accurate, especially for applications not depending on graphicscard performance (i.e. games). I'll get back to you if I find a real test of a low frequency Athlon64!VERiON wrote:Where did You found A64@800Mhz test?
I've serched THG.com and found nothing in:
All CPU Charts Part 1:
http://www.tomshardware.com/cpu/20041220/index.html
and the latest
All CPU Charts Part 2:
http://www.tomshardware.com/cpu/20041221/index.html
I did some passive cooling tests with my Winchester
(look here: http://forums.silentpcreview.com/viewto ... c&start=30
in the middle)
It's a Winchester 3000+ passivly cooled by a Zalman 7000CU with these temperatures under full load (prime):
1.6GHz @ 1.15v: ~ 63 °C (needs 14.2 watt)
1.4GHz @ 1.05v: ~ 55 °C (needs 10.9 watt)
0.8GHz @ 0.95v: idle near 32 °C (2-3 watt)
There might be some fluctuations between different CPUs (of the same type), so it could be possible that other Winchesters need more power.
But my CPU is not a great overclocker (stable till 2070 mhz at 1.4v).
If you find a better overclocker you can get a lower voltage at a given speed, so the wattage could be even lower.
What I wanted to say with this post:
There is no need to go down to 800 MHz with the Winchester, it can easily be cooled passivly e.g. at 1400 MHz when undervolted.
(look here: http://forums.silentpcreview.com/viewto ... c&start=30
in the middle)
It's a Winchester 3000+ passivly cooled by a Zalman 7000CU with these temperatures under full load (prime):
1.6GHz @ 1.15v: ~ 63 °C (needs 14.2 watt)
1.4GHz @ 1.05v: ~ 55 °C (needs 10.9 watt)
0.8GHz @ 0.95v: idle near 32 °C (2-3 watt)
There might be some fluctuations between different CPUs (of the same type), so it could be possible that other Winchesters need more power.
But my CPU is not a great overclocker (stable till 2070 mhz at 1.4v).
If you find a better overclocker you can get a lower voltage at a given speed, so the wattage could be even lower.
What I wanted to say with this post:
There is no need to go down to 800 MHz with the Winchester, it can easily be cooled passivly e.g. at 1400 MHz when undervolted.
I concur - the wattage posted for the XP-M2500+ @1.1V, 500MHZ is way too high. I've seen figures like 16W floating around for 1G @1.1V, and "CPU Power" reports 17.1W for 1033MHz @1.1V, for a 2500+ (not mobile). I'd expect the wattage to be even lower for the mobile (or is that not the case, just that it can get to these specs whereas the inferior wafer selection non mobile can't?). @500MHz, CPU Power reports 8.2W.
For ~ zero external airflow cooling, I'd expect the Zalman's, at least the older flower models to outperform the Thermalrights - wish there was a review on that, and not with the typical 5V fan _directly mounted to the heatsink_. I believe the thermalrights (all models - they all have closely packed fins, at least all the ones I've seen) will perform drastically better if you can _direct_ the airflow so that it passes through the fins - say with a duct attached to where the fan would normally be. With no vectorization (if that's a word of the airflow, cooling probably differs quite a bit from case to case even with identical net airflow passing through the case.
For ~ zero external airflow cooling, I'd expect the Zalman's, at least the older flower models to outperform the Thermalrights - wish there was a review on that, and not with the typical 5V fan _directly mounted to the heatsink_. I believe the thermalrights (all models - they all have closely packed fins, at least all the ones I've seen) will perform drastically better if you can _direct_ the airflow so that it passes through the fins - say with a duct attached to where the fan would normally be. With no vectorization (if that's a word of the airflow, cooling probably differs quite a bit from case to case even with identical net airflow passing through the case.