Memory Controller, or is it the Northbridge - someone correct me if I am wrong.
Just idling with an ambient of 22c (taken with DigiDOC5 outside of case) and case temps of 22c (yeah good cooling
I have these following temps for CPU, RAM, and Memory Controller. The temps for the RAM and memory controller are with DigiDOC5 thermal diodes on the heatsink of the memory controller and wedged between the substrate and the heatspreader on the RAM. I am not sure why one RIMM is 2c cooler than the other. Maybe it's the diode. I will be playing with this a little bit more and I will move the diodes around. I'm also gonna try and take a temp measurement of the memory controller directly (stick the diode between heatsink so it touches the side of the processor).
My System
1.6a @ 2.4 1.7v
2x Samsung 256 PC800 RIMMS
450FSB (150 x3)
*IDLE*
CPU: 26c (two 120 Panaflo L1As right behind the CPU sucking air out)
Mobo: 22c
RIMM1: 29.3 (above CPU)
RIMM2: 31.6 (to the right of CPU)
Memory Controller: 34.0
Now, load her up with Prime and I get:
*AT LOAD*
CPU: 37c
Mobo: 22c
RIMM1: 32.8
RIMM2: 35.2
Memory Controller: 40.0
Looking at these temps I think the memory controller definately needs better cooling. That's 40c on the heatsink alone - not the core. The memory is a bit cooler, but with some cooling I could probably get it down under 30c with a 22c case/ambient temp and better cooling.
These are some preliminary temps. But they show items which must be considered when overclocking. I am just barely overclocking the ram and memory controller at 450 FSB. Maybe I'll jump out and do some testing at 540 FSB to see what it's like then.
This is helping me tons - let me know if it helps you guys any.
shadow
**************************************************
OK!!! 544FSB (136x4)
I pushed the diode for RIMM1 farther in, and had to move the memory controller diode a little bit. I put it back in place as well as possible.
*AT LOAD*
CPU: 36c
Mobo: 22c
RIMM1: 34.0c
RIMM2: 35.3c
Memory Controller: 38.6c
The ram temps seem the same, the memory controller a little less. It is running at 136 instead of 150. That could be it. I'll have to mess around with it some more and get it to touch the core instead of the heatsink.
See you later with another update.
************* UPDATE #3 **************
The ambient fluctuates in this next test, but I still think you can make some good inferences from the data. I'll wait for it to warm up till I take another set of measurements at 22.6c ambient.
These test use Sandra's Burn-in Wizard to loop the memory testing, I set it for 100 iterations and took measurements after 60-80 of the iterations. I left out the cpu, cuz it's not important to this discussion.
A hub? So, it's not a Northbridge? I'm talking about the chip with the passive heatsink to the right of the CPU. I'll start calling it a hub now instead of Memory Controller.
133FSB x 4 or 532mhz
ambient: 22.6c
mobo: 22c
RIMM1: 36.4
RIMM2: 36.2
Hub: 39.3
(aka memory controller)
150FSB x 3 or 450mhz
ambient: 21.1c
case: 21c
RIMM1: 34.7
RIMM2: 35.7
Hub: 39.6
The difference in the ambient temp can account for the lower mem temps, however I still want to see what it is with similar ambient temps.
The hub however is hotter - even with the lower ambient temps. This implies that the hub's heat production (and hence usage/clocking) is dependant on the FSB, not the memory speed. So, if you are having trouble in general with FSB speed, and moving to a x3 multiplier does not help, then your Hub might be suspect.
I am currently limited to 133FSB at the Quad (x4) memory multiplier. I tried 136 and after around 60 or so iterations with an ambient of ~22.6c it crashed sandra. So this would mean that my memory is the culprit, and this makes sense. I have 8piece 256meg Samsung. What I really need is the 16piece Samsung 256.
Will cooling the hub give me increases? I suspect, that even though the memory seems to be the culprit, I might still get something out of cooling the hub. Unfortunately, this will have to wait. Maybe until the summer, when school lets out.
I have the 600mhz DRCGs. I know there are many people out there who are having trouble overclocking their TH7-IIs and P4T-Es. These people get to the mid 120s with the x4 multiplier, and need to drop to the x3 multiplier to get to 133. There are several theories seeking to explain this. The first, and the one most people tend to agree with is the 400mhz DRCGs are holding people back. The second is that the memory itself is what is holding people back.
To muddy the waters even more, there are individuals with the 400mhz DRCGs who are overclocking the living daylights out of their systems. This would reinforce the second hypothesis. That it is poor memory which is holding people back.
Would 600mhz DRCGs help these unfortuante souls out? Yes I think they would, at least in part. We will find out soon enough, a member of the [H]forums, psyco, has received shipment of some 600mhz DRCGs and will begin shipping them out soon. We will see if it's the DRCGs or the memory.
I think it's a little bit of both, more on the memory however. But, of course, there are so many variable to consider. THe two biggest ones are cooling and power. I have great cooling (4x120mm L1As) and a great power supply (Enermax 431watt). Another variable might be the memory controller, who knows.
If you think the above discussion about DRCGs and RAM is a bunch of hogwash, reply and let me know. I hope, no matter what you think of my conclusions and thinking that you find this data at least somewhat useful in your overclocking pursuits.
enjoy,
shadow
Just idling with an ambient of 22c (taken with DigiDOC5 outside of case) and case temps of 22c (yeah good cooling
I have these following temps for CPU, RAM, and Memory Controller. The temps for the RAM and memory controller are with DigiDOC5 thermal diodes on the heatsink of the memory controller and wedged between the substrate and the heatspreader on the RAM. I am not sure why one RIMM is 2c cooler than the other. Maybe it's the diode. I will be playing with this a little bit more and I will move the diodes around. I'm also gonna try and take a temp measurement of the memory controller directly (stick the diode between heatsink so it touches the side of the processor).My System
1.6a @ 2.4 1.7v
2x Samsung 256 PC800 RIMMS
450FSB (150 x3)
*IDLE*
CPU: 26c (two 120 Panaflo L1As right behind the CPU sucking air out)
Mobo: 22c
RIMM1: 29.3 (above CPU)
RIMM2: 31.6 (to the right of CPU)
Memory Controller: 34.0
Now, load her up with Prime and I get:
*AT LOAD*
CPU: 37c
Mobo: 22c
RIMM1: 32.8
RIMM2: 35.2
Memory Controller: 40.0
Looking at these temps I think the memory controller definately needs better cooling. That's 40c on the heatsink alone - not the core. The memory is a bit cooler, but with some cooling I could probably get it down under 30c with a 22c case/ambient temp and better cooling.
These are some preliminary temps. But they show items which must be considered when overclocking. I am just barely overclocking the ram and memory controller at 450 FSB. Maybe I'll jump out and do some testing at 540 FSB to see what it's like then.
This is helping me tons - let me know if it helps you guys any.
shadow
**************************************************
OK!!! 544FSB (136x4)
I pushed the diode for RIMM1 farther in, and had to move the memory controller diode a little bit. I put it back in place as well as possible.
*AT LOAD*
CPU: 36c
Mobo: 22c
RIMM1: 34.0c
RIMM2: 35.3c
Memory Controller: 38.6c
The ram temps seem the same, the memory controller a little less. It is running at 136 instead of 150. That could be it. I'll have to mess around with it some more and get it to touch the core instead of the heatsink.
See you later with another update.
************* UPDATE #3 **************
The ambient fluctuates in this next test, but I still think you can make some good inferences from the data. I'll wait for it to warm up till I take another set of measurements at 22.6c ambient.
These test use Sandra's Burn-in Wizard to loop the memory testing, I set it for 100 iterations and took measurements after 60-80 of the iterations. I left out the cpu, cuz it's not important to this discussion.
A hub? So, it's not a Northbridge? I'm talking about the chip with the passive heatsink to the right of the CPU. I'll start calling it a hub now instead of Memory Controller.
133FSB x 4 or 532mhz
ambient: 22.6c
mobo: 22c
RIMM1: 36.4
RIMM2: 36.2
Hub: 39.3
(aka memory controller)
150FSB x 3 or 450mhz
ambient: 21.1c
case: 21c
RIMM1: 34.7
RIMM2: 35.7
Hub: 39.6
The difference in the ambient temp can account for the lower mem temps, however I still want to see what it is with similar ambient temps.
The hub however is hotter - even with the lower ambient temps. This implies that the hub's heat production (and hence usage/clocking) is dependant on the FSB, not the memory speed. So, if you are having trouble in general with FSB speed, and moving to a x3 multiplier does not help, then your Hub might be suspect.
I am currently limited to 133FSB at the Quad (x4) memory multiplier. I tried 136 and after around 60 or so iterations with an ambient of ~22.6c it crashed sandra. So this would mean that my memory is the culprit, and this makes sense. I have 8piece 256meg Samsung. What I really need is the 16piece Samsung 256.
Will cooling the hub give me increases? I suspect, that even though the memory seems to be the culprit, I might still get something out of cooling the hub. Unfortunately, this will have to wait. Maybe until the summer, when school lets out.
I have the 600mhz DRCGs. I know there are many people out there who are having trouble overclocking their TH7-IIs and P4T-Es. These people get to the mid 120s with the x4 multiplier, and need to drop to the x3 multiplier to get to 133. There are several theories seeking to explain this. The first, and the one most people tend to agree with is the 400mhz DRCGs are holding people back. The second is that the memory itself is what is holding people back.
To muddy the waters even more, there are individuals with the 400mhz DRCGs who are overclocking the living daylights out of their systems. This would reinforce the second hypothesis. That it is poor memory which is holding people back.
Would 600mhz DRCGs help these unfortuante souls out? Yes I think they would, at least in part. We will find out soon enough, a member of the [H]forums, psyco, has received shipment of some 600mhz DRCGs and will begin shipping them out soon. We will see if it's the DRCGs or the memory.
I think it's a little bit of both, more on the memory however. But, of course, there are so many variable to consider. THe two biggest ones are cooling and power. I have great cooling (4x120mm L1As) and a great power supply (Enermax 431watt). Another variable might be the memory controller, who knows.
If you think the above discussion about DRCGs and RAM is a bunch of hogwash, reply and let me know. I hope, no matter what you think of my conclusions and thinking that you find this data at least somewhat useful in your overclocking pursuits.
enjoy,
shadow
Last edited:
