Overclocking Sandbox: Tbred B DLT3C 1700+ and Beyond

[hitechjb1]

Tbred B 1700+ DLT3C 2.61 GHz

HSF: SLK-800U, Vantec Tornado.

It is stable to finish PCMark 2002. Will try with 2V.

Highest Sandra CPU: 9758 MIPS/3921 FLOPS, model rating 3306, PR rating 3785.
The CPU integer computing power is equivalent to P4 running at 3.34 GHz.

 

[EspElement]

you got water cooling or is that temp not loaded?

 

[hitechjb1]

you got water cooling or is that temp not loaded?

It is still air cooled. The temp is lightly loaded at 2.61 GHz after running PCMark 2002. I am working on to get it stable running prime above 2.6 GHz.

 

[hitechjb1]

Testing a Fortron 530W PSU

I have been thinking about upgrading to an Antec TP 550 or a Fortron 530W PSU, hoping to get better voltage stability for oc. Many ppl have highly recommend the Fortron 350W and 530W. Some even say the Fortron 350W is enough to do most job and can deliver 454W (I do not agree on this number).

I usually like to look at spec and testing, rather than just based on word of mouth. The Antec TP550 or TC550 has better specification, +-3% line tolerance instead of 5% of Fortron, and also higher current in the 12V line of 24A compared to 18A of Fortron. In additional Antec has better fan control and fan location to vent the CPU.

Anyway, for its cheaper price ($40-50 less), I got a new Fortron 530W PSU and tested on the Tbred B 1700+ setup, with a SLK-800U + TT SFII.

Here is some measurement of the voltage lines at idle and running prime95, @1.92V, 2544 MHz.

1. Voltage lines at idle

The 5V line dropped to 4.703 V, and flutuated between 4.703 - 4.977 V.

The 12V line was at 12.42 V, this was too high and might be too much for hard drives, CDRW, DVD, ...

For the Antec TP430, 12V line is around 11.93 V, +- 0.04 V.

See the two figures in next post for snapshot of voltages, ....

2. Voltage lines with Prime95

I saw the 5V fluctuated to a high of 4.977 V from 4.703 V, which is 0.274V. This fluctuation is more than the 5% line tolerence of the PSU. If taking potential measuring uncertainty of 5% of the ASUS probe, that fluctuation would still be 0.274/5 * 0.95 = 5.2% (almost in line with the spec tolerance).

Also I saw the Vcore bounds around between 1.85 to 2.0 V (in the snapshot it bounds between 1.888 V and 2.0 V), also this is outside of the AMD Vcore tolerence of +- 50 mV.

See the three figures in next post for snapshot of voltages, ....

To my surprise and many's recommendatation, it is disappointed to see the Fortron 530W performs like this. The 12V too high (12.42 V), and % of tolerence is more than 5% of spec, or right at the margin if 5% measurement uncertainty is taken into account.

My system does not have excessive hardware (many may even think a Fortron 350W would do it): 1 mb, 1 CPU, 1 TT SFII, 4 80mm fan, 1 DVD, 1 HD, 1 video card.

The Antec TP430W that I have been using has smaller tolerance on the various lines and within spec, for powering exactly the same system.

Due to the voltage fluctuation of the Fortron 530, I got BSOD even running at lower CPU clock, compared to that when using the Antec TP430.

I don't know whether it is a bad PSU or others have looked at the Fortron 530W carefully. I hope I have not done anything wrong with the setup ???

I am going to return this PSU and ask for full refund. It seems there is no other choice except an Antec TP/TC 550.


Link to discussions, measuremnts, comparisons, more about Vcore regulator and PSU:
Comparing FS 530 and TP 550.

CPU Power Model

 

[hitechjb1]

Tbred B 1700+ DLT3C 2.54 GHz voltage idle (Fortron 530W PSU testing)

 

[hitechjb1]

Tbred B 1700+ DLT3C 2.54 GHz voltage under Prime95 (Fortron 530W PSU testing)

 

[Kdog154]

hitechjb1- Did you mod the CPU any or stick it in @ stock w/out touching anything?

Is the SLK-800 worth the extra money? Do you mod the HS in any way? I read that some people do something to it to get rid of some of the little bubbles.


For a Serial ATA MB, can normal ATA/100 HD's work w/ it or you have to buy the new Serial ATA HD's?

 

[EspElement]

hey.... does it do prime all night?
whats thre temp then?

 

[hitechjb1]

hitechjb1- Did you mod the CPU any or stick it in @ stock w/out touching anything?

Is the SLK-800 worth the extra money? Do you mod the HS in any way? I read that some people do something to it to get rid of some of the little bubbles.


For a Serial ATA MB, can normal ATA/100 HD's work w/ it or you have to buy the new Serial ATA HD's?

I did not mod the CPU at all, neither any bridge or any wiring trick for altering the default Vcore or high/low multipliers. Unless there is really no other alternative, I am always concern about a human mistake would kill the CPU or the Vcore regulator. E.g. doing a wiring trick, just a simple wrong pin connection can short the Vcc to Vss and kill the motherboad (Vcore generator).

SK7 is around $20, SLK-800 is around $35. The different between a SK7 and a SLK-800U is only about 0.02 C/W in thermal resistance at high CFM (~5500 rpm 80mm fan). 0.23 C/W compared to 0.25 C/W for the SLK-800U and SK7 respectively. This equates to only about 2-3 C at high degree of overclocking (100-150 W CPU thermal power, or around 2.5 GHz @ 2V). That 2-3 C would become immaterial by using better case air flow and high CFM fan.

So SK7 is much more economical. That 2-3 C difference would tanslate to around 20-30 MHz at exterme oc on air. !!!!

I did not do any mod on the SLK-800U HS I used.

ATA serial drive is different from the traditional ATA IDE drive w/ parallel interface. ATA serial drives are new and expensive for the performance gain.

A RAID 0 using traditional ATA IDE drives (esp with the 8MB cache drives) is very fast, and I have been using RAID 0. One bad thing about the recent nforce2 mb are that they don't support RAID (also no RAID 1, 0+1) for the traditional ATA 100 drives (unless you add the parallel/serial interface), and serial ATA drives are not yet common and expensive.

There are some interace one can get to hook up ATA IDE drive (w/ parallel interface) to the serial drive interface on the mb. But those interface costs about $20 a piece which I consider also expensive. The NF7-S mb comes with one such interface, and you need to get another one, so you can hook up two ATA IDE parallel drives to the onboard serial interface to run RAID.

 

[hitechjb1]

I found that besides sufficient wattage reserve which has been paid much attention and talked about, tight line regulation (1% PC P&C, vs 3% fo Antec TP, vs 5% for the rest) is important for system and Vcore stability for the last 50-100 MHz of CPU overclocking.

Good PSU for overclockin besides wattage reserve, with the Vcore regulator should be able to handle the huge CPU current changes and loading:

- CPU active current ~ 70+ A (at high oc), the largest component in a system
(to get an idea: fan ~ 0.2 - 0.7A, HD ~1.3A (WD 120 GB 8MB), CDRW, DVD ~ 1A, ...)
- CPU is not a pure resistive load, rather a non-linear R, C, L deivce (harder to regulate than pure resistive load due to transients)
- CPU requires fast response time regulator and PSU. Some regulators has more than 2 phsase regulator, e.g. ABIT usually has 3 phrases, some even has 6 phrases
- current fluctuation between different load level is huge

Tighter rail has an advantage for CPU overclocking stability for the last 100 MHz. I learned that from overclocking the Tbred B 1700+ DLT3C, and it should be of relevance to other CPU overclocking too.

I must point out that for that 100 MHz to get a new PSU is very expensive for price/performance, and may not justify to chase for that performance in most cases.

It has been pointed out (by Since87, ... in a PSU thread Comparing FS 530 and TP 550 and posts about CPU Power Model) that CPU huge load change (min to max) and current surge can occur in a short time of 3 CPU cycles. And I think such requires fast transient response of the switching regulator and PSU to regulate the Vcore line (+- 50 mV max from AMD spec) and the underlying PSU line to within small fluctuation for system and CPU stability.


This is a link to PSU review saying that Antec TP480 is especially suited to exterme overclockers, maybe of some interest to CPU owners, ...

Quote:

"In the power-use class, the 450- to 520-Watt league, the THG lab technicians only arrived at one recommendation: the Antec True 480P, which produced the most power in the test. This power supply also delivers stable voltages on all lines and comes with a good equipment package. The Antec True 480P is the power supply we recommend for high demands, and is especially suited to extreme overclockers."

Reference: June 9, 2003 High Power For Power Users - 13 Power Supplies In The Spotlight

 

[hitechjb1]

Overview: Some thoughts on choosing PSU for overclocking recent motherboard, CPU, GPU, ... (page 8)


What size PSU do I need?

Let's do some estimates. An estimate is better than just guessing, if it is not correct, we can refine it. By doing calculation, I find out the main demand is on current, and not on the total power that most people seem to emphasize on and pay attention to when looking at PSU:

E.g. Tbred B 1700+ DLT3C, 1.9V 2.5 GHz
CPU current = 71 A max, power = 135 W max
At rated, CPU current = 29.9/32.9 A, power = 44.9/49.4 W (typ/max)
It is the largest component in a system.

fan ~ 0.2 - 0.7 A, 12 V
HD ~ 1.3 A (WD 120 GB 8 MB), 12 V
CDRW, DVD ~ 1 A, 12 V

I tried but cannot find spec for the following, so just making some guess (they are not big number compare to CPU, so end result should not be much different if guess is off).

mothboard = 40 W (includes NB, SB, ports, floppy, ...)
9700/9800 pro video card = 45 W (GPU load) + 15 W (card)
memory module per dimm = 3A @ 3.3V ~ 10 W

So assuming system as follow (assume air cool, adjust according for extreme cooling):
1 motherboard = 40 W (10 W 3.3 V, 30 W 5 V)
1 overclocked CPU at 1.9 V 2.5 GHz = 135 W
2 memory modules = 2 x 10 W = 20 W (3.3 V)
1 AGP video card (such as 9700/9800 pro) = 60 W (15 W on 3.3 V, 30-45 W on either 5V or 12 V)
1 CPU fan = 8 W (12 V)
4 case fans = 4 x 4 W = 16 W (12 V)
2 HD = 2 x 19.5 W = 39 W (0.8 A 5 V, and 1.3 A 12 V)
1 DVD = 12 W (1 A 12 V)
1 CDRW = 18 W (1.2 A 5 V, 1 A 12 V)

Total = 348 W

You may adjust these numbers according to your system, and level of overclocking. For good design, you may add 20% tolerence, or you may agrue some PSU's have hidden juice to cover this.

The above assumes overclocking as specifed above. If you don't overclock, then the CPU component would be smaller; ~ 85 W can be taken off based on my estimate. Or if you plan to overclock Vcore to 2.0V or above, you have to add more.

CPU active power and current are the major compoent due to overclocking. One thing to be careful is about the current on either the 5V or 12V PSU rail. This is the major current number you have to go after at the spec of the PSU whether they can source enough current for stable Vcore for CPU plus other peripherals.

Besides looking for sufficient power reserve, one thing important is to look at the current rating for sourcing CPU current and whether that line has tight regulation (1% vs 3% vs 5%) to proviode stable line for Vcore stability, and hence oc stability. For AMD mb, some mb's use 5V and some use 12V to generate Vcore.

1. If mb uses 5V to generate Vcore (e.g. A7N8X), the 71A will draw about 1.25 x 71 x 1.9 / 5 = 33.7 A on the 5V line (assume regulator efficiency 80%). Then

5V_current_rating > 33.7 + current_for_other_5V_components
Also assuming video card uses 30W on 5V
current_for_other_5V_components ~ (30 + 30 + 8 + 6)/ 5 = 14.8 A

Total 5V current requirement = 33.7 + 14.8 A = 48.5 A (if both Vcore and video card use 5V)

So if both CPU and high end video card use 5V, it is very overloading the PSU !!!! That is why it is highly recommended to get a motherboard that derives Vcore from 12V. Best scenario is motherboard uses 12V for Vcore, and video card uses 5V for GPU (I think the 9700 uses 5V for GPU).

2. If mb uses 12V to generate Vcore (e.g. NF7-S, 8RDA3+), the 71A will draw about 1.25 x 71 x 1.9 / 12 = 14.3 A on the 12V line (assume regulator efficiency 80%). Then

12V_current_rating > 14.3 + current_for_other_12V_components
Also assume video card uses 30W on 12V
current_for_other_12V_components ~ (30 + 8 + 16 + 31 + 12 + 12) / 12 = 109 / 12 = 9.1 A
Total 12V current requirement = 14.3 + 9.1 = 23.4 A

0308 finding - Video cards (9700/9800) mainly use 5 V for GPU:

12V_current_rating > 14.3 + current_for_other_12V_components
Also assume video card uses 30W on 12V
current_for_other_12V_components ~ (8 + 16 + 31 + 12 + 12) / 12 = 79 / 12 = 6.6 A
Total 12V current requirement = 14.3 + 6.6 = 20.9 A

Add: If video card does not use 12V, then the 12V current would still be around 21-24 A, depending on the number of HD's.
E.g. I estimate a system that uses 12V for Vcore, with a Tbred B 1700+ DLT3C oc to 2.5 GHz 1.9V, 4 HD's, 1 DVD, 1 CDRW, 4 case fans, 1 CPU fan, 9700 video card, .. would require about 23.5 A on the 12 V line.

From the above calculation, it looks like as far as wattage, one can agrue any good 350 W power supply will do the job. But due to the high current requirment of the CPU and GPU, the main demand and requirement are shifted to current requirement, and not just on wattage.

For new nforce2 motherboard such as NF7-S and 8RDA3+, they both use the 12 V to generate Vcore, which I think is better for more stable Vcore. This put the main demand on PSU for overclocking.

Let's look at some popular PSU on 12V line:

Fortron 530 W = 18 A (line regulation 5%)
Antec TP 430 = 20 A (line regulation 3%)
Antec TP 480 = 22 A (line regulation 3%)
Antec TP 550 = 24 A (line regulation 3%)

I am not going into the reserved power, hidden juice, ... debate of which PSU to get. You can use these numbers as guideline and make your own decision.



Some useful design points:

1. A P4 1.8 overclocked to 2.4 GHz 1.675V, 4 HD's, without highend video using 12V, ... (~ 20.8 A) link.

Originally posted by hitechjb1 (06-21-2003 11:53 PM)
...
I try to estimate what 12V current you need:

P4 1.8a overclocked to 2.4 GHz @ 1.675 V
from spec 1.8 GHz 1.75 V Icc = 50.6 A

At 2.4 GHz, 1.75 V
Icc = 50.6 (2.4) / 1.8 = 67.5 A
power = 1.75 (67.5) = 118.1 W

At 1.675 V
Icc = 67.5 (1.675/1.75) = 64.6 A
power = 1.675 (64.6) = 108.2 W

The 12 V current for Icc assuming 80% efficiency on Vcore regulator
I_12V_1.75 = 1.25 x 67.5 x 1.75 / 12 = 12.3 A if Vcore = 1.75 V

I_12V_1.675 = 1.25 x 64.6 x 1.675 / 12 = 11.3 A if Vcore = 1.675 V

4 HD = 1.3 A x 4 = 5.2 A (based on WD 120 GB 8MB, 12 V)
4 fan = 0.4 A x 4 = 1.6 A (typical fan 0.3 - 0.7 A, 12 V)
CPU fan = 0.7 A (12 V)
1 DVD = 1 A (12 V)
1 CDRW = 1 A (12 V)

So 12 V current
12V_current = 12.3 + 5.2 + 1.6 + 0.7 + 1 + 1 = 21.8 A (Vcore 1.75 V)
or
12V_current = 11.3 + 5.2 + 1.6 + 0.7 + 1 + 1 = 20.8 A (Vcore 1.675 V)


Let's assume motherboard, video card do not drawn from the 12 V line at all. The power for them on the 3.3 V and 5 V is not significant.

Others (5 V or 3.3 V)
Mothboard = 40 W
2 memory = 2 W
1 video card = 30 W (7500 AIW not using 12 V ?)

So the main thing to look at is the 12 V current which is the most important number to look at. You will need a PSU that can supply about 20.8 A on the 12 V line. This is my estimate, pls take a look at it, comment and check the calculation are welcome.

In another thread, I have done similar calculations for powering recent systems. I found that we should pay attention to the 12 V current rating instead of just pure wattage in the PSU. I did some calculation in another thread, in terms of power 350 W should be enough. But as far as 12 V current, the story is very different, many PSU may not have sufficient 12 V current.

So please check the 12 V current rating of your PSU, and see whether it can supply the above current estimate. Does the above current estimate reflect your system reality.

If you have any question, pls post.

2. P4 2.8C overclocked to 3.7+ GHz, 9700 pro, 2 SATA HD's, fans, ..., 12V current ~ 20 A link.

Originally posted by hitechjb1 (07-20-2003 09:37 AM)
Let's look at the 12V current. From the Intel P4 data for a 2.8C at rated 1.525V, Icc is between 23 to 55.9 A. I estimated when overclocking to 3700 MHz at rated Vcore 1.525V, the current would be somewhere betwen 66.5-73.9 A. For P4 and any CPU, around max overclocking, raising Vcore (than necessary) may make things worst. If raising Vcore to 1.6V, at 3700 MHz, the max current would be somewhere between 69.7-77.5 A. And it would translate into 11.6-12.9 A on the 12V line of the PSU (assuming 80% efficiency).

Assumiing each HD takes about 2 A on the 12 V (I don't have the detail spec for the SATA HD). The few fans take 1-2 A on the 12 V line. The optical drive takes 1 A. So the max total 12 V current would be around 20 A. The TP 480 PSU is rated 22 A on the 12 V.

3. P4 2.4C overclocked to 3.4 GHz 1.6 V, 1 HD, fans, ..., 12V current ~ 18 A (2 HD's would be 20 A)
link.

Originally posted by hitechjb1 (07-22-2003 09:27 AM)
The major demand on PSU for your system is the 12V current due to CPU, HD, CDROM, fans, ... from your list. Let's do some estimate.

P4 2.4C at rated 1.525 V, the current is between 23 - 52.4 A (from data sheet). At your target frequency of 3.4 GHz, I estimate the max current would be somewhere between 64.7 - 74.2 A at 1.525 V. In case you have to raise the Vcore to 1.6 V to acheive the 3.4 GHz, I estimate the max current woudl be between 70.0 - 81.7 A. At 1.6 V Vcore, assuming a 80% efficiency Vcore regulator, it would draw a max current around 11.7 - 13.6 A on the 12 V line of the PSU.

Other 12 V current are: 2 A per HD (average), 1.5 A for the fans, 1 A for CDROM drive. So the total max 12 V current is estimated to be around 18.1 A.

Other the current and power for the 3.3 V and 5 V should not be a concern.

The 12 V current estimate of 18.1 A is already higher than the rated current of most 350 W solid PSU. If you need to add a HD in the near future, it would be about 1.5 - 2 A more per drive, and bring the 12 V current requirement to 20 A. I estimate for a typical overclocking P4 and AMD system that use 12 V for Vcore, the 12 V current requirement is at least 20-24 A (from a variety of cases).

Even if taking some potential hidden reserve of a PSU into account, due to overclocking of the CPU, the 12 V line current rating of most 350 W PSU is working at the edge of, or slightly below your system demand. If you already have the PSU or on an absolute budget for a 350 W, then try it first and it may just limit your CPU overclocking by 200-300 MHz. If you plan for some growth for your system, I would get a solid PSU with 20-24 A on the 12 V, possibly with smaller line regulation.
...

 

[NoxioN]

i think they just trying to sell more PSUs the TP330 and up should be fine.

 

[hitechjb1]

i think they just trying to sell more PSUs the TP330 and up should be fine.

But apart from these reviews and commerical. If you look at the few posts earlier, I like to bring out the important points for extreme overclocking for CPU:

1. CPU at high level of oc draws more than twice the nominal rated current and almost three times the rated power
2. tight rail and Vcore help overclock stability
3. CPU is a non-linear device and its current can surge within a short period of time, hence requires fast respond Vcore regulator and PSU

 

[hitechjb1]

Impact of higher ambient temperature on CPU clock frequency

Summer is almost here in the northern hemisphere, and many of us are finding that our room and system ambient temperature are 5-10C higher than that in the best air-overclocking season winter. Here I attempt to find out what is the impact of such ambient temperature increase on CPU max frequency, so we can have an excuse/explanation that the CPU is not doing so well recently than used to be.

For air cooling, we usually keep the die below certain temperature say, 45, 50 or 55 C (for whatever probe we use w/ certain accuracy). The system and CPU stability are indicators whether max CPU overclocking has been reached. I found that for the DLT3C TB B, beyond 50 C 1.9-1.95 V, it is harder and harder to get the CPU stable due to leakage current and heat which prevent CPU from running faster to meet the timing.

General speaking, HSF and high CFM are most important for air cooling. On top of that, if system ambient tempeature is higher, and since CPU die temp is controlled under certain temp (for stability and speed menitoned above), that will reduce the active power to a lower number and hence the resulting max CPU frequency.

So when system ambient temp is higher by x degree C, keeping the same HSF and max temp, the active power has to be lowered by

del_active_power = - x/K

where K is the HS thermal resistance. E.g. for SLK-800U at high 5500 rpm, K = 0.23 C/W. If ambient temp is increase by 2 C, del_active_power = - 2/0.23 = -8.7 W

P = C V^2 f (the active power component that varies w/ freq)

Also I estimate for every 100 mV Vcore under 40 C, del f/del V ~ 100-140 MHz / 100 mV. That is 100 mV increase gets you about 100-140 MHz increase in CPU clock, or vice versa.

So P = a V^3 (where a is a constant)
or dP / P = 3 dV / V

E.g. For Tbred B 1700+ DLT3C, using SLK-800U at high 5500 rpm, max CPU thermal power at 2.5 GHz 1.9V ~ 137 W.

The -8.7 W is equal to 6.35% lower in active power, and I estimated to be about 6.35%/3 = 2.1% lower in CPU clock (df/f = (1/3)dP/P). A 2 C rise in ambient system temp will lower the CPU clock by 2.1%. So at 2.5 GHz overclocking, it is about 53 MHz lower in CPU clock.

A 5 C rise in ambient system temp will lower the max active power by about -5/0.23 = - 21.7 W, which is 15.8% lower in active power. And it equates to about 15.8%/3 = 5.3% lower in CPU clock. A 5 C rise in ambient system temp will lower the CPU clock by 5.3%. So at 2.5 GHz overclocking, it is about 132 MHz lower in CPU clock.

Does this estimate seem to agree with your experience in higher room and system temperature?

 

[Gautam]

The numbers are perfect, but you seem to be implying that higher temperatures will actually force people to lower their overclocks by as much as afformentioned. With a case like yours, i.e. a DLT3C on air, in which the cooling is the only restriction in clock speed, around 100mhz is to be expected. However processors that overclock this well are rare. For people with lower batch processors, or people with better cooling, temperature is not the restricting factor in stability, rises in ambient temperature may mean nothing to their overclock. You are considering what effects ambient temperature would have supposing that one distinct core temperature would have to be maintained, however in real world experiences, one can often afford to let their core temperatures simply get higher along with ambient temperatures. Before I went to water, I was running my 1700+ at over 55°C, with no stability problems. Just thought I'd point this out.

Btw, I'd like to commend you on your absolutely enlightening information. Your posts are always very interesting to read; it's very interesting to see the laws of physics being applied to overclocking. Very helpful, too.

 

[Audioaficionado]

My DLT3C will seem to run fine fully loaded with FAH3 in the low 50s but will eventually lock up in the peak heat evening hours between 1800 to 1900. After I lowered the vcore and fsb a little, the CPU stays in the mid to high 40s and hasn't locked up since. So heat seems to be my main limiting factor.

 

[Gautam]

Yeah, heat obviously plays a very very large role as a limiting factor in overclocking, especially with the new DLT3C's; with these procs, its the only factor, considering that they fly with such low core voltages.

 

[hitechjb1]

Gautam,
Thanks for posting these questions, you have good insights into the problems, .... For overclockers, the CPU's are usually at max overclocking and max die temperature for stable system at full load, hence the assumption. And let me try to go a little bit further and answer them, ...


Comparing different HSF's and impact of ambient temperature on max CPU frequency (HSF 0.30 C/W)

To do analysis, we need certain model and assumptions. The assumption is that
- At any Vcore and max die temp (system stable), the CPU is clocked to its max level at same work load (e.g. running Prime95).
That is:
- It can be a not so good CPU, e.g. a 1700+ DLT3C topped out at 2.3 GHz
- Or other CPU such as a Barton or a P4 2.4C, ...
- It can be a different HSF (top line or no so good, different thermal resistance), ...

I think it can be extended to water cooling, or other extreme cooling, .... but I don't want to calculate for them now, since I don't have experience with them yet.

I have done the calculation using a SLK-800U (few post earlier). Now, for example, using a 1700+ DLT3C and a Vantec aeroflow (with TMD fan) or Volcano 7+, let's see what happens.

From the heat sink page, Vantec aeroflow with fan at 5300 rpm has a thermal resistance of K_aeroflow = 0.30 C/W. BTW, a Volcano 7+ has similar thermal resistance of 0.3 C/W.

So assuming setting a max temp (w/ system stable) for the die at 50 C (it can be 45 or 55 C), the ambient system temperature to be 25 C (e.g. winter)

Thermal resistance K = (maxT - ambT) / P

max CPU thermal power Pmax_50C = (50 - 25) / 0.3 = 83.3 W

This means a CPU under such HSF cannot be clocked as high as one with a SLK-800U under 50 C, since the latter would be able to clock to dissipate (50-25)/0.23 = 108.7 W.

At max 83.3W using aeroflow/Volcano 7+, I estimate a Tbred B 1700+ DLT3C can be clocked to around 2200 - 2288 MHz (50C load, 25C ambient). It is estimated as follow:

P = C Vcore^2 f
f / Vcore ~ 100 - 140 MHz/100 mV (piecewise linear at higher temp 30-50 C)
where
P is CPU active power,
f is CPU frequency
C is the equivalence capacitor to model CPU switching for a given load (there are 10-100 millions cap charged/discharged per cycle depends on the CPU load)

f_aeroflow_50C_25C = 2500 (83.3/122.8)^(1/3) = 2500 (0.88) > 2200 MHz
f_aeroflow_50C_25C = 2500 (83.3/108.7)^(1/3) = 2500 (0.92) < 2288 MHz
(at 50C with 25C ambient, consuming 83.3 W, compared to 2500 MHz 1.9V consuming 108.7 to 122.8W)

Here we can see how the overclock result of two different HSF's, since the one with higher thermal resistance limit the max power to the CPU for a given max die temperature. A difference of 212-300 MHz between a top end and a medium good HSF.

Ambient temperature rise

With this setting, for a x degree C (e.g. 5 C) rise in ambient system temp, the drop (delP) in max thermal power in order to maintain max die under 50 C would be
delP = -x / K = - 5 / 0.3 = 16.7 W (for the aeroflow)

which is 16.7/83.3 = 20% of the max power at 25 C system ambient. As a result, the max CPU clock has to be lowered by 20%/3 = 6.7% which is about 2200 (6.7%) = 147 MHz. I.e. the new CPU clock with a 5 C increase in ambient system temp would have to be lowered by 6.7% from 2200-2288 MHz to around 2053-2141 MHz to maintain the max die temperature at 50 C.

 

[hitechjb1]

My DLT3C will seem to run fine fully loaded with FAH3 in the low 50s but will eventually lock up in the peak heat evening hours between 1800 to 1900. After I lowered the vcore and fsb a little, the CPU stays in the mid to high 40s and hasn't locked up since. So heat seems to be my main limiting factor.

What HSF are you using?

If you look at the calculation. The difference between a SLK-800 and a aeroflow/Volcano 7+ can be as much as 200-300 MHz with the same temperature setting for max die 50 C and system ambient of 25 C.

On top of that, during hot season, a 5 C increase in ambient will further cut the CPU clock down by another 6.7% or roughly 147 MHz.

 

[L337 M33P]

This is the thread to end all threads...

First and foremost, VERY NICE OC!

Hmm The XP1700+ is a very overclockable chip, but may I suggest a Peltier or Watercooling setup, since you are already approaching the limits of air cooling. You could probably get a few extra hundreds of Mhz out of that chip.

I will buy one of these from PClincs in the UK - they allow you to specify the stepping codes

Must remember this - JHIUB 0310 XPMW DLT3C write it down, write it down...