I have a question: Namely, I have a rated 400W PSU, and I heave OC the CPU to a 2.4 GHz and Vcore=1.55V, stable on air cooling. It can go to 2.5GHz
but it's not stable. Since the PSU is bundled with the tower, btw, it is from it-colors,
can the futher overclocking be achieved using another PSU or can the system be stable at 2.4 but with lower Vcore? In other words, can I get a stable system with less voltage and higher current? How is the Venice 3200+ S939,
current dependent?

Thanks for any reply!

While a low quality PSU could hold you back by failing to provide constantly stable current under load, there are more likely sources of unstabilities.

I'd like to know more about the settings you used for the 2.4 and 2.5GHz overclocks such as the HT link frequency, the HT link multiplier, the memory frequency, the memory divider and any relevant timings.

Same goes for your hardware, knowing exactly which motherboard and memory (please be exhaustive on that one, down to the model and revision along with the rated timings) you are working with will be of great help.

I'll be exhaustive as possible:

MB: MSI K8N Diamond, nFORCE 4, SLI, nforce voltage = 1.65V

Memory:2*512 DDR400 Dual-Channel A-Data CL(2.5-3-3-7, 1T), down to 166MHz, MEM voltage = 2.7V

CPU:Athlon 64 3200+, S939, Venice Core, Stepping E6 at 1.55V

FSB: 240, HTT: 4x, CPU Multiplier: 10x, CPU/MEM=1/12, The MOBO has no
options for memory divider, I guess it sets up automatically based on the cpu frequency or it is locked by the MOBO manufacturer to 1/12.

PCIex16=101MHz

MSI 6600GT TD 128E @ 550/1100 128MB GDDR3

Phoenix, AWARD Bios, ver.1.A the latest from MSI site, dating from april 2006

There should be settings to change those multipliers/dividers as I have a K8N Platinum and i know that i have changed them but i belive you have to change setting to full manual. you may need to change your ram to 2t also to increase stability for further overclocking.

Okay, your system becomes unstable at 2.5GHz because your memory starts running out of specs (2500MHz / 12 = 208MHz), try to lower your memory down to 133MHz, this will result in the CPU/15 memory divider being used, that way, your memory won't be holding you back until you exceed 3GHz.

An even better scenario would be to set your memory to 150MHz if it is an available option (it is a locked BIOS option for most motherboards but modded BIOS usually feature it), it would use the CPU/14 memory divider which will be good until you exceed 2.8GHz.

Don't forget to set your HT multiplier to 3x in order to keep the HT link frequency below 1GHz.

Now all you have to do is to overclock your CPU, if your system becomes unstable, it means that you took it as far as it physically could.

As far as memory is concerned, it has no impact on the system stability. I have tested it and it is very easy to overclock to 2.22 GHz, having in mind that memory operates at 222MHz, and 444MHz effective, so I think that it is not an issue till I am keeping to 450MHz margin, I wouldn't go further. At 2.5GHz I can run SuperPi 1.5, MemTest is ok, but fails on Prime 95. Also, when i got a stable overclock, at 2.4 GHz, the voltage must have to be set at 1.55V, lower than that and Prime 95 fails. Reminder: All the other voltages are as stated above.

I think, maybe I'm wrong, it is the power supply that limits me for a stable overclock to 2.5 or 2.6 GHz. I think that I should get a new PSU with higher current rating at voltage rails if I want to achieve that.

Here is another question: Is DDR memory a bottleneck for the CPU?

If it is, why trying to OC, when the CPU must wait for data to come in from HTT link and memory, being processed at 2.8GHz speed, and sent to an output device at speed which is halfe or even more 1/4 to that of the CPU.

Oh, I had no idea that you already tested your RAM's OC headroom so I assumed a worst case scenario where your memory would hardly have any tolerance, my bad for not specifically asking for it in my earlier post.

As for the PSU being the culprit, the only way for you to test your hypothesis it to replace it with a new one.

Don't worry about the impact of memory bandwidth, after all, a 2.8GHz FX57 is meant to run with dual-channel PC-3200, hinting that I/O is not a problem for the k8 uarch.

So, do you agree with me about the new PSU or there are other considerations to be taken into account?

Thanks!

Well, topping out a Venice core at 2.4GHz with 1.55V seems awefully low IMHO, what about your cooling system and temps under a sustained P95 load (Use Core Temp if possible).

Since nothing infirms nor confirms that your PSU is holding you back, I won't encourage nor discourage you to get a new one...

There are other considerations to be taken into account.
It is just luck.
I made 15 identical systems:
Asus A8N5X s939 NF4
3200+ Venice E6
2x512MB PQI CL 2.5-4-4-8
Asus 7600GS 512MB

All had different overclocking abilities from 2.1GHz the worst case and 2.8 the best. I also have very similar config and my CPU is able to perform SPi1M at 2.8 and Prime at 2.7Ghz stable. So, I guess it is luck

Temperatures are: The worst case scenario: At 28C room temperature:

Idle: 38C Full Load: 48C

So, to conclude, as gOJDO said, it's just luck with the chip, though I didn't got an answer how the cpu is current dependent. I'll lookup that issue in the technical reports and white papers.

Thanks to you all!!!

Actually, for a Venice core 3200+, thats not that bad off for an O/C. I have seen worse. There seem to be a lot of variations in the quality of the earlier Venice cores...

Sorry, I used now Core Temp and it says that CPU = 1.45V and Temp = 39C

In my last post I've used CoreCenter from MSI.

Random programs give different results.

Was that 39C with Core Temp your idle or load temp ? BTW, the 1.45V figure is your CPU stock Vcore.

As for the impact of Vcore on CPU stability, an higher current artificially increases the transient response of transistors, allowing them to operate at an higher frequency.

Here is an informative article on the subject.

an higher current artificially increases the transient response of transistors

How can a physical flow of current be artificiall???

If you have voltage and no current, what are you going to get???

Transient response of the transistors is fixed, i.e determined by the technology of production, you can't change it!

So, what is overclocking? Simple, telling the transistors to switch on/off more quicker and that is not done by increasing CPU voltage, but with adjusting the internal clock. To maintain the stability, transistors ask for more voltage and consequently more current at their inputs. They want more and more as you go higher on overclocking. When there is no power anymore, they are saturated and stability breaks.


You are right, 1.45V is the stock that I have adjusted in the bios, the default is 1.4V. I have also enabled override of the Cpu_Vid to gain 1.55V.

39C is idle, 48C is at load.

I read the article after posting my previous text. I was surprised how accurate was my explanation, though I think that the author made a mistake in the part: Putting it all together, where it says and I quote:

we reach a point at which the cycle time is less than the transient time. In other words, the cycles are so short that the voltage doesn't have time to reach the levels it should. Consider the image below.

And, When we have VCC+ (or the intended voltage) set to a higher value, it takes less time for our voltage to reach the actual VCC needed by our logic gates. Now, if we decrease the cycle time, our logic highs still come through as logic highs, thanks to the higher core voltage.

With the new VCC+, there is a new threshold voltage aplied to the gates, so they will want to reach full VCC+ and not the actual VCC. This not works in practise, since you are forcing another design technology that is not implemented by the logic gates.

But whatever, maybe I'm in a mistake about the above stated, but this article also proved that there is need for higher current (current source power supply).

Thank you about the article!

I misused the term "current", a simple brainfart on my part, I was refering to the core Voltage, not Amperage. Same goes for me using the term "artificial" to describe the effect of a higher Vcore on transistors when the cycle time is close to the transient time, it may not be the most adequate term but I thought that it was accurate enough.

As for more Vcore resulting in an "artificial" increase in transistors transient response, it is a simplistic but valid statement. Even though the transient response is fixed, a greater gap between Vss and Vcc will allow the transistors to reliably reach the threshold between logical 1's and 0's when operating at higher frequencies, this is clearly illustrated it the article that I linked to in my previous post.

I was not carefull in reading the article, my mistake, it is ok.

Heh, no problems, don't sweat it !

There is nothing wrong with questioning data and putting hypothesis foward for the sake of discussion.

If you're disappointed with your OC on that Venice and think that you could use a dual core, you may want to consider an entry level Manchester such as the X2 3800+, those have a lot of headroom, especially if you grab one that was fabbed in 2006.

I'm thinking about that! If the store from where I've bought my system is willing to make an xchange between my CPU and X2 3800+, of course with payment by my side to compensate with the difference between my used cpu and their's new x2 3800+ it'll be very, very nice!

I've never heard of any store doing that though, then again I never bothered trying, I just give my old hardware to close friends and family members anyway.

Be sure to check out This Thread, I posted how to find the datecode of a CPU, if given the choice between multiple CPUs, try to get the most recent one.