A64 overclocking virgin. (3200+, 939)

[Snoochems]

Hi all.

As the thread title implies, I have had no experience with A64's. I only got it last week. I've upgraded from an AthlonXP 2500+, that was able to go up to 2.35GHz on air. So, I want to see what I can pull off with these 90nm CPU's.

Here are the system schpeks.
Athlon A64 3200+ socket 939 (stock cooling)
Gigabyte GA-K8NS-ULTRA (nforce 3 ultra)
1GB corsair TwinX DDR400 CAS2
Antec 480W PSU
X800XT-PE
Plenty-o-case cooling

So... I want this baby to fly faster than it already goes.
what do I do?

I've read the guide to overclocking an A64 that everyone I ask seems to link me to. (see http://www.dugu9tweaks.net/guides/a64oc/index.html#setup).

I think I know what to do, but have a few queries, and things I didn't understand fully from that guide.

1. What version of ClockGen should I use for this procedure? I went to the homepage, and no socket 939 boards are listed as 'supported'. Hmmm... I heard the gigabyte Easytune one I have been mucking around with is crap. This right?

2. How do I tell if my PCI is locked or not?

3. For stock cooling, what is the average/expected overclock for one of these 939 CPUs.

4. Why is it my CAS2 rated RAM is unstable at CAS latency 2 timings? In my old AthlonXP system, it was stable. Now it has to be on 2.5 in order to be 100% stable. (it would give me BSOD occasionally with CAS2...)

5. What effect does the HyperTranport Bus have on system performance? Any? My max is supposed to be 1000MHz. I assume it is safe to go under that speed, but not (too much) over.

This is my plan as it stands...
Change the LDT to 4x. (5x being stock)
I have DDR400 RAM. So, is it an idea to say, put the "MaxMem Clock" to 166, and then increase the HTT?
The way I understand the guide, if I set that "MaxMem Clock" to 166Mhz, I don't actually limit the RAM speed to 166MHz do I? It just means that the desired memory speed sits at 5/6 of the HTT speed.

Right?

If I am correct, that 166MHz thing is very misleading. I initially assumed that it would mean that my RAM speed could not go over 166MHz.

Anyways, if, say, I manage to then increase the HTT to 250MHz, my CPU will then run at 2.5GHz (10 being the stock multiplier). Right? And if I change the CPU multiplier to 9, I will get a CPU speed of 2.25GHz. And my RAM will run at 208MHz.

CPU speed = HTT x Cpu Multiplier
=9 x 250
= 2250MHz

HyperTransport = 4x250MHz
=1000MHz


Desired Ram Speed = 5/6*HTT
=208.33MHz

But, from the guide, I’m told that this is not the case. The ‘divider’ will need to be 11 to get close to as possible to the desired RAM speed. Right?

2250/11 = 204.54MHz

Which I know is 100% stable for my RAM.

Now am I right to go ahead with this? What voltages may i need to increase?

Thanks in advance.

 

[Tim]

Desired Ram Speed = 5/6*HTT
=208.33MHz

If you run 10x CPU multiplier and 166MHz RAM max setting the RAM will run with a divider of 12:

2000/12=166.66

If you then overclock the baseclock to 225, the CPU will run at 2.25 and the RAM at 2250/12= 187. 2.4GHz is very realistic for a Winchester with a 2.4GHz clock you will get 200MHz RAM speed. With a 250MHz baseclock you will reach a CPU speed of 2.5GHz and 208.33MHz memory - 2.5GHz is possible with many Winchester cores but it is not a sure thing (you might need to up the voltage a notch).

If you drop the CPU multiplier to 9 the RAM multiplier will drop to 11 (if the max 166MHz setting is used), with a baseclock of 244 you will get a 2.2GHz CPU clock and 200MHz memory. With a baseclock of 254 you will reach 2286MHz CPU speed and a memory speed of 207.8MHz.

Edit:
I don't think you need to raise anything other than you core voltage, and only if you go for the 2.4+ GHz overclock. I would not raise the voltage beyond 1.5V or maybe 1.55V the 1.65V that some does not seem very safe.

 

[trinibwoy]

This might not be the best board for overclocking advice. Check out ocforums or amdforums.

 

[PatrickL]

There is a guide for that here but don't know if you will like it as it is in french.

 

[T2k]

FYI: I've updated my primary work PC to the same mobo w/ 3500+ "Winchester" A64 (512KB-L2, the new 90nm SoI, 1.4Vcore) and currently it's running 10x 240MHz FSB (faster than a 3800+) with no problems with +0.1V only.

Home system - check sig - running on similar FSB (238 maybe, can't recall) with 11x multiplier but purely on water. 8)

PS: I'd replace your memory. I use DDR500 Corsairs and yet I give them +1V (home +2V, otherwise it's acting weird) anyway, as well as for the HT. GB tends to undercut when it comes to voltage.

 

[Moffell]

If we are continuing with the overclock talk, a friend of mine just built three 90 nm AMD 64 boxes. One with a 3000, two with 3200's and they all hit 2.4 Ghz with stock cooling and the standard heatsink. Not bad at all IMO.

Is there anyone who knows when AMD will bring Strained silicon on their 90nm line? They used it on FX-55 so it shouldn't be that far off.

 

[Mize]

If we are continuing with the overclock talk, a friend of mine just built three 90 nm AMD 64 boxes. One with a 3000, two with 3200's and they all hit 2.4 Ghz with stock cooling and the standard heatsink. Not bad at all IMO.

Is there anyone who knows when AMD will bring Strained silicon on their 90nm line? They used it on FX-55 so it shouldn't be that far off.

Stock cooling?! Wow. Winchesters be bad. Maybe I don't need that 3200+ after all. You are talking about the lame HS and wimpy fan that AMD provides, right?

 

[Snoochems]

Cool. Thanks ppl. Currently i'm up to 2.33GHz. Stock cooling

 

[T2k]

If we are continuing with the overclock talk, a friend of mine just built three 90 nm AMD 64 boxes. One with a 3000, two with 3200's and they all hit 2.4 Ghz with stock cooling and the standard heatsink. Not bad at all IMO.

Is there anyone who knows when AMD will bring Strained silicon on their 90nm line? They used it on FX-55 so it shouldn't be that far off.

That's actually a bad crap, so I really hope they never gonna do that here.

They did it for FX-55 solely because they couldn't make it on 90nm.

 

[T2k]

My primary work PC also running with stock cooling.

 

[Moffell]

Is there anyone who knows when AMD will bring Strained silicon on their 90nm line? They used it on FX-55 so it shouldn't be that far off.

That's actually a bad crap, so I really hope they never gonna do that here.

They did it for FX-55 solely because they couldn't make it on 90nm.

The facts simply doesn't add up for that conclusion IMO.

First; AMD doesn't (didn't) have ready phase masks for a 1 Mb 90 nm.

Second; FX-55 is high end and very low volume. Since the FX is basically the same as the Opteron it would have to go through a lot tougher validation trials on 90 nm than the regular A64 before release. Releasing 10 000 cpu's on 130 nm with SS is what, 200 wafers? Slightly more than a trial run and to me that looks like they're gearing up for full time SS production.

Third; SS isn't a matter of reconfiguring the transistor layout AFAIK. You get the process right for SS-wafers and go.

Fourth; SS added 200 Mhz with relative ease with better overclockability to boot. No temperature increases running away nor higher voltages or more leakage so why would switching from 130 nm to 90 nm all of a sudden nullify all the good stuff? Why would AMD and Intel bother with it if it sucked.

Bah, not so many fact after all mostly speculation and observation. I guess SS-cpu's spotted in the wild this far can come down to Prescott=bad( but with a ton more transistors) and FX-55=good.

He who lives shall see.