[REQ] Overclocking - CPU(AMD Phenom II x4 945) + MOBO(ASUS M4A79T Deluxe)

[trents]

Thanks!

but wait a second... that means the timings I had set originally in my CPU-Z screen shots were at 6.666 so if i change the timings to 9.999 wouldn't that not work? I guess I'm still confused with how setting the ram timings to 9-9-9-24-34 would help

Not sure what you are referring to when with "6.666 and 9.999". That doesn't look like a memory frequency, a memory timing or a memory divider.

"Timings" or "latencies" refer to pauses in the flow of data as it is being handled by the memory. Think of it as the flow of a cattle being herded through a stock yard and passing through a series of gates on the way to a railroad car. There are a series of digital gates that are closed for a time interval and then opened. The lower the timing or latency number value, the smaller the time is that the gate is closed. Maybe all the bovine data bits have to be present and accounted for before the gate is opened. I don't understand it either beyond a certain level so my homespun explanation may not be technically correct. If the gate is opened to fast some of the bovine data bits get excluded before they have reached the gate and the heard moves on without them and corruption has occurred.

RAM frequency, on the other hand refers to how wide the the path is to the gates. The wider it is the less crowded the data heard is and the faster it can get to the gate. But because more data cows are now present at the gate it will take longer to count them all. This is why as frequency increases latencies must be relaxed.

 

[trents]

"Dividers" refers to standard set points for memory frequency in bios. It may be expressed as an actual frequency or a ratio like 1:2, 2:3, 3:4, or 4:5. You can see this in the FSBRAM line in the CPU-z memory tab. CPU-z in it's recent versions has kind of messed this up so in the JEDEC columns of the SPD tab the frequencies are non-standard dividers but you can still see principle nonetheless and the timings still approximate well.

 

[5n4p]

Not sure what you are referring to when with "6.666 and 9.999". That doesn't look like a memory frequency, a memory timing or a memory divider.

"Timings" or "latencies" refer to pauses in the flow of data as it is being handled by the memory. Think of it as the flow of a cattle being herded through a stock yard and passing through a series of gates on the way to a railroad car. There are a series of digital gates that are closed for a time interval and then opened. The lower the timing or latency number value, the smaller the time is that the gate is closed. Maybe all the bovine data bits have to be present and accounted for before the gate is opened. I don't understand it either beyond a certain level so my homespun explanation may not be technically correct. If the gate is opened to fast some of the bovine data bits get excluded before they have reached the gate and the heard moves on without them and corruption has occurred.

RAM frequency, on the other hand refers to how wide the the path is to the gates. The wider it is the less crowded the data heard is and the faster it can get to the gate. But because more data cows are now present at the gate it will take longer to count them all. This is why as frequency increases latencies must be relaxed.

That made sense to me - trust me your paragraphs are a lot more user friendly than anything I've been reading haha.

To update you:
3.9ghz (stable thus far)
54c-59c steady under prime 95 blend

.... I think this may be a stopping point?

Says 62c max temp under the specs

 

[trents]

Is that 54-59c a core temp?

Yeah, you are probably pretty close to the "wall" for that CPU under conventional cooling methods. You've done well.

 

[5n4p]

Is that 54-59c a core temp?

yes ... I'm assuming that's too hot?
What are your recommendations when it comes to voltages?
Specifically cpu-nb and cpu... any suggestions of safe ranges? Obviously it varies... but curious as to what you suggest.

 

[trents]

No, that core temp is not necessarily too hot. It's not hot enough to damage the CPU and if it's stable it's fine but you probably are at the limit temp wise. But I would make sure things are good and stable with your current overclock settings by running the Prime95 blend test for at least two hours.

If things are stable I wouldn't necessarily mess with voltages and frequencies but let's have those three CPU tab pictures once again to check things out with where you presently are in your overclock.

 

[5n4p]

No, that core temp is not necessarily too hot. It's not hot enough to damage the CPU and if it's stable it's fine but you prbably are at the limit temp wise. But I would make sure things are good and stable with your current overclock settings by running the Prime95 blend test for at least two hours.

If things are stable I wouldn't necessarily mess with voltages and frequencies but let's have those three CPU tab pictures once again to check things out with where you presently are in your overclock.

This ram wont allow for a 4.0ghz OC.... lol at least not with this processor.
The math just doesn't add up for it

To hit 4ghz OC on this processor with a 15x multiplier I need at least a 267 FSB freq.
This puts the DRAM freq at 1068 = 5.34*267 = 1425.78

I'm assuming there is no way around this without replacing the RAM.

I am now a sad panda, but I feel like I have learned a lot today thanks to all of you!
This whole experience is greatly appreciated.

 

[trents]

I'd like to see you get your ram up to the full 1333 mhz it's rated for. And there's a way to do it you probably haven't thought about or were aware of. You see, even though your processor has a multiplier that is locked in the upward direction it is not locked in the downward direction. If you lowered the multiplier a little (even .5x) you can continue to increase the FSB in order to get the ram up to 1333 mhz.

So let's see some pics of CPU-z tabs and see what that might look like.

 

[5n4p]

I'd like to see you get your ram up to the full 1333 mhz it's rated for. And there's a way to do it you probably haven't thought about or were aware of. You see, even though your processor has a multiplier that is locked in the upward direction it is not locked in the downward direction. If you lowered the multiplier a little (even .5x) you can continue to increase the FSB in order to get the ram up to 1333 mhz.

So let's see some pics of CPU-z tabs and see what that might look like.

But it's not really worth sacrificing CPU power for it though =/

or ... is it?

 

[trents]

But don't forget, as you increase the FSB the CPU speeds up. So, as you begin increasing the FSB the CPU will speed up again. You want to get the best balance between CPU speed and memory speed.

 

[5n4p]

But don't forget, as you increase the FSB the CPU speeds up. So, as you begin increasing the FSB the CPU will speed up again. You want to get the best balance between CPU speed and memory speed.

now you're getting complicated...hehe I like it.
Even though the FSB does increase the CPU speed - it takes a bigger hit from the multiplier =/ so you're just sacrificing CPU speed for mem speed at that point.

or my math may be off - I'll run some calculations real quick.

 

[trents]

You are correct in your thinking but you would have to do the math to see how it actually works out. There is also the distinct possibility that if in order not to lose CPU speed you could increase the FSB to preserve that element but overclock the ram slightly. Usually, you can overclock ram at least a little especially if you increase the ram voltage say from 1.5 to 1.55 or even relax the timings a bit. And you have Corsair ram which usually overclocks well.

 

[5n4p]

You are correct in your thinking but you would have to do the math to see how it actually works out. There is also the distinct possibility that if in order not to lose CPU speed you could increase the FSB to preserve that element but overclock the ram slightly. Usually, you can overclock ram at least a little especially if you increase the ram voltage say from 1.5 to 1.55 or even relax the timings a bit. And you have Corsair ram which usually overclocks well.

so the DRAM Frequency which is calculated by x4 the FSB is what you want me to get to 1333mhz with optimal CPU speed?

 

[trents]

Yeah, if I understand what you are saying I think that's correct, or even a little higher than 1333 mhz is what I'm saying. That ram may well do 1400 mhz if you overvolt it a bit and if you need it to in order to get the CPU speed back where you want it if you lower the CPU multiplier a bit.

 

[I.M.O.G.]

I can simplify, or at least give an example, having done the math.

If you stick with the 1000 divider on ram, you want to try to hit a 266 fsb. That would mean 3.99ghz on the cpu tho (if you use 15x cpu multiplier), if it can do that. This would get your memory close to the rated clocks of 1333mhz, with about the same cpu clock.

Or you could drop the cpu multi to 14, then you could probably pull off an fsb of 278. This would mean at the 1000 divider, your ram would be at 1392, slightly over what its rated for, but it would probably run fine there.

There are other combinations. Just a couple examples for what could get your ram back up around its rated speed, now that you know the cpu is pretty happy at 3.9ghz and the current voltage.

I am not sure if you understand this yet, but what matter for cpu speed is multiplier times fsb. It doesn't matter if cpu multiplier is bigger or smaller, just what the final number is when combined with the fsb.

 

[5n4p]

I can simplify, or at least give an example, having done the math.

If you stick with the 1000 divider on ram, you want to try to hit a 266 fsb. That would mean 3.99ghz on the cpu tho (if you use 15x cpu multiplier), if it can do that. This would get your memory close to the rated clocks of 1333mhz, with about the same cpu clock.

Or you could drop the cpu multi to 14, then you could probably pull off an fsb of 278. This would mean at the 1000 divider, your ram would be at 1392, slightly over what its rated for, but it would probably run fine there.

There are other combinations. Just a couple examples for what could get your ram back up around its rated speed, now that you know the cpu is pretty happy at 3.9ghz and the current voltage.

I am not sure if you understand this yet, but what matter for cpu speed is multiplier times fsb. It doesn't matter if cpu multiplier is bigger or smaller, just what the final number is when combined with the fsb.

This is what I ended with... I'm assuming you all are saying I can do better..? I feel like something isn't adding up... I ran through some math and I don't see how you can get a better Dram FREQ than this with a cpu speed above 3.8

with your example of:
278 FSB - 14 multiplier = 3.892

Your divider cant be 1,000 - my BIOS only allows 1112 - 1483 - 1853 - 2224 (x4,x5,x6,x7 multipliers for the RAM divider)

With a 1112 divider
5.56*278 = 1545.68

cannot do

Unless I'm seriously mistaken - I'm ready for another learning curve haha

Given: I could just get new RAM lol but I don't see how it's possible with the current RAM
(I'm sure I can push it like a couple mhz more but it seems the gain is so minor now)

Correct me if I'm wrong.

I love you guys.

 

[trents]

Try a divider of 5x and FSB of 267. That should give you a CPU speed of 4.05 ghz and a ram speed of 1335 mhz. Don't know if it would be stable at the current CPU voltage, however, and you may need better cooling to pull that off if you need to increase the CPU voltage, that is.

 

[5n4p]

Try a divider of 5x and FSB of 267. That should give you a CPU speed of 4.05 ghz and a ram speed of 1335 mhz. Don't know if it would be stable at the current CPU voltage, however, and you may need better cooling to pull that off if you need to increase the CPU voltage, that is.

and.... typing it out i just found my blooper.... lol
but a 5x of a 267 FSB = 1424mhz?
Do you ever increase the DRAM Voltage so your RAM can hit higher frequencies? above 1.7?
and is a processor safe above 1.6v if temps are still ok?

 

[trents]

I don't know. I get all confused with trying to figure out the math. But you could try what I suggested (along with the extra DRAM voltage) and see what happens. If you can get into Windows and look at CPU-z we would know for sure.

5x267-1335.

267x15=4050.

Maybe my method is wrong.

 

[5n4p]

I don't know. I get all confused with trying to figure out the math. But you could try what I suggested (along with the extra DRAM voltage) and see what happens. If you can get into Windows and look at CPU-z we would know for sure.

5x267-1335.

267x15=4050.

Maybe my method is wrong.

oh then the dividers aren't modifiers... no idea what they are
under 267 we got - 1068 | 1424 | 1780 | 2136