PC3200 memory won't run full speed on NForce3.

[eno]

Epox 8KDAJ NF3 250GB - motherboard

Corsair PC3200 TwinX 3200C2PT(2x1024) or Kingston HyperX PC4000(3x 512mb)

AMD64 3200+ (754)


Alright, I have always noticed with my Kingston HyperX PC4000 that when I saw the post screen it would display my memory running at 208mhz even though my CPU FSB was set at 250fsb. I started looking into it this last week and found that my stock 3200 speed settings that the memory would clock down to 166mhz. If I ran only one stick and had memory settings on auto it would run at 200mhz. I have tried both brands of memory and the board just does not run stable if I set the memory setting to 200mhz and then try to raise the FSB of the CPU. I have to put it to auto or 166 then raise the CPU but at the same time my memory just runs at low 208mhz speed and for PC4000 memory its a waste. I ordered up the Corsair 3 days ago and just got it from Newegg, if I put in both sticks 2x1gb then my memory shows at 166mhz, take out a stick and now 1 stick will run at 200. So I do not want to get a new board but this is just very upsetting that I spend money for overclockablity like a good Thermalright cooler, good fast memory, good case cooling and a board that is supposed to allow me to tweak and get some more performance out of it. I am planning on a new system in the next year which will not use this any of this hardware but I am now stuck between running this hardware and stock speeds for stablity or trying to get my full money's worth with a better chipset or board. I know that this board doesn't like Double Sided memory in pairs, only 1 stick but I am wondering if I got to so called "good overclocker" DFI Lan GF3 board if I would really get my memory to run at its rated speeds while at the same time maximize the CPU overclock with my good cooling. I really hate to spend more money on this 754 setup but am not happy that I can't get basically anything out of these parts that "should" be capable of more.

Do you think its the board or chipset not allowing the fine adjustments of FSB/CPU/Memory speeds??

Should I get the DFI GF3 250 board to allow me to open up the rest of the hardware I have?

Do you think it could run at least the 200mhz memory speeds while at the same time overclock the CPU to its 2.5-2.6mark? Just don't want to spend more money on system if I don't have to but then again I already spent money in other areas for higher performance like memory and cooling, I would like to use it or I should of just used stock cooling and cheaper memory. Thanks!!!!!!!!

 

[mechBgon]

I see you posted this in Motherboards at the same time. FYI, that's called "cross-posting" and is sort of a nuisance. Give your first thread a chance next time

(for people who are wondering, this thread was in GH and seems to have been moved here)

Edit: on the topic at hand, read your EPoX manual and see if you're using too many "sides" of RAM for it.

 

[eno]

I know what cross posting is and sorry if I accidently did that but where is my other post? I posted on another website for DFI boards but I don't see where I posted this on anandtech already? Yes I know its annouying if I did that but if I did I didn't mean to and I don't see where I did. Please explain.

 

[Peter]

It's for sure not a chipset issue, simply because the RAM controller is in the processor, not the chipset.

Besides, there is no "CPU FSB" for exactly that reason. What you think is the CPU FSB is the system reference clock. CPU core speed multiplies UP from there. RAM bus speed divides DOWN from CPU _core_ clock speed. What I'm trying to say is, your math is wrong, you're still living in northbridge centric world.

What's happening?

* You set the refclk to 250 MHz, CPU thus runs at 2750 MHz but thinks it's running at 2200 (because it assumes refclk=200).
* So for PC3200 mode, it chooses a divider of CPU/11 for the RAM, which would actually turn out to be 250 MHz.
* You setting the RAM target speed to 166 MHz makes it choose CPU/14, which actually turns out to be 196 MHz.

This is not an "issue", it is expected behaviour within _realistic_ operating parameters. In other words, RAM labelled "PC4000" is generally to be regarded as marketing bull, and your experience just confirms that.

 

[eno]

Originally posted by: Peter
In other words, RAM labelled "PC4000" is generally to be regarded as marketing bull, and your experience just confirms that.

Woah woah woah. I might not have a grasp on this area, thats why I am posting , but I know PC4000 isn't marketing bull when I was running it in my 865PE chipset, I was able to run the CPU/Memory at 1:1 without needing to use a divider. Thanks for your help.

 

[eno]

Originally posted by: Peter

* You set the refclk to 250 MHz, CPU thus runs at 2750 MHz but thinks it's running at 2200 (because it assumes refclk=200).
* So for PC3200 mode, it chooses a divider of CPU/11 for the RAM, which would actually turn out to be 250 MHz.
* You setting the RAM target speed to 166 MHz makes it choose CPU/14, which actually turns out to be 196 MHz.

Yes that makes sense, thats what I have dealing with today. I see that the Ram Target Speed is basically the divider. If you set it to 200, then your CPU is in a 1:1 setup? Setting the Ram Target Speed to 166 is basically making the memory run at 83% of whatever your reference clock is. RIght? Thats at least what I have seen with todays overclocking. Its not that I want to put it to 166 Target Speed, its just the system doesn't want to play nice if I set R.T.S to 200 and then try to raise the reference clock higher. Where is the stress being put on there? Isn't that stress going to the CPU then and not so much the memory since its supposed to handle 200mhz easy. Or is that still Northbridge thinking?


Edit : Actually the stress is on the memory since its dual sided dimms and the board recommends only 1 dual sided dimm. So are these overclock numbers I am hitting normal good numbers or does running two dual sided dimms interfering with its overclocking limits?

 

[Peter]

There is no "async" and no 1:1, simply because there is no FSB. The memory controller is directly inside the CPU, operating at CPU core speed. The externally visible RAM _bus_ clock is ALWAYS generated by dividing down from RAM _controller_ clock, which happens to be CPU core clock.

The "target speed" of 200/166/133/100 you set is telling the RAM controller to choose a bus frequency divider that keeps it below or exactly at this target - always assuming the reference clock is 200.

So yes, if you look at just the numbers, when you say "166" it'll give you memclk = 83% of refclk. However, if you also look at how the frequencies actually are generated, you'll see that the rise in CPU core (and thus RAM controller) clock will put its own limitations onto overclockability.

My guess: It ain't working because your 2750 MHz setup is plainly asking too much of either the CPU core or the RAM controller.

Why did I say "PC4000 is marketing bull"? Simple: Because there is no such standard. Any DDR1 speed grade beyond PC3200 is entirely made up. In particular, there is no standardized environment and method for verification of such a speed grade. For example, any RAM vendor can make up its own "perfect" environment wherein their DIMM achieves that PC4000 speed grade. In the real world, mileage will vary downward-only. A lot. Whereas the parameters for PC3200 compliance are well standardized - voltage, total max DIMM and chip count per channel, timings, everything.