whats holding back my FSB???

Man, im already ****ed...

I can run my cpu at 174x10.5=1827 at 1.775v while prime95 torture test for one straight day without a hitch. As you should note, i have a corsair pc3200 cas2 stick far below spec. With this speed it can run fastest timings and 1t flawless.

Anyway, if i do sandras benchmark, my memory bandwidth sux!!! i cant keep up with pc2700 score . so i thought lets change multi and higher fsb.... well, even being the cpu at the same speed like 9.5x192=1824MHz and the memory at rated specs (2-3-3-6 1t) prime 95 gives out errors after 5 min... i sad f*ck it, lets try most lagged timings.... SAME RESULT!!!!

Now, the cpu isnt the problem since its proben at that speed, corsair is corsair for christ sakes!!! i cant have a bad sample... not even at lagged timings. That leaves out maybe the northbridge... ill try to put a small 40mm fan on it and keep you guys posted.

Any other ideas on what is holding me back???

Forgot to mention, at 192mhz fsb my sandra benchmark is barely on par with pc2700 even with most agressive timings... GRRRRRRRRR

well, through research I have done, I've found that pc2700 uses faster chips (nanosecond wise) than any higher (3200/3500) ram. this may be the cause of your bandwidth problems. I would try and raise the vcore a notch or two also, my 1600 didn't like to run at stock speed @ 190 fsb without a bump.

PhobMX said:

Man, im already ****ed...

I can run my cpu at 174x10.5=1827 at 1.775v while prime95 torture test for one straight day without a hitch. As you should note, i have a corsair pc3200 cas2 stick far below spec. With this speed it can run fastest timings and 1t flawless.

Anyway, if i do sandras benchmark, my memory bandwidth sux!!! i cant keep up with pc2700 score . so i thought lets change multi and higher fsb.... well, even being the cpu at the same speed like 9.5x192=1824MHz and the memory at rated specs (2-3-3-6 1t) prime 95 gives out errors after 5 min... i sad f*ck it, lets try most lagged timings.... SAME RESULT!!!!

Now, the cpu isnt the problem since its proben at that speed, corsair is corsair for christ sakes!!! i cant have a bad sample... not even at lagged timings. That leaves out maybe the northbridge... ill try to put a small 40mm fan on it and keep you guys posted.

Any other ideas on what is holding me back???

Forgot to mention, at 192mhz fsb my sandra benchmark is barely on par with pc2700 even with most agressive timings... GRRRRRRRRR

Click to expand...

Two weeks ago I got from googlegear a brand new Corsair 3500C2 which run 20 MHz below stock frequency, and gave thousands of memtest86 erros at stock speed 217 MHz 7-3-3-2 timing. I don't know what statistics is for getting a bad Corsair module, given its good name and premium price (30-40% more $$ for 5% more performance). I wasted lots of time and money on it. I confirmed that it was a bad stick using memtest86 ver 3.0.

ADD: My MB can run stable at 211-217 MHz with another memory stick.

I think you should download memtest86 ver 3.0 onto a diskette. Set the stock timing for Corsair 3200C2 which is 200 MHz 6-3-3-2 in the BIOS. Then just run memtest86 for a few passes to see whether such module is error free or not.

This test runs on bare hardware, it bypasses Winodow XP, 3D mark, prime95, ... So if the module fails, you know is not the MB, CPU, fsb, ... at stock freq, most likely is the memory stick. In my case, it gave thousands of errors at memory address all over the places, ... believe it or not on a premium Corsair stick.

In case your MB cannot run in SYNC at 200 MHz for memtest86. Run it ASYNC so that memory speed is still at 200 MHz.

Like to hear back what you get from the test.

LittlePiggie said:

well, through research I have done, I've found that pc2700 uses faster chips (nanosecond wise) than any higher (3200/3500) ram. this may be the cause of your bandwidth problems. I would try and raise the vcore a notch or two also, my 1600 didn't like to run at stock speed @ 190 fsb without a bump.

Click to expand...

PC2700 is slower than PC3200 slower than PC3500 in terms of cycle time (ns).

If the module is not overclocking the DRAM chips under the heat sink,
PC2700 ----- 6ns DRAM chip
PC3200 ----- 5ns ..
PC3500 ----- 4.7ns ..
PC3700 ----- 4.3ns ..

Re: Re: whats holding back my FSB???

hitechjb1 said:

I think you should download memtest86 ver 3.0 onto a diskette. Set the stock timing for Corsair 3200C2 which is 200 MHz 6-3-3-2 in the BIOS. Then just run memtest86 for a few passes to see whether such module is error free or not.

This test run on bare hardware, it bypasses Winodow XP, 3D mark, prime95, ... So if the module fails, you know is not the MB, CPU, fsb, windows, ... it is the memory stick.

Click to expand...

Untrue. If you plugged your floppy into the memory stick alone, you could rule out motherboard or other concerns. As it is you are running the computer, albeit in dos. As such problems with the motherboard, power supply, BIOS settings, or CPU may indeed affect the results of your test. There is no doubt your corsair stick was bad, rather than any other component. But an error in memtest establishes only that there is a problem, not the offending component. The memory stick depends on the rest of the system for its operation, and therefore cannot do its job if there is a problem elsewhere. A second memory stick and/or another machine to test it in is required to fix blame on the ram with all certainty.

Re: Re: Re: whats holding back my FSB???

larva said:

Untrue. If you plugged your floppy into the memory stick alone, you could rule out motherboard or other concerns. As it is you are running the computer, albeit in dos. As such problems with the motherboard, power supply, BIOS settings, or CPU may indeed affect the results of your test. There is no doubt your corsair stick was bad, rather than any other component. But an error in memtest establishes only that there is a problem, not the offending component. The memory stick depends on the rest of the system for its operation, and therefore cannot do its job if there is a problem elsewhere. A second memory stick and/or another machine to test it in is required to fix blame on the ram with all certainty.

Click to expand...

Agree that the best control test is to swap with another MB and RAM and test. The memtest86 is a necessary but not sufficient test.

I made an assmption that the given MB, CPU, buses, ... can run at the stock frequency of the RAM in question, otherwise why would one buy expensive RAM to run on MB that cannot handle such speed.

Also MB, w/ CPU set to much lower speed than stock are less error prone than the full system, since most unstablility factors such as windows, 3D video, HD paging, ... are already taken out.

Usually DOS in minmal hardware can run fsb 10-20 MHz higher than stable windows with 3D video, ... and Memtest86 does not even need dos to run.

I am tending to think it is the northbridge also, or Corsair is beginning to put out a lot of bad ram.... take your pick?

Ed or one of the writers here at oc.com wrote an article on the northbridge holding back the oc a month or so ago. I've looked and looked for that article and now I can't find it. Anyone know where it is?

Anyhow I added me a 2nd stick of XMS3500 this past week and now I can't oc as much as I could before. Well only 1 mhz difference at 158, was 159 over 166 strap. So that's about 212 mhz oc on my ram. But I had to use the most lax timings possible. 2.5-3-3-8. This shouldn't be for XMS3500! Even if 2 sticks of ram oc's less than 1, this is too much! Anyhow I tested the ram with DocMem at 158 and it checked ok. Never checked it at 159 with 2 sticks. However earlier when I was first oc'ing my new mobo, cpu and ram my system started giving freezes and locks due to memory erros (and according to DocMem) above 160. Actually I could boot into WinXP and run for awhile at 165 before I would lockup. (With 1 stick of XMS3500 running at 159 mhz I still had to run lax timings of 2-3-3-7. Also my DRAM voltage is set 2.9v)

All this was similar to what the article described about the northbridge being too hot and or holding back your oc. I to am thinking about putting a fan on it and maybe some AS2 under the NB heatsink. Still I wished I could find that article.

Cheers,
Mike

What MB are you running with, AMD or P4?

It looks like yours is a P4 MB running under 166 MHz. Regardless of how high you can oc, P4 will benefit from the dual channel bandwidth improvement.

If it is an AMD XP, I found that 1 module overclock few MHz better than two modules in single channel which overclock few MHz better than two mouldes in dual channel. Reason unknow to me, I suspect is the memory controller (in NB) related, hardware and/or driver. IMHO, the higher overclocking MHz outweighs the goodness of dual channel and bank interleaving, hence translates into better memory bandwidth and 3D performance.

Asus P4PE P4 2.53 @ 3002
In BIOS memory set to 166
Then 158 for oc

Cheers,
Mike

Mike360000 said:

Asus P4PE P4 2.53 @ 3002
In BIOS memory set to 166
Then 158 for oc

Cheers,
Mike

Click to expand...

Are you running fsb and memory bus at a ratio? If so, what ratio?

hitechjb1 said:

Are you running fsb and memory bus at a ratio? If so, what ratio?

Click to expand...

Well actually this is my first Asus mobo and I've been use to Abits.
And well, the memory options in the Asus are different than the Abit so I'm having a little time explaining them.

In my Asus BIOS I have 2 settings for memory:
1 is my "Memory Overclock" which I have set 158, and which was set to 159.

2 is settings called "memory Frequeny" which is set to or at 421 mhz. This is the setting that appears to be most like the Abit setting of "Bootstrap" which regulates the ration at 233/266/333.
But it isn't configured exactly like the Abit setting.

Anyway a doublecheck with Sandra shows it to be 211 or 421mhz and an oc of my cpu of a 158 multiplier, giving me 3002 mhz.

Matter of fact I just quit the above, rebooted, went into BIOS and raised my multiplier to 159, went back into WinXP and into Sandra and copied this:

System Chipset
Model : ASUSTeK Computer Inc 82845PE Brookdale Host-Hub Interface Bridge (B0-step)
Front Side Bus Speed : 4x 159MHz (636MHz data rate)
Max. Bus Bandwidth : 5088MB/s (estimated)

Logical/Chipset Memory Banks
Bank 0 Setting : 256MB DDR-SDRAM 2.5-2-2CL
Bank 1 Setting : 256MB DDR-SDRAM 2.5-2-2CL
Bank 2 Setting : 256MB DDR-SDRAM 2.5-2-2CL
Bank 3 Setting : 256MB DDR-SDRAM 2.5-2-2CL
Memory Bus Speed : 2x 212MHz (424MHz data rate)
Max. Memory Bandwidth : 3392MB/s (estimated)


BTW, the memory timings shown above does not match the timings in my BIOS.

Cheers,
Mike

1. fsb = 159 MHz
memory bus = 212 MHz
so fsb to memory ratio = 159 / 212 = 3 / 4 = 0.75

Max P4 fsb bandwidth = 159 x 4 x 8 = 5088 MB/s (x 4 because quad pump fsb, x 8 because 8 bits per byte)
Max memory bandwidth = 2 x 212 x 2 x 8 = 6784 MB/s (x 2 because DDR and dual channel, x 8 because 8 bits per byte)

2. The efficiency = 5088 / 6784 = 75%
Actually this can be computed just based on the ratio of fsb to memory.

Efficiency = ratio = 0.75 = 75%

In order to get a higher efficiiency, one has to run them in SYNC (but may not be possible at high fsb though).

You need to plug in the actual SiSoft benchmark number in order to get the actual effective efficiency and memory bandwidth.

3. In order to get higher bandwidth at this fsb/mem ratio, you have to overclock the fsb higher (if you can do it), since your Corsair 3500 is spec'd at 217 MHz and you are now only at 212 MHz.

Well that's basically what I understood, not considering all the math. I knew my divider was maxxed, so I couldn't improve that.
So as you said only oc'ing my fsb was left. And that was where my problem is. I can't really go any higher! My ram timing are as lax as they will go at: 2.5-3-3-8! Isn't this too lax for Corsair's specs?

Anyhow I just raised my multiplier back up to 159 to see if I could run stable there now. (Will chack out a game and or do a burn in test in Sandra.) I hadn't tried this combination with my ram stick in Dimm 1 & 3. I couldn't get Dimm 2 to work at all.(I posted on this in another thread. Can't remember if you saw it or not.) So what you see here is why I'm wondering if my Northbridge may be hampering my oc any.

"You need to plug in the actual SiSoft benchmark number in order to get the actual effective efficiency and memory bandwidth."

I did run the actual benchmark. I get around 3200 in both mem tests. I forget what my efficiency was though. I just rechecked and bandwith efficiency is estimated at 93%.


Cheers,
Mike

Mike360000 said:

I did run the actual benchmark. I get around 3200 in both mem tests. I forget what my efficiency was though. I just rechecked and bandwith efficiency is estimated at 93%.

Click to expand...

3200 MB/s bandwidth looks like is not getting any benefit from dual channel.

A single module running at 200 MHz (DDR 400) has already the bandwidth of 3200 MB/s, minus some overhead would still be 3000 MB/s (I can get 3000 MB/s in an AMD MB running single channel at 200 MHz)

I am not familiar with your MB, make sure the mem are plugged into the right dimm pair.

hitechjb1 said:

3200 MB/s bandwidth looks like is not getting any benefit from dual channel.

A single module running at 200 MHz (DDR 400) has already the bandwidth of 3200 MB/s, minus some overhead would still be 3000 MB/s (I can get 3000 MB/s in an AMD MB running single channel at 200 MHz)

I am not familiar with your MB, make sure the mem are plugged into the right dimm pair.

Click to expand...

Oh no, you mis-understand hitechjb1...
I have a P4PE which is not a dual channel mobo.
It's the Intel PE chipset.
Anyhow my concern was I think either my ram timings are not up to Corsair's specs or my Northbridge is holding me back.
Well I was able to play a couple rounds of UT2k3 at 159fsb/3021mhz without my puter freezing.

Cheers,
Mike

I thought your MB was a dual channel one, ...

You can run a memtest86 or something like that with one memory module at a time at your favorite memory slot. Memtest86 runs on bare hardware without DOS and Windows, eliminating the uncertainty and instability of drivers, 3D, ... Most likely different dimm will give you difference max memory freq (by a few MHz !!!).

You can set your MB with a ratio (even lower than what you are now using, such as 50% or 66%) such that fsb, CPU speed are below spec and only stressing the memory frequency to Corsair spec of 217 MHz 7-3-3-2. Then run memtest86 for a few passes to see what happens. If it performs to spec, you can then try higher frequency or tighter timing.

hitechjb1 you seem familiar with memtest86. To use that program must you load it on to a bootable floppy and run from a DOS prompt?

Farwalker2u said:

hitechjb1 you seem familiar with memtest86. To use that program must you load it on to a bootable floppy and run from a DOS prompt?

Click to expand...

The version that I use (memtest86 Version 3.0) creates a diskette to boot from it. I think it is not even using DOS to boot. So it runs on bare hardware bypassing all the uncertainty from Windows drivers, HD paging, 3D video instability, ...

Also I usually lower the CPU speed and just set the memory speed to whatever frequency and ras/cas timing you want to test. If you can run a ratio of fsb/memory speed on your MB, you may lower the fsb too and run the memory speed higher, so only stressing the memory of your system during memtest86 runs.

alright dudes, back to my problem i just installed a small 40mm fan on my northbridge, and i still have problems

whats next to try??? the memory slot position?? unplug pci cards or swap em until i can increase fsb??? im running out of ideas...

i have a 400w antec smart power psu, i dont think there might be voltage problems, ill post my voltages in a while anyway..

PhobMX said:

alright dudes, back to my problem i just installed a small 40mm fan on my northbridge, and i still have problems

whats next to try??? the memory slot position?? unplug pci cards or swap em until i can increase fsb??? im running out of ideas...

i have a 400w antec smart power psu, i dont think there might be voltage problems, ill post my voltages in a while anyway..

Click to expand...

Wished I knew. I tried the same thing you did, and I applied AS3 to the heatsink. Didn't do a bit of good. I noticed my northbridge was cool before adding the fan anyway, so I took it back off. Really for the ram I'm running, Corsair xms 3500, I should be getting higher fsb. And to add insult to injury I'm already running the ram with as lax a timings as possible. If I raise any last thing to higher standards my computer will end up freezing up on me.

Cheers,
Mike