DDR Dilemma - What would you do?

[Red3]

Egged on by Tom, I just built my first PC based on an A7N8X.
How could I resist the features: Dual Channel DDR, SATA, DD5.1 encoding, onboard SP/DIF out, onboard Firewire...

But unlike Tom, I'm on a budget and not 100% up with the FSB, memory bus and all that Jazz, so I just go for the best bang for the buck solution:
- Athlon 2400+,
- 2x512Mb Crucial PC2700 CAS2.5
- (and a GF 4200Ti - couldn't resist that one!)

The guy who sold me the memory asked me if I was planning on Dual Channel DDR use. "Of course, Tom said it's the best thing since DDR!" I replied.
"Oh right, I'd better make sure these puppies are twins then." He goes into the back room and comes back with two serial number matched Crucial chips. "You sure you don't want Corsiar matched pairs?" he asks.
"What, pay the same price for half the RAM - you crazy?" I replied.
"Well I'll give you 7 days to change your mind." he says - like he KNOWS something I don't...

So here's the problem: I build my PC, everything seems fine and then suddenly about 10 minutes into browsing the net my computer just resets itself! No blue screen, no error, just plain shuts down and restarts.

So my first suspicion is the memory. I play around with a few settings in the bios. Same thing every time. Eventually, after some sniffing around the net, I work out that either I've got to
1) pull the RAM out of slot 3 and give up on Dual DDR, or
2) clock my RAM down to match the 133MHz FSB of the Athlon 2400.
I take option 2. Reboot, system's solid as a rock.

OK I can handle that. It only cost me $10 more for PC2700 over PC2100. But really I'm now running PC2100 RAM, right?
But then it dawns on me. How sweet it would be if I could now have DDR333 running in synch with the FSB.
Maybe I should up my Athlon? So I go back to my supplier's website. Suddenly I notice the 2500+ - only $40 more and running at 166MHz FSB. A match made in heaven? Not quite. The clock speed is only 1.8 GHz. How can I give up my 2.0GHz chip? Do I want FBS speed or processor speed? Arrggghh! The frustration - I just don't knwo anymore!
(The 2600+ is too expensive - another $100 price hike on top of the 2500).

So here's the dilemma:

Do I go through the hassle of taking my system apart and 'up' the processor to a 'slower' 2500+?

Do I keep what I have and be happy that my new system is very fast and now stable, but not SMOKIN' fast like it could be?

Do I try to solve the lock-ups and run asynch 133MHz FSB and optimum 166MHz DDR333 PC2700? CAN I solve the lock-ups?

Do I give up on Dual Channel DDR and run the RAM off one memory controller? Is Dual Channel really all it's cracked up to be? I have read that the ACTUAL performance gain is nowhere near the theoretical 2x on the A7N8X. Certainly not worth the premium price of Corsair TWINX RAM to me...

Or do I cave and trade in my 1Gb Crucial CAS2.5 RAM for 512Mb of TWINX Corsiar CAS2?

And finally the million dollar question - what REAL performance difference does it make?
1%? 5%? 10%? 100?
(Remember I'm not after performance at any cost - I want the best price/performance solution.)

It's a tough one - I did get the A7N8X to take advantage of the performance features after all, but I'm not really into overclocking...


Red-3 Standing by...<P ID="edit"><FONT SIZE=-1><EM>Edited by Red3 on 03/28/03 03:19 AM.</EM></FONT></P>

 

[bum_jcrules]

Reguardless to say I don't even have the budget that you do to build an A7N8X system. Anyways, as for your motherboard...

I downloaded the manual. If your memory setting is set for "By SPD," then the SPD chip on the memory modules will tell the BIOS to run at 166.667MHz in accordance to DDR333/PC2700.

By leaving the the Memory Timings to "Optimal" should be fine unless you want to use "User Defined" and set the Latencies however you like.

However try the 125% setting to see if this helps your stability. You mentioned that you played around with a few of the settings... What did you do so I don't recommend something you have already done.

<A HREF="http://www.crucial.com/store/listparts.asp?Mfr+Productline=Asus++Motherboards&mfr=Asus&cat=RAM&model=A7N8X&submit=Go" target="_new">Here is the current listing of memory that works with a A7N8X from Crucial.</A> I don't know if your part numbers match, but my guess is that they do.

I would like to know more of what you did.





<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=28752#28752" target="_new">THGC meeting? Would you go if there was one?</A>

 

[Red3]

Thanks for the link to Crucial. I'll check the memory part when I get home from work.
The BIOS was using "By SPD" as default and this was causing me grief because it was out of synch (too fast for) my 2400's FSB.

Are you saying that I should be able to run at 125% and increase memory performance?
I think my mem settings are currently 6-3-3-3-2.5


Red-3 Standing by...

 

[bum_jcrules]

The memory bus frequency is being generated as a multiple of the system clock. (CPU CLock) So if the system clock is 133.333MHz or (400/3)MHz, then for the memory clock to run at 166.667MHz or (500/3)MHz that is 125% figure of the system clock. ((500/3)/(400/3) = 1.25 x 100 = 125%)

Nothing really will change but I have seen screwier things make boards work.


You might have the tRP setting too low. Set it to 7. In the BIOS it is labled as "SDRAM Active Precharge Delay." Now theoretically you should be fine with 6 but then again you never know.



To answer your "and increase memory performance?"" question by being out of sync. Yes.

The MCH0 and MCH1 run on thier own buses. The FSB is on its own. What the MCHs have to do is generate a translational frequency to sync both the FSB signal and the memory bus signals together. It is like the transmission on a car.

Anyway, that is the same thing that happens when there is only one DIMM being occupied by something that runs at a speed faster than the FSB. The main system clock is generating the base signal. The MCH's signal generators operate at some multiple of the system clock. Overall there are three signals. The system clock, the memory clock, and the translational clock.

You will get better results from dual DDR333 than single channel DDR400. The gains are because of the bandwidth. The FSB is the one keeping up with the memory bus. That isn't a bad problem. Bieng in sync is the best but the FSB can't handle it nor the rest of the motherboard components at too high of a speed.

So try it at 125% which is 166.667MHz for the memory bus speed. Then try it some of the others if that does not work.

You might have to up the RAM voltage to 2.7V. That is under "DDR Referance Voltage." I leave that one up to you.

Your settings look okay but you can try what I stated above about the tRP setting up to 7 cycles.


Now all we can do is wait and see what happens.



<A HREF="http://forumz.tomshardware.com/community/modules.php?name=Forums&file=viewtopic&p=28752#28752" target="_new">THGC meeting? Would you go if there was one?</A>

 

[LtBlue14]

Bum said some good stuff, but all in all if you can't fix the problem, running at pc2100 speeds is ok. for one thing you can probably lower your memory timings and get you to cas2, maybe 2-3-3-5-1T
also, running your memory in sync with your processor may help with latency. AND pc2700 in dual channel won't give you much (if any) extra performance over pc2100 because your fsb bandwidth is 2.13GB/s at 133mhz (which is, incidentally, identical to single channel pc2100), while in dual channel your bandwidth doubles to ~4.27GB/s with pc2100. with pc<i>2700</i> your bandwidth is a whopping 5.33GB/s, far bigger than your fsb and really fairly useless except for minor latency improvements (but i think cas2 and synchronous operation offset these)
in any of the changes you make, we're talking about maybe a few percentage points difference, it's nothing huge and in my opinion nothing to worry about. i bet your pc2100@cas2 outperforms pc2700@cas2.5
well, it's possible anyway

<A HREF="http://www.tweaktown.com/document.php?dType=guide&dId=120&dPage=1" target="_new">WinXP tweak guide</A>
<A HREF="http://www.tweaktown.com/document.php?dType=guide&dId=145&dPage=1" target="_new">WinXP tweak guide 2</A>

 

[Red3]

Thanks for your input. So far I haven't had any luck running the memory above 133MHz in the Bios. I was however able to run stable at 6-3-3-2T, which I think has at least reduced latency to CAS2, or am I read it wrong? I'll have to run a benchmark on it.
I think I'll get a bit more familiar with it before attempting to o/c my CPU to try to meet my RAM somewhere in between 133 and 166, but it's encouraging to know that I can try that in future.
It seems there's no way to run this RAM asynch, though. System reboots after about 10 minutes use.


Red-3 Standing by...

 

[Crashman]

It's a scam. He sold you modules he knew wouldn't work in order to extort you into paying protection fees on modules that would. Have you ever considered buying your Crucial memory FROM CRUCIAL? It's probably a hell of a lot cheaper there, and full lifetime warranty, free shipping. <A HREF="http://www.crucial.com" target="_new">http://www.crucial.com</A>

<font color=blue>Watts mean squat if you don't have quality!</font color=blue>

 

[bum_jcrules]

Ahhh... Did I hear the voice of consumer reason? Why yes! It is, but can it be? Yes, Yes it is! Crashman to the rescue.


I have one question. Have you checked the PCB's for the model numbers and/or did you use a SPD checker? If they are Crucial then they are Crucial. I thought that their warranty was there even if you bought memory through a reseller. Is that the case? Now I know you would know that one for sure.




<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[Crashman]

Yes, the warranty should be there no matter where you bought it. But there is a chance one or both of the modules is not what the seller claimed them to be, or that the seller kept around 1 or 2 bad modules rather than exchange them, simply so he could pull this scam.

Of course if you buy directly from crucial you'll get matched pairs without paying extra, because crucial will pull them from the same bin.

<font color=blue>Watts mean squat if you don't have quality!</font color=blue>

 

[Red3]

I very much doubt the integrity of the dealer is in question. They have a very good reputation, and they actually sold them at a better price than Crucial. They are aware that there have been problems with TwinBank RAM on the A7N8X and were very up front and honest about it BEFORE I made the purchase.
I think if he had wanted to get me to buy Corsair TWINX, he would have tried to do it then and there. In fact I emailed customer support and they said they would gladly exchange or upgrade if I found the product to be faulty. They recommended using memtest86 to test it and inform them of the results.
I will check the serial numbers this evening.

I ran an SIS Sandra test on the system using the settings above and found that the memory running at PC2100/133 rate using the CAS2 timings above performed at about the same rate as PC2700/166. It trailed PC2700 twinbank scores somewhat.
I am not sure if it was testing just one bank of RAM or both, but I think the test was showing TwinBank in synch with the FSB @ 133MHz timed at CAS2.
I'm going to try pulling out a stick of RAM and also running both in banks 1 and 2 to see the differences. I'll get the numbers for you this evening.

Red-3 Standing by...

 

[Red3]

OK, finally managed to test the RAM.

I checked that it's Crucial brand and matches the part number on their website. Everything on the two chips is identical except for one number: BG111BM.CQ vs BG111BM.CV
Could this be the batch number?

I did some benchmarking in Sandra with different configurations and this is what I got:

In Sync with FSB
PC2100/266 CAS2.5 timings (7-3-3-2.5T) (Slots 1&2)
Int: 1896
Float: 1785

PC2100/266 CAS2 timings (6-3-3-2T) (Slots 1&2)
Int: 1901
Float: 1792

PC2100/266 CAS2 timings (6-3-3-2T) DUAL CHANNEL (Slots 2&3)
Int: 2012
Float: 1927

Async from FSB
PC2700/333 CAS2.5 timings (7-3-3-2.5T) (Slots 1&2)
Int: 1933
Float: 1853

All these three configurations ran stable. The only one I can't seem to run is Dual Channel PC2700/DDR333 async., which resets the system after just a few minutes.

As you can see, I'm getting only slightly better performance from Dual Channel PC2100 with aggressive CAS2 timings, but this does seem to be the best configuration.

But there's not much in it... Any comments?

Red-3 Standing by...

 

[bum_jcrules]

Just to rule Crashman's suggestion of foul play, have you tested the modules for errors? Memtest, etc?

Until we know they are truely good modules, that will always be a question.



<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[Red3]

I ran memtest86 v3.0 overnight and came down in the morning to see quite a few memory errors. I'm not one to jump to conclusions, so I'd like to understand the results a little better.
I'm running PC2700 async mode in slots 1&2 with aggressive CAS2 timings at present.

I got these errors:
Test 0: 0
Test 1: 2
Test 2: 0
Test 3: 52
Test 4: 4685 (!)
Test 5: 690
Test 6: 66

[That's all that had run before I left for work this morning.]

So what next? Run memtest on default settings? i.e. PC2700 CAS2.5

Or can I deduce from this that my memory needs replacing?


Red-3 Standing by...

 

[bum_jcrules]

Yes, you should run it at default. Those are the manufacterer specs and more errors can occur when running above spec.

My quess is that you will still find a lot of errors even at spec. Replacement is advised.


If you can get a screenshot of the errors and print it our, maybe you can get the Corsair for the same price.




<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[Red3]

Of course, the irony here is that by default the RAM is unstable (PC2700/DDR333 running asynch with my 133MHz FSB Dual Channel...this is how my supplier set it up in the shop.)

I suppose a fair test would be PC2700 in Slots 1&2 with Optimal timings (i.e. NOT Dual Channel)...

I'm pretty disappointed with the performance increase with Dual Channel RAM anyway - it's certainly not twice the bandwidth... 'optimistic marketing' by nVidia/ASUS?

Red-3 Standing by...

 

[LtBlue14]

technically it does provide twice the bandwidth, but your FSB can't handle it all, so in the end it really only helps with latency and "filling in the gaps" of information so to speak. as i said before, your 266mhz fsb gives you 2.13gb/s of bandwidth into the processor, and the dual channel ddr gives you 4.27gb/s at 2100 speeds. your processor's front side bus isn't wide enough to handle all that data. <A HREF="http://www.overclockers.com/tips00157/" target="_new">go here</A> for more

<A HREF="http://www.tweaktown.com/document.php?dType=guide&dId=120&dPage=1" target="_new">WinXP tweak guide</A>
<A HREF="http://www.tweaktown.com/document.php?dType=guide&dId=145&dPage=1" target="_new">WinXP tweak guide 2</A>

 

[Red3]

Thanks for the link.

I would say 'theoretically', not 'technically', since the A7N8X is NOT CAPABLE of providing double the bandwidth. Therefore the marketing hype is purely that. 120% DDR is perhaps more accurate?

Anyway, I tested my RAM last night at basic PC2100 rates (not Dual DDR) and received NO errors from Memtest86. Not one error in 1Gb RAM throughout all the tests.
I then clocked my RAM up to PC2700 (basic safe CAS2.5 timings 7-3-3-2.5T) and got 86 errors before I had to come to work. (I think it was about halfway through.)
So does this mean that my PC2700 RAM is not up to spec? It seems to do fine at PC2100 rates, but chokes at spec rates.

I just can't fathom it all out...

Red-3 Standing by...

 

[bum_jcrules]

The problem is the theoretical verses the real world. It is twice the bandwidth but with latencies you will never achieve optimal throughput. (Page misses, etc.) The FSB is still stuck at 2133.333 MB/s where as the memory bus is a true 5333.333 MB/s.

I don't know why he set it Asynch. Even if the memory bus is running at a higher frequency than the FSB the MCH will sync the signals using a third signal as an intermediary. So??? Why he did that, I don't know???

For simplified performance the single channel, one DIMM filled with DDR400 setup will do almost as well as a dual channel DDR333 setup in terms of bandwidth.

So if you were going to exchange the memory modules maybe you might just want to stick with one 256MB or larger DDR400 module.



<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[slvr_phoenix]

Of course there is still a small possability that it is the power supply and not the RAM...

The voltages on each rail <i>are</i> at spec, right?

<font color=blue><pre>If you don't give me accurate and complete system specs
then I can't give you an accurate and complete answer.</pre><p></font color=blue>

 

[bum_jcrules]

Good possibility. He never stated if it was a total new build or if it was a system upgrade.

<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[bum_jcrules]

I got these errors:
Test 0: 0
Test 1: 2
Test 2: 0
Test 3: 52
Test 4: 4685 (!)
Test 5: 690
Test 6: 66

Here is what I can tell you about your test. I didn't have time yesterday to respond so here we go today. I will only comment on the tests that found errors. Some tests are more critical or should I say more accurate than others.

Test 0-4 use inversion algorithims that use differing sized patterns of zero and/or ones to test the memory's physical status.


Test 2: 2

This test found two "Hard Errors." Basically it is checking the ability to change from a zero to a one and vise versa. You had two errors, that is not a good thing on this test. It is a very basic test for the ability of the memory to change. (Hysteretical evaluation) Overclocking the memory of setting the timings too fast can cause these. But with the way you set you timings I would blame the memory itself and not your timinigs. They aren't too aggressive. Now if your precharge settings were really low that will cause them but you were running at 2 cycles which those modules should be able to cope with. If you had lower grade modules you would see a lot more errors in this test if the precharge setting were too low.


Test 3: 52

Another reason I suggest that the memory is bad is because of this test's output. It is similar to Test 1 but uses 8-bit wide blocks.


Test4 : 4685 (!)

This is also the same. It uses 32-bit patterns. You are missing a lot of data. These modules are not good. They are functional but bad.


Test 5 uses move commands to take 8 mb of data from one area of your memory to another.

Test 6 uses another algorithm that finds errors that can go undetected under an inversion algorithm test because of cache buffering. Notice that you have more errors detected in test 6 than in test 3.

If you want to know more about the testing process post the rest of you results and we can go through them. Either way the modules are not good.


Sorry to say, it looks like Crash was right. You were probably scammed.






<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[bum_jcrules]

I got these errors:
Test 0: 0
Test 1: 2
Test 2: 0
Test 3: 52
Test 4: 4685 (!)
Test 5: 690
Test 6: 66

Here is what I can tell you about your test. I didn't have time yesterday to respond so here we go today. I will only comment on the tests that found errors. Some tests are more critical or should I say more accurate than others.

Test 0-4 use inversion algorithims that use differing sized patterns of zero and/or ones to test the memory's physical status.


Test 2: 2

This test found two "Hard Errors." Basically it is checking the ability to change from a zero to a one and vise versa. You had two errors, that is not a good thing on this test. It is a very basic test for the ability of the memory to change. (Hysteretical evaluation) Overclocking the memory of setting the timings too fast can cause these. But with the way you set you timings I would blame the memory itself and not your timinigs. They aren't too aggressive. Now if your precharge settings were really low that will cause them but you were running at 2 cycles which those modules should be able to cope with. If you had lower grade modules you would see a lot more errors in this test if the precharge setting were too low.


Test 3: 52

Another reason I suggest that the memory is bad is because of this test's output. It is similar to Test 1 but uses 8-bit wide blocks.


Test4 : 4685 (!)

This is also the same. It uses 32-bit patterns. You are missing a lot of data. These modules are not good. They are functional but bad.


Test 5 uses move commands to take 8 mb of data from one area of your memory to another.

Test 6 uses another algorithm that finds errors that can go undetected under an inversion algorithm test because of cache buffering. Notice that you have more errors detected in test 6 than in test 3.

If you want to know more about the testing process post the rest of you results and we can go through them. Either way the modules are not good.


Sorry to say, it looks like Crash was right. You were probably scammed.






<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[bum_jcrules]

I got these errors:
Test 0: 0
Test 1: 2
Test 2: 0
Test 3: 52
Test 4: 4685 (!)
Test 5: 690
Test 6: 66

Here is what I can tell you about your test. I didn't have time yesterday to respond so here we go today. I will only comment on the tests that found errors. Some tests are more critical or should I say more accurate than others.

Test 0-4 use inversion algorithims that use differing sized patterns of zero and/or ones to test the memory's physical status.


Test 2: 2

This test found two "Hard Errors." Basically it is checking the ability to change from a zero to a one and vise versa. You had two errors, that is not a good thing on this test. It is a very basic test for the ability of the memory to change. (Hysteretical evaluation) Overclocking the memory of setting the timings too fast can cause these. But with the way you set you timings I would blame the memory itself and not your timinigs. They aren't too aggressive. Now if your precharge settings were really low that will cause them but you were running at 2 cycles which those modules should be able to cope with. If you had lower grade modules you would see a lot more errors in this test if the precharge setting were too low.


Test 3: 52

Another reason I suggest that the memory is bad is because of this test's output. It is similar to Test 1 but uses 8-bit wide blocks.


Test4 : 4685 (!)

This is also the same. It uses 32-bit patterns. You are missing a lot of data. These modules are not good. They are functional but bad.


Test 5 uses move commands to take 8 mb of data from one area of your memory to another.

Test 6 uses another algorithm that finds errors that can go undetected under an inversion algorithm test because of cache buffering. Notice that you have more errors detected in test 6 than in test 3.

If you want to know more about the testing process post the rest of you results and we can go through them. Either way the modules are not good.


Sorry to say, it looks like Crash was right. You were probably scammed.






<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[bum_jcrules]

I got these errors:
Test 0: 0
Test 1: 2
Test 2: 0
Test 3: 52
Test 4: 4685 (!)
Test 5: 690
Test 6: 66

Here is what I can tell you about your test. I didn't have time yesterday to respond so here we go today. I will only comment on the tests that found errors. Some tests are more critical or should I say more accurate than others.

Test 0-4 use inversion algorithims that use differing sized patterns of zero and/or ones to test the memory's physical status.


Test 2: 2

This test found two "Hard Errors." Basically it is checking the ability to change from a zero to a one and vise versa. You had two errors, that is not a good thing on this test. It is a very basic test for the ability of the memory to change. (Hysteretical evaluation) Overclocking the memory of setting the timings too fast can cause these. But with the way you set you timings I would blame the memory itself and not your timinigs. They aren't too aggressive. Now if your precharge settings were really low that will cause them but you were running at 2 cycles which those modules should be able to cope with. If you had lower grade modules you would see a lot more errors in this test if the precharge setting were too low.


Test 3: 52

Another reason I suggest that the memory is bad is because of this test's output. It is similar to Test 1 but uses 8-bit wide blocks.


Test4 : 4685 (!)

This is also the same. It uses 32-bit patterns. You are missing a lot of data. These modules are not good. They are functional but bad.


Test 5 uses move commands to take 8 mb of data from one area of your memory to another.

Test 6 uses another algorithm that finds errors that can go undetected under an inversion algorithm test because of cache buffering. Notice that you have more errors detected in test 6 than in test 3.

If you want to know more about the testing process post the rest of you results and we can go through them. Either way the modules are not good.


Sorry to say, it looks like Crash was right. You were probably scammed.






<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>

 

[bum_jcrules]

Yes!


My first ever quad-post!!!





<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>