Disaster with 512 meg sticks on Infinity

[mrlobber]

Ok, maybe I miss experience in something but here is the developing story:

 

My Infinity has run 240fsb stable on any bios tested so far with 2 kits of 2x256Mb sticks from Mushkin, both Level1 and LevelII as well (CH-5 and BH-5), and nothing more. That was so far.

 

Now, bought a 2x512 MB HyperX 3000 BH-5 kit from a pcper.com member. He has run them Pifast and Superpi stable at 235fsb 3.2volts on his Socket754 board, and I didn't have any reason not to believe him (as he provided a screenie too).

 

But now the most interesting part follows.

 

Received the sticks, put them in, 3.2v 220fsb CPC on 2-2-2. Got a BSOD in Windows. Hey I thought, not so fast man, do some testing with memtest first. Went to it - loads of 5th test errors. Then to CPC off. The same. Then down to 200fsb. The same. Then started to test the sticks separately.

 

CPU was ruled out with the 9th multiplier. I was running 6/19 bios with 7 additional RGone's stability timings, Vcore as needed, various Vdd (from Auto to 1.8), various AGP (from 1.5 to 1.8) tested.

 

One stick will run 200fsb memtest stable only at 3+ Vdimm :confused:, and its rated 187 fsb with 2.8 volts (not 2.5 as rated by Kingston). So it seems despite of the guy is telling me, the initial results lead to believe the stick is not up to its specs, if not simply bad?

 

Or has anyone heard about incompatibilites between HyperX3000 and DFI Infinity?

 

The other stick went memtest stable up to 225fsb 2-2-2 3.1 Vdimm CPC on. Any higher fsb gives errors but that might be attributed to the infamous CPC bug of the nForce2 chipset. So either I have to tweak more or live with that. That's acceptable and I didn't hope for the impossible. The same happens if I pair this stick with my 2x256 mb of CH-5 and try to run, for example 220fsb. It's memtest stable... but what confuses me in this case, is that Sandra bandwidth is 93%/87% instead of the usual 96%/90%? And that's with CPC on, CPC off gives 87%/77%. Wtf? I know that 512Mb sticks on nForce2 never are going to overclock to 256MB stick levels but do they experience some kind of performance hit regardless of CPC on/off in bios, as illustrated by Sandra and some other benchmarks I run as well?

 

I'm going to continue testing but any input would be appreciated.

[Popcicle]

We need to remember that the Socket 754 system and the

Socket A system are two different horses. Two entirely

different memory bus systems. The 754 memory bus system

was designed and copied after the Intel HTT memory bus.

 

RAM will play entirely different on these two types of boards.

 

Your testing and trials have been the correct method of

testing and checking the RAM that you are dealing with. And

your considerations towards 256 sticks vs. 512MB sticks are

correct.

 

The stick that went stable to 225 is the keeper. I would try

to obtain (find) a replacement on the other stick.

 

Also, we need to keep in mind that SiSandra testing and

benching are strictly sysnthetic results at best.

 

Luck.

 

 

Pops.

[RGone]

...somewhat the 'infamous' well it did X in X's board but not mine.

 

First the 754 board was running at 3.2V which makes a tremendous difference in 'his' stability and screen capture. Even more of a difference at the time of 'his' testing was the 754 board is a single channel board. Two huge differences in his reported situation and the one you have now put the memory in.

 

Sandra is a lovely little toy that we can play with but in the practical world of reality; it is just that little more than a toy. Sorry, but I have 'raced' too many folks on Sandra for high Memtest scores and is all adjusting for each bench that can make that bench higher.

 

I have been running San2003 because I run certain tests and San2004 has not allowed me to put the tests 'I want' first instead of having to scoll or move down the menu every time to use the test I want. Yep, they changed the proggie but not for 'my' benefit. I have never been able to find that % figure as relates to memory bandwidth. But then I am a MS red-nake. HEhehe. Anyway if you took the the % reading or 'what' it was likely supposed to measure in the first place it should be accurately picturing the possible bandwidth of setup versus the actual bandwidth. That would be what I would 'want' it to do and do very accurately. Yeh, right!

 

The bandwidth score that Sandra gives is 'extremely' influenced by FSB speed. More FSB and tighter timings will cause Sandra to give you an elevated MemTest score. Always has since I have used it. So if you lowered the FSB to put the 3 sticks in then the score went down and so too, the %figure it will quote. And stop and think of 'all' the problems with the two memory controller setup of the NF2 chipsets from the get go and how much better they perform with only a single stick>2x256 sticks>2x512sticks and 'down' the line of configuration. Certainly Sander is going to take a look at the already known to be less efficient memory configuration and say its' % figure is and will be lower. That is a fact. But none of that is really your problem.

 

Your real problem and the 'only' one that would concern me is that the 'two' of your recently purchased sticks perform SO differently from each other and that will lessen their performance 'together'. That is the real question in my mind. What to do about that? Pair em up and live with them? Take the best one and intersperse it in a 3 stick setup with your level2 stuff and RMA the weaker 512 stick IF it fails to perform as advertised by Kingston?

 

In pairing sticks of memory; I usually want them to be within 3FSB or each other in their performance when tested singlely and with about the same voltage on them. That is what you really need to sort if you ever plan to 'pair' them and try to run them past 225FSB while running in harness together.

 

Sorry for the length of post but so many trees that it is hard to see the forest.

 

Sincerely, RGone...:confused:

[mrlobber]

Thanks guys for the feedback.

 

RGone, as you might have noticed, the initial testing was done with 3.2 Vdimm as well.

 

I think, you a little misunderstood me. As far as I understand from Sandra, the percentage I was worried about, depicts the actual "determined" bandwidth versus " theoretical bandwidth" for a given fsb. Theoretical remains the same for a given fsb, actual - varies depending on memory timings, configuration etc. For 2x256 sticks, even in single channel configuration, I've never seen it to drop under 94%/87% even in sigle channel 2-3-3 configuration in the range from 200 to 240 fsb. Now with the 2x256 + 1x512 in dual channel, tha bandwidth is less for all the same hardware, except for one dual-banked memory module. As I said, maybe that's only because my knowledge is too less and I have missed an important point, but that's why I asked, could this be so and have other people who are using 512mb sticks, experienced similar behaviour.

 

Anyway, I'm not a Sandra membench whore as I have seen many cases where benchmarks like 3dmark which certainly is a more "real world" benchmark than Sandra (although not perfect as well) have given identical results while Sandra has reported suspicious "up and down" acitivities despite of one changed timing in the bios. I think, you understand, my bewilderment is just one of the consequences to my frustration that the sticks I thought, might have done enough fsb (although my board limits me to 240) just as the famous BH-5 usually does with enough volts, together do not perform nowhere as good as they "might" do - and this is regardless of the CPC on/off issue, as I was stopped in my overclocking road of them even before I run into that barrier.

 

Of course, the Socket 754 vs nForce2 might be the decisive factor in this story about the "expected performance". Once again, I didn't expect any wonders because I knew how picky the nForce2 chipset is about memory, and you, RGone, have already written countless novels on that topic .

 

And as I said, the story is developing and I might find some new results soon. Most of all, I'm going to run some "not-so-synthetic" benchmarks to see whether the performance differences are there indeed. But the real problem was indeed nicely summed up by you, RGone. Nevertheless, I'd still appreciate input from people perhaps more experienced as me, just to see the missing points in the mosaic, as it needs to be put together in another case of mysterious Infinity behaviour.

[uwackme]

Noooobeeeee alert.

 

Noobie to KHX3000 that is. I got 6 sticks, everyone of which wouldnt run over 185Mhz..... even at 3.3Vdimm... untill I burned them in for 72 hours at 3.3Vdimm on memtest #5. All 6 sticks got to 240Mhz+.

 

Also, they dont run CPC on FOR . EVER. Dont fool yourself if it boots, it will crash. The errors are few on mine now, but it only takes ONE. So even though 230Mhz CPC ON sorta works... not really. While 240+Mhz CPC OFF works 24/7 100% stability.

 

Preoccupy yourself with getting stability, then burning in, THEN later worry about the top end. Otherwise you'll just loose hair worry about it.

 

I didnt believe in "burnin" till I saw it with my own eyes. I doubt your seller did any real burnin.

 

BTW, after they started to get "up to speed" OFF came the heatspredders and stickers, as they are heatblankets in reality.

 

You can always CONFIRM they are BH5 by putting them in the freezer, and taking off the spredders (they come off wicked easy frozen) and see the #'s on the chips. D328DW-45 or 50 from date code 0225 to 0317 and you are golden. As they ARE then BH5 confirmed, you dont put the spredders back on, cause you aint EVER gonna RMA them. Even a bad stick is worth MORE than the RMA value as spare parts to fix OTHER BH5 sticks ;-) An SMD reworker can pull the good chips off and save them for repairs.

 

BH5, esp the Kingstons, are hipo animals and the proper care/feeding takes effort.... the rewards are worth it.

[mrlobber]

Originally posted by uwackme

Noooobeeeee alert.

 

Noobie to KHX3000 that is. I got 6 sticks, everyone of which wouldnt run over 185Mhz..... even at 3.3Vdimm... untill I burned them in for 72 hours at 3.3Vdimm on memtest #5. All 6 sticks got to 240Mhz+.

 

Also, they dont run CPC on FOR . EVER. Dont fool yourself if it boots, it will crash. The errors are few on mine now, but it only takes ONE. So even though 230Mhz CPC ON sorta works... not really. While 240+Mhz CPC OFF works 24/7 100% stability.

 

Preoccupy yourself with getting stability, then burning in, THEN later worry about the top end. Otherwise you'll just loose hair worry about it.

 

I didnt believe in "burnin" till I saw it with my own eyes. I doubt your seller did any real burnin.

 

BTW, after they started to get "up to speed" OFF came the heatspredders and stickers, as they are heatblankets in reality.

 

You can always CONFIRM they are BH5 by putting them in the freezer, and taking off the spredders (they come off wicked easy frozen) and see the #'s on the chips. D328DW-45 or 50 from date code 0225 to 0317 and you are golden. As they ARE then BH5 confirmed, you dont put the spredders back on, cause you aint EVER gonna RMA them. Even a bad stick is worth MORE than the RMA value as spare parts to fix OTHER BH5 sticks ;-) An SMD reworker can pull the good chips off and save them for repairs.

 

BH5, esp the Kingstons, are hipo animals and the proper care/feeding takes effort.... the rewards are worth it.

 

Actually, I don't mind being called newbie if it later contains some good points

 

I noticed that as well - they run awfully hot, I've never seen a memory stick running that hot. I considered a fan over them, now this idea seems to be confirmed.

 

And the Kingstons being very capricious sticks - my first PC3500 ever were KHX3500 - and they were complete crap on my first nForce2 mainboard - Epox 8rda+ - some kind of incompatibility, later recognized by Kingston as well.

 

So you did a 180 fsb 3.3 Vdimm burning marathon? It seems my sticks are scheduled to receive Olympic games soon

 

Thanks for the help.

[RGone]

...HOW MUCH, uwackme has caused my electric bill to go up in the last 3 weeks since he jumped up there with that 72hours burn-in thing he posted a few weeks ago in a thread. Hehehehehe. :nod:

 

But I used to burn in 'some'>now a lot. And I have gained about 12FSB on two sticks over 39hours burn time on them and an average of 3 useable FSB across 3 other pairs of memory. Hehehehehehe. It is hard to walk off for at 'least' 12 hours and leave memtest86+ just a churning. Thank goodness for another puter. Hehehehee. The good lawd must be trying to teach me patience because the 'goodness' of burn-in/out/up/down requires mucho patience.

 

Thanks "uwackme". Heheheheheheheheehehe. :angel: :shake:

 

RGone...:nod:

[mrlobber]

Originally posted by RGone

...HOW MUCH, uwackme has caused my electric bill to go up in the last 3 weeks since he jumped up there with that 72hours burn-in thing he posted a few weeks ago in a thread. Hehehehehe. :nod:

 

But I used to burn in 'some'>now a lot. And I have gained about 12FSB on two sticks over 39hours burn time on them and an average of 3 useable FSB across 3 other pairs of memory. Hehehehehehe. It is hard to walk off for at 'least' 12 hours and leave memtest86+ just a churning. Thank goodness for another puter. Hehehehee. The good lawd must be trying to teach me patience because the 'goodness' of burn-in/out/up/down requires mucho patience.

 

Thanks "uwackme". Heheheheheheheheehehe. :angel: :shake:

 

RGone...:nod:

 

Would be very hard, indeed. To leave the system on for 3days without being able to access this great forum, that takes a whole lot of nerve cells dead :nod:

[uwackme]

HEHEHE ty......

 

It is a nice luxury to have 4-5 systems around..... at the time I task'd MY Inf with the burning, while using CH5 sticks to set everything else up on my brothers system...raid/Wnxp etc. Time fly's with all that other work to pre-occupy you with.

 

ML.... you really need to lok into putting a FAT 120mm fan in the side door of your case, centered on the AGP/DIMM intersection.

 

Its a weird thing, but the airflow/convection on the DFI boards really ends up with a dead zone in that area if you dont.

 

This impacts your video ram sinks also. My underside/lower 2 videorams were in a total deadzone with the Vdimm regulator (funny square chip) on the Infinity "cooking" my 9800pro. Putting a mosfet sink on it only made it do a better job cooking the videoram :nod:

 

By using a side fan... BIG FAT SLOW one.... without making things noisey I solved all the heating issues. now the sinks are all just warm to the touch, and Im stable at full speed on much warmer days with NO AC in the lab. I havent bothered since we've been BELOW avg temps so far, but looks like the break is over and JULY looks like it is finally coming.

 

Im gonna recommend a change in my alpha timings too.... on reading, and cross correleating with the NF7 and my past experiments, both the (W2P) and (R2P) settings should be "4" at high FSB to make the 9800pro happiest and most stable.

 

This corresponds to the 11,2, "3", 2.0 required with CH5 on the NF7-S to make 3D stable. The AGP card is the guy that needs this, not the CPU. So CPU testing will see no diff, but I believe that the AGP 3D stability will be added too.

 

On the cooling front.... Im really liking a new MDF case, made with 8 papst 120mm silent fans, positioned w/2 blowing over the HDisks, w/2 blowing over the AGP/memory area, w/2 on back panel exhausting, w/1 mounted to the CPU hs, and 1 in the Superflower, all running off a single knob.....7V-12V Duct the intakes and exhausts so they dont have a direct path to the outside, lined with sound abosorber. Give the Superflower it's own intake path. Hmmm......thing is, even with watercooling, you still need a fan blowing well onto the mosfet/dimm area and videorams. hmmm

[mrlobber]

Originally posted by uwackme

ML.... you really need to lok into putting a FAT 120mm fan in the side door of your case, centered on the AGP/DIMM intersection.

 

Its a weird thing, but the airflow/convection on the DFI boards really ends up with a dead zone in that area if you dont.

 

This impacts your video ram sinks also. My underside/lower 2 videorams were in a total deadzone with the Vdimm regulator (funny square chip) on the Infinity "cooking" my 9800pro. Putting a mosfet sink on it only made it do a better job cooking the videoram :nod:

 

By using a side fan... BIG FAT SLOW one.... without making things noisey I solved all the heating issues. now the sinks are all just warm to the touch, and Im stable at full speed on much warmer days with NO AC in the lab. I havent bothered since we've been BELOW avg temps so far, but looks like the break is over and JULY looks like it is finally coming.

 

On the cooling front.... Im really liking a new MDF case, made with 8 papst 120mm silent fans, positioned w/2 blowing over the HDisks, w/2 blowing over the AGP/memory area, w/2 on back panel exhausting, w/1 mounted to the CPU hs, and 1 in the Superflower, all running off a single knob.....7V-12V Duct the intakes and exhausts so they dont have a direct path to the outside, lined with sound abosorber. Give the Superflower it's own intake path. Hmmm......thing is, even with watercooling, you still need a fan blowing well onto the mosfet/dimm area and videorams. hmmm

 

A good read. Messing around with my AGP would be the next thing to do after I'd finally settle the issues with memory... a custom cooler project is underway.

 

Btw, despite of many people saying soutbridge doesn't need active cooling, without a small fan on it, the heatsink goes really hot. I think, this correlates with your findings about that dead zone on Infinity without a side case fan. And now when those HyperX's start cooking in that area as well, the things can just go worse.

My case needs further modding, it seems.

 

What's a MDF case? Your own built one? (those acronyms can get you crazy if English is not your native language...)

[uwackme]

MDF... medium density fibreboard. Just dense particle board, still easy to work with but DAMN is it heavy. 4x8" sheet is over 80LBS!!!! The perfect material for silent PC cases. My case will be SHORT and WIDE with air duct in the front and out the back so the "fans" do not have a straight path to the outside....all lined with sound absorber. Think of it as an outer-chamber for a normal PC inside. Just the CD/Floppy and Connector area are exposed to the outside.

[SAE]

Yeah... those heatspreader's "heatpads" are really heat insulators trapping the heat between the spreader and the chips...

 

I never understood why kingston used that crap on their high speed ram... I made the KHX naked a while ago A fan suffices.

 

And the really great thing to see is the BH-5 chips 335WF on both of them

 

P.S. I never did this kinda burnin - 72hours+... :eek: Woah! My electrical bill would kill (me and my lil budget)

 

Imagine what these sticks could do with a REAL burnin