DDR1 speeds... is this really possible?

[Formula350]

Got a DFI LanParty UT NF4-DR Expert for the cost of shipping since the owner said it was probably dead. It'd POST but BSOD when loading Windows or when trying to reinstall (which I replicated). I just needed it to diagnose why my PC wasn't POSTing at ALL, either dead CPU, bad RAM or PSU. Eitherway, everything turned out being OK and the old rig runs just fine now, whatever the problem ended up being :\ Now I figured I'd try to get the DFI running and flash an older BIOS to it, but in the end it didn't need it as it as well magically came back to life lol

Been tinkering with it, tweaking my old components to see what I could get out of them with much more precise tuning options. Figured with the ability to change DDR voltage not only much higher than my Gigabyte board but also in about +0.3v increments, I'd see what more this memory could do. One thing I did find out, much to my dismay and annoyance, was these Corsair sticks run PROMOS chips, which after some googling turns out to be pretty crappy :\ I had managed to get them stable around 480mhz though, seemingly more than most, but I typically ran them at 450mhz. Never was sure at the voltage my GB board was giving them, but it was around 2.70-2.72v, lower than the 2.75v Corsair tested them at. Started at 2.8v and attempted JUST running the memory @ 233mhz (9 divider) with everything else default settings. Bottom line, not stable. Would boot into windows and then almost immediately BSOD or hard lock. The one time I was able to load A64Tweaker to check the speed, it was running @ 223mhz, which I know to be stable not only in this board but in general. Figured it must not like the divider of 9 for whatever reason, so I decided to go for the gusto @ 250mhz (8 divider) and... it posted! I had once gotten 500mhz but wasn't stable, yet it appears to be just fine in the DFI board. From there I started increasing the HTT (FSB) by 2mhz and running Sandra and a few passes of 64K and 128K SuperPi to make sure it was stable enough. Managed to get it up to 265mhz (DDR530) before the system puked, figuring that was my limit, but with the DFI board's odd nature of crashing for no reason after upping the HTT speed so much I tried starting windows @ 215mhz HTT (DDR530) which didn't work out. Ran some MemTests at those speeds and since Test 5 always showed errors while 1-4 didn't, I started there. Got only a few, 12 errors on the first pass, in comparison to a few thousand I had once. Finally found that running at Command Rate of 2T gave stability, where I've always ran @ 1T. Changed that in the BIOS, leaving the rest alone (still at DDR530), and windows booted up fine. Ran some benches, which Sandra's mem score was much lower which I expected, but it was stable. So I began again increasing the HTT again by 2mhz and finally am at 280mhz (DDR560!!).

NOW I'm questioning if I'm actually running at these speeds... 1) Sandra, SysTool and MemTest show between 133-147mhz (266 DDR), with SysTool showing what initial speed is selected being 133mhz 2) A64Tweaker and HW32Info show 250mhz (or in this case 280) 3) Sandra's memory score reflects that of 500+mhz memory. My GB board was netting around 6,050mb/s with tight timings with CR 1T, at around 265mhz 1T on the DFI I was netting 6,500-6,600mb/s! Though @ 280 2T it's around 6,100mb/s.

Anyone know what speed I'm really running them at?! I really can't imagine them pulling off these speeds. While I do have a 40x20mm fan blowing on them, they are EXTREMELY cool, only what I would consider lukewarm to warm after repeated mem bandwidth tests. Yet if it was running at the lower reported DDR266ish speeds, why can't I run @ 1T when I can when I know I'm @ DDR460? Also why do I hit a wall when I try to increase the speed over 285 (DDR570)? I haven't even tried to loosen up the timings either, in an attempt to go further. Currently timed @ 3-3-3-8 2T, Trrd-2, Trwt-4, Twr-3, Twtr-2, Twcl-1. I'd even be willing to bet that it'd be fine @ 3-3-2-8....

I haven't run CPUID yet, but I plan to, and can post a screenshot later. I have to run some errands now though... Just all seems a bit unreal to me lol

 

[sneekypeet]

first off the DR expert was amazing with ram, so your clocks arent surprising to me at all.

Second, whatever CPU-z shows for ram speed is what you are running, Sisoft and everest dont always read correctly.

Third, if you want to play around some more, google TMODS DFI bios disc, has every bios on it, even the beta's and special bios' for TCCD and BH5 ram IC's.

fourth 3-3-3-8 is where I ran my Kingston 4GB kit at 280mhz 24/7 (560 effective). was able to push to 300 at times but it was unstable w/o a ton of voltage, which again the DFI has plenty of love to push ram farther

 

[Wrigleyvillain]

Wanna buy my BH-5?

 

[sneekypeet]

IIRC crucial was big on BH-5 and my Tracers were the shit, but it was only a 2X512 kit. Also keep in mind that with certain bios' on that DFI, the ram slots were "seperated". meaning the yellows ran TCCD better and the oranges ran other IC's better. Play around with it, AthlonX2 used to put up insane numbers with that board in the original AMD overclockers club.

 

[Wrigleyvillain]

Yeah I only have 2x256 Corsair XMS. Benching sticks, good for 2-2-2-5 450+

 

[Flyordie]

IIRC crucial was big on BH-5 and my Tracers were the shit, but it was only a 2X512 kit. Also keep in mind that with certain bios' on that DFI, the ram slots were "seperated". meaning the yellows ran TCCD better and the oranges ran other IC's better. Play around with it, AthlonX2 used to put up insane numbers with that board in the original AMD overclockers club.

I have a 256MB stick of TCCC. From HP, but has the Samsung Sticker on it still as well.
I also have a 512MB stick of UCCC (Crucial, but direct from Samsung as it still has the Samsung sticker on it)

The HP stick ran up to DDR-600 @ 2.8V 3-4-4-11-2T
The Crucial stick ran up to DDR-610 @ 2.95V
--
The set that I love the most however, is my OCZ's. 2-2-2-6-1T @ DDR-400 with just 2.6V = Win. 2x1GB set and they can both do it. The odd part is though- Some say the OCZs use Qimonda DDR, while some use Samsung. I haven't peeled back the XTC heatsink on them and don't intend to, but I have found that they are more likely Qimonda as Samsung wasn't heavy on the Low Latency DDR in high density modules.
I am guessing my OCZs are Qimonda.
-----

 

[sneekypeet]

thats because "certain" companies like to use revisions on their ram sticks. Release some really great Ic's on the first one or two series of the sticks, then as the revisions got bigger in number the IC quality drops as well OCZ arent the only ones. who still to this day, do this.

 

[Formula350]

thats because "certain" companies like to use revisions on their ram sticks. Release some really great Ic's on the first one or two series of the sticks, then as the revisions got bigger in number the IC quality drops as well OCZ arent the only ones. who still to this day, do this.

Yea, seems Corsair (who I adored until finding this out the passed week) apparently did this as the majority of theirs use Promos IC. I was kind surprised to see that the version of these 2x1gb sticks is 8.1 heh I can't complain any though, as I had the 2x512mb kit and when I RMAed them, they said that they are obsolete and had to upgrade me to 2x1gb ^_^ It's just surprising to me that they, after all I've read, are able to do 560.

Other than the odd couple of issues with this board, I do like quite a bit about it. Yet if I can't get the thing to boot past 260mhz HTT, that's 100-200mhz off my CPU I can't get back. I still don't see why this thing has an issue with that low of a HTT speed, considering the GB board that didn't have any descernable voltage modifier for whatever controls the HTT speed aside from HT Link (or however it was worded) voltage or Chipset PCIe (again however it's worded exactly) voltage. The latter was worded slightly different, making it seem like it could be either for one thing OR another heh I may knock the HT multiplier and memory down to FULLY remove them from the equation and try upping the HTT speed then, and who knows there may be some weird quirk with this thing that doesn't like a memory speed or certain divider, that I'm running into with the 265mhz HTT

What's the max voltage one would want to go on the HT (LDT) bus and NF4 chip? The NF4 I'm able to keep under 40C, currently idling at 35C with the fan on low. LDT Voltage I've got at 1.3v which the BIOS was reading 1.33v but SysTool reports 1.30v. NF4 I can't recall what it's sitting at right now, 1.6v I believe, but I can crank it up passed 1.9v if I wanted heh

Also according to what I've read, my CPU apparently is one that fixed the Errata94, but has Errata123 which has something to do with bypassing L2 cache and going straight to DRam. So if you disable the BIOS fix for it, you should have instability, yet here it sits happy as a camper >_> Was hoping that would fix the 265HTT issue, but Auto or Disable didn't help anything :\

I'll try tracking down that BIOS pack too, thanks

Here's screenies and link to CPU-Z
http://valid.canardpc.com/show_oc.php?id=1002003

And Sandra's little hiccup, which put me at 147% of bus bandwidth lol:
Logical/Chipset Memory Banks
Bank 2 : 1GB DDR 3.0-3-3-8 2-16-3-2 2T
Bank 3 : 1GB DDR 3.0-3-3-8 2-16-3-2 2T
Channels : 2
Bank Interleave : 2-way
Memory Bus Speed : 2x 150MHz (300MHz)
Multiplier : 1/15x
Width : 64-bit
Integrated in Processor : Yes
Cores per Memory Controller : 2 Unit(s)
Maximum Memory Bus Bandwidth : 4.69GB/s

 

[Jor3llBR]

Had my NF2 running 24/7 with 2x256MB Kingston Hyper-X week 38 BH-5 PC3200 one of the very best weeks for BH-5.

They would take 3.5vddr to 3.8vddr and still work fine today...


273mhz ddr1 2-2-2-9:

 

[Mussels]

my record was 2x1GB at 633MHz on a DFI NF4


All those other programs (not CPU-Z) are reading 'stock' settings - stock FSB with the divider selected. So if you start at 266MHz before you OC, they will read 266MHz regardless of what the real clocks are.

 

[Formula350]

Had my NF2 running 24/7 with 2x256MB Kingston Hyper-X week 38 BH-5 PC3200 one of the very best weeks for BH-5.

They would take 3.5vddr to 3.8vddr and still work fine today...


273mhz ddr1 2-2-2-9:
[url]http://img407.imageshack.us/img407/4663/4knsxns.th.png[/URL]

heh I guess maybe I should've clarified with "This really possible on Promos equipped DDR" I mean because I've seen reports from folks running DDR600, but that was with the GOOD stuff

 

[Jor3llBR]

heh I guess maybe I should've clarified with "This really possible on Promos equipped DDR" I mean because I've seen reports from folks running DDR600, but that was with the GOOD stuff

Yes with PROMOS would be very unlikely! DDR600 only on NF4 on NF2 would be VERY hard.

 

[Formula350]

Yes with PROMOS would be very unlikely! DDR600 only on NF4 on NF2 would be VERY hard.

Guess my sticks have been touched by god

 

[Formula350]

Here's another thing that now makes me question this board actually running the memory at/above 250mhz... I figured since the memory is supposedly capable of 280mhz, and I know that I could at least get the system working with a 250mhz HTT (was able to actually goto 260, but wanted to play it safe), so an extra 250mhz on the CPU while keeping the memory a fair amount under it's limit would be a much better route. So I dropped the memory divider to 1:1 (or 2:2 technically) and increased the HTT to 250, along with a small NF4 volt bump just to ensure it... Nothin. Doesn't even get to the CMOS backup check-point after the POST. I get the device/IRQ screen but it just stalls and the board reboots itself. Fine, dropped it down 5mhz and same result same for 235 and same thing. Finally dropped it down to 225 and 2:2, bupkis (nothing, for those not familiar with that Yiddish term lol)... Tried different timings at higher and lower HTT speeds, with a last attempt I put it back to 225 and put the memory to 180 (which I believe is a divider of 11) and STILL wouldn't go I want to say it has something to do with my dual PSU setup, but at the same time I can't figure how that could be the case since running the second brings the 5v rail up from 4.7v (can drop to 4.65v under memory load) to 4.82v (only drops to 4.79v under memory load). Temps are all normal, and if it wasn't a pain in my ass to swap boards I'd make SURE the CPU is fine by running it back on the Gigabyte. What makes it a pain is that damn IHS creates SUCH a friggen suction to the heatsink, that it will literally pull the CPU out of the locked socket... HOW it's not ripped pins off the CPU, I haven't any idea, but that's the last time I plan on pulling that CPU as I've removed it probably 10 times in it's life with all but maybe twice not coming out stuck to the heatsink And yes, I've tried prying up a corner of the heatsink in an attempt to pop the suction, no luck. I'm sure lapping and polishing both surfaces only made that bond stronger

Only option I have is trying the memory in the GB board with my old 3200+ Venice, but the problem there is that I can't give the memory the same amount of voltage as the DFI, so it's kinda a moot point :\

So irritating....

 

[Formula350]

Just an update, I decided to try and take these sticks the opposite extreme, not so much of low speed but low timings. What amazed me was these were done on stock voltage and @ DDR416 (yea only 8mhz over, but still heh).

Timings were 1-2-0-8 1T 7-9-0-2-2-1-1 with the bold timings being only the ones that I couldn't put down to their lowest. Or in the case of 8, just didn't want to bother since 8 seemed to result in the best performance. Either 2 dropping to 1 resulted in an instant hang. To be fair though, the last 1 is default value :\ And yes, this was stable, and surprisingly up to DDR422! Which may or may not have been a voltage limitation, as I didn't bother trying for higher at those timings. Performance was nothing even worth mentioning hence me saving it for last. I think I net 5,600mb/s in Sandra, with the rest of my benches being all MHz hungry so being basically stock CPU core (had to reset CMOS trying to do CAS1.5 as the board doesn't like 3.5, 4, 4.5 or 1.5).