DFI 965-S is here! For sale @ ewiz

[nealh]

What I remember of my crucials compared to my g. Skills was their default voltage is 2.2 for their entire Ballistix line. I would not think that under 2.1 would be possibe @ any frequency above PC2-5300. You are simply undervolting your ram, thus it's failing.

ran P95 blend....
well I upped vdmm to 2.2
set ram DDR 952...4:5 divider with 380 fsb
ddr mode3
gtl mode2
fast
fast
dq auto
auto
auto

ran for awhile ..left came back to a black screen and reboot to cmos checksum error...clear cmos with jumoer and starting over again..blah

[Praz]

set ram DDR 952...4:5 divider with 380 fsb

You will probably have to raise voltages for 4:5. The board is not optimized for that divider.

[The Nemesis]

I didn't know you were on the 4:5 divider. With that in mind, what voltage do you have set for the nb? This board struggles to run the 4:5 divider. For most people it won't even boot. When successful, it takes a significant amount of voltage on the nb 1.5v+. This board is optimized for 1:1 up to 500FSB with very little tweaking or voltage. I know it's not what you want but try 1:1 same settings. Also I still don't think the gtl needs to be changed @ such a low clock.

[MADMAX22]

well so far i am having great luck with my board, ram is the cellshock (d9gkx)
settings
mode3 mode 2 fast fast auto auto auto

ran orthos at 500x8 for over 8 hours and good to go, i just installed my new NB cooler the extreme spirit II which dropped my NB temps by 4c under full load, i know most of the guys like longer like 12 to 30 hours orthos but i dont wanna spend the time with that right now, still going up bit by bit to see what i can get stable,




edit: oh and the squealing noise i got when running blend test orthos at these same settings but with large fft's no squealing

[The Nemesis]

Looks good MADMAX22. I have some OCZ DDR2 PC2-8000 Titanium Alpha VX2's on the way. I'm hoping they're d9gkx's. I always do 8 hour runs on a new board until I get to what I think is the max. Especially when I already know what the cpu has done on previous boards. Keep raising that FSB, it lets me know what to shoot for when I get more ram. I'm currently running orthos 7 X 500 1:1 to prove this board's 500FSB is stable without raising voltage.

[nzbleach]

either the board or CPU is limiting me. I did the 266fsb pin mod to my E2140 but the FSB wall still remains at about 410 fsb. Not sure if its the DFI board or the CPU.

[ziddey]

you setting nb strap in the bios to 266?

[doom_st]

Your chips must be based on Micron D9GKX (B6-25E)

for real ! (cellshock pc8000 cl5 week 30 06)

@ madmax22 try this up to ~530fsb vdimm 2.25v

mode3
mode2
fast
fast
dq 5
auto
auto

4
4
4
9
4
29
1
9
9
2
8
6
5
5
4
4

i hope to be stable like mine.

[MADMAX22]

doom_st hey very nice chip you got there

thanks for the ram timings works great initial, will stable test tonight/morning and see if they are stable enough at 505 fsb, seems good so far

[doom_st]

for >515 fsb you have to set cpuvtt 1.42-1.5v in order to prevent random reboots @ orthos blend/3d (if any, due to high cpu frequency)

[nealh]

I didn't know you were on the 4:5 divider. With that in mind, what voltage do you have set for the nb? This board struggles to run the 4:5 divider. For most people it won't even boot. When successful, it takes a significant amount of voltage on the nb 1.5v+. This board is optimized for 1:1 up to 500FSB with very little tweaking or voltage. I know it's not what you want but try 1:1 same settings. Also I still don't think the gtl needs to be changed @ such a low clock.

I just tried 4:5 to DDR 952..as a quick test

I had been 1:1....went back to 1:1

ran P95 for 8hrs..except speedfan gave some weird error message and core 2 was not under full load..weird ..prime still was running on both cores but explorer.exe was using 10% cycles and P95 was getting 90%..so it failed again at

3420 (380x9)...380 fsb 1:1 DDR 760 5-5-5-15-30. with 2.1v(I know not 2.2v..just trying to keep voltage down)....all other ram stuff at auto

it appears to be stable for only about 5hrs...

[The Nemesis]

Okay, now I know my board will do 500FSB stable with no change in voltage. Now to see how far before I need more.

[ziddey]

congrats. wish I could say the same. and again I'll say that that's one hell of a cpu you've got there. one hell of a cpu!!!

[ziddey]

Whoops. Forgot to mention that the screenie is of 6x500. Guess that'd make it easier than to have you figure out actual fsb based on my sig and the "nbcc" given by memset


I don't quite believe it. Just when I was about to throw in the towel too.. I couldn't sleep and so I went at it again. Check this out:


Basically, I did the same old crank up voltages and see what happens approach. This time I even went a little further, with 1.7vsb and 1.5vtt. Still absolutely no dice. Wouldn't even escape post properly. I even tried setting the memory advance features to normal, but still no help.

So, I went and switched everything I could in dram page to auto, except for the default 8665. Then, in the main page, I tried 500fsb and 100pcie again. Nope. No dice.

So, that got me to thinking. What's the one thing I never played with? I've diddled with dq, ddr mode, and all that until my synapses started misfiring. But what didn't I touch?? 8665!!! Change it to 8666 and try again.

Holy moly would you look at that?

I posted that particular screenshot because it's a little peculiar. It seems to be setting timings a little more aggresively than what spd calls for. Now I recall in the past, when I first got the board, it used values of 5-5-5-18 like a good little boy. I think one of the memory enhance options actually may not adjust internal latencies, but sort of advances timings one table in the spd. It may do more than that though.

I noticed the same behavior with my p5b vanilla. When I set static read to fast (or maybe it was faster, can't recall which option it had), it also forwarded timings one table. But I did also notice that when I manually set timings, and just switched around with static read, I still got a little boost in bandwidth.

Subtimings are still seeming to be different for different boards. With the p5b, I'd get 4-28-2-11-9-2-8-4-4. With this, I'm getting 4-50-2-15-9-2-8-4-4. But that's a minor detail.

Now you might be wondering why I'm doing small fft in the screenshot, since blend is the one notorious for locking up right away. I wasn't anticipating cas4, and so only gave 2.1v instead of the 2.2v these sticks need. So blend errors out rather quickly (read, it's an error out instead of a lockup!).

So now it's time to bump up vdimm slightly and see what happens then. I did a quick mbench and saw latency of about 55. When I had memory enhancements to fast, I saw in the mid-upper 40's. When I had them as normal, they were in the low 60's. So this 55 seems to be some sort of middle value. Interesting to say the least!


edit: Just for comparison sake, here's fast/fast 4-4-4-10-4-28-2-11-9-2-8-6-6-5-4-4 vs auto/auto 4-4-4-12-4-50-2-15-9-2-8-6-6-6-4-4

Access latency 47.9 ns (192 clocks)

Read datarate (INT) 8070 Mb/s
Write datarate (INT) 3751 Mb/s

Read datarate (MMX) 8712 Mb/s
Write datarate (MMX) 3773 Mb/s

Read datarate (SSE) 9158 Mb/s
Write datarate (SSE) 9131 Mb/s


Access latency 54.9 ns (220 clocks)

Read datarate (INT) 7137 Mb/s
Write datarate (INT) 3418 Mb/s

Read datarate (MMX) 7979 Mb/s
Write datarate (MMX) 3421 Mb/s

Read datarate (SSE) 8174 Mb/s
Write datarate (SSE) 7890 Mb/s

While the results are indeed incredibly subsatisfactory, it is worthy to compare to a p5b:
4-4-4-4-4-25-2-10-9-3-8-x-x-x-4-4 static read fast(er)

Access latency 60.0 ns (240 clocks)

Read datarate (INT) 6797 Mb/s
Write datarate (INT) 2773 Mb/s

Read datarate (MMX) 7421 Mb/s
Write datarate (MMX) 2767 Mb/s

Read datarate (SSE) 7821 Mb/s
Write datarate (SSE) 9022 Mb/s


Interesting that the p5b did so well for sse write.

But most importantly, if you look simply at the clocks/latency, you'll see what I'm trying to get at.


update: bumped vdimm to 2.2v, and blend is chugging

update:
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.
Torture Test ran 8 minutes 17 seconds - 1 errors, 0 warnings.
Execution halted.

wonder what it could be. vmch bump probably




And now for some humor. normal/normal.

Access latency 66.8 ns (268 clocks)

Read datarate (INT) 6079 Mb/s
Write datarate (INT) 2567 Mb/s

Read datarate (MMX) 6935 Mb/s
Write datarate (MMX) 2600 Mb/s

Read datarate (SSE) 7096 Mb/s
Write datarate (SSE) 7839 Mb/s


I think I might know why it is setting 4-4-4-12 instead of 5-5-5-18. It's 1:1 divider and 266strap, so it looks for 266mhz spd programming. Perhaps when I saw 5-5-5-18, it was at stock, and using 2:3 divider.

[ZhaoYun]

I have spent the last 4 days trying to get this motherboard Orthos blend stable. Tried different sticks of ram and all it does is prolong the hourly wait towards an inevitable error. I have also tried changing all the voltages / timings and still no dice. You name it I've done it. After all this work once everything is up and running, in the end the squealing may not be worth all the hassle. On load, the squeal is louder than all the fans in my case combined. I can't believe I waited weeks for this to go in stock.

[Rattle]

lol I hear ya buddy, mine doesnt squeak but I cant get anything over 367x9 and 400x8 orthos stable for more than 2 hours...

[flyfisher]

I couldn't get mine stable with 2x1GB Firestix PC2 6400 no matter what settings I tried. Prime always failed eventually. So I tried a set of the cheap Super Talent PC2 5400 from Ewiz, set the timings to 5-5-5-15 @2.2v, the rest to auto. Primes all day at 500x7 with 1.35v mch and +137.5mv Vcore, the rest stock. Maybe not the fastest memory but at least it works.

[Rattle]

yeah i think that this mem might be my problem

[nealh]

anyone have issues running Prime95 and speedfan??

seems to be causing issues for me..I went back to SG(for temp monitoring) and Prime95 is still running at 3hrs of blend..speedfan would error out of system would reboot..not sure whether I am having a software conflict

When I run Prime95...I usually have coretemp 0.94, speedfan and windows taskmanager running

[ziddey]

hmm never really thought of that before. i always throw the slew of programs open and put them on my second monitor. sg, sf, ct, setfsb, orthos, memset, cpuz


edit: what if you go down to stock settings? can you run p95 and speedfan together?

[nealh]

hmm never really thought of that before. i always throw the slew of programs open and put them on my second monitor. sg, sf, ct, setfsb, orthos, memset, cpuz


edit: what if you go down to stock settings? can you run p95 and speedfan together?

never tried stock(like 30min)for long..generally crashes seems to occur after 30min to 1hr or so

the odd thing is when speedfan crashes ..it gives some exception error..did this 3 times(I figured it was simply a voltage issue on chipset/vtt/ram/cpu)..but sometimes, the system just does a hard-reboot

I am still not sure this is a real issue but it maybe...was wondering if anyone else had problems...I am using speedfan 4.32

it could be coretemp and speedfan together....

I am at 4hrs of Prime95 v251 blend at e6600@ 3420(8x380) 1:1 w/ Ballistix at DDR760 4-4-4-12 @ 2.2v...set auto-auto-fast-fast-auto-auto
nb 1.30/vtt 1.275/cpu +175mv...these were tested earlier to day with speefan open and it reboot....now no speedfan and using SG/coretemp 0.94/windows taskmanager...going strong at 4hrs

Hmm...I may have to reassess all my voltages???

wonder if speedfan is conflicting with something

[ziddey]

I'm noticing some anomalies. Currently, I'm at 450fsb fast/fast. This is what I get from mbench:


Access latency 53.9 ns (195 clocks)

Read datarate (INT) 7174 Mb/s
Write datarate (INT) 2973 Mb/s

Read datarate (MMX) 7799 Mb/s
Write datarate (MMX) 2982 Mb/s

Read datarate (SSE) 8167 Mb/s
Write datarate (SSE) 8066 Mb/s

Now, let's drop the multi from 8x to 7x with rmclock and continue to boost fsb up to 500 with setfsb. Run mbench again:

Access latency 55.8 ns (202 clocks)

Read datarate (INT) 6843 Mb/s
Write datarate (INT) 3076 Mb/s

Read datarate (MMX) 7429 Mb/s
Write datarate (MMX) 3085 Mb/s

Read datarate (SSE) 7750 Mb/s
Write datarate (SSE) 8086 Mb/s

Got worse???

[The Nemesis]

That's probably the effect of 100 Mhz less overall clock speed. I'm sure I can run the same test with similar results. I will if you like.

[ziddey]

Well I mean, when I was able to boot 7x500ff, or even 6x500ff, I saw:


Access latency 47.9 ns (192 clocks)

Read datarate (INT) 8070 Mb/s
Write datarate (INT) 3751 Mb/s

Read datarate (MMX) 8712 Mb/s
Write datarate (MMX) 3773 Mb/s

Read datarate (SSE) 9158 Mb/s
Write datarate (SSE) 9131 Mb/s

Monster latency/bandwidth.

I manipulate multiplier in windows with rmclock, while I use setfsb to change around. The actual cpu speed doesn't really affect memory bandwidth much at all. However, that said, I do see a very clear drop in memory performance as I'm scaling fsb up. Regardless, I'm at a loss of words for 500fsb. Just can't get it stable anymore

Hopefully we see some p35 boards with memory enhancement features out soon.

[The Nemesis]

You got me curious so I slap the system back together for a quick run.

Intel P6 processor (CPUID = 6f6) @ 3508 MHz 7X 500
Instruction set support : MMX SSE SSE2

Access latency 49.9 ns (200 clocks)

Read datarate (INT) 7652 Mb/s
Write datarate (INT) 3615 Mb/s

Read datarate (MMX) 8428 Mb/s
Write datarate (MMX) 3656 Mb/s

Read datarate (SSE) 8907 Mb/s
Write datarate (SSE) 8965 Mb/s




Intel P6 processor (CPUID = 6f6) @ 3805.4MHz 8 X 475
Instruction set support : MMX SSE SSE2

Access latency 52.3 ns (199 clocks)

Read datarate (INT) 7601 Mb/s
Write datarate (INT) 3434 Mb/s

Read datarate (MMX) 8207 Mb/s
Write datarate (MMX) 3444 Mb/s

Read datarate (SSE) 8621 Mb/s
Write datarate (SSE) 8548 Mb/s


Intel P6 processor (CPUID = 6f6) @ 4009.1MHz 8 X 501
Instruction set support : MMX SSE SSE2

Access latency 48.1 ns (193 clocks)

Read datarate (INT) 8122 Mb/s
Write datarate (INT) 3619 Mb/s

Read datarate (MMX) 8858 Mb/s
Write datarate (MMX) 3631 Mb/s

Read datarate (SSE) 9138 Mb/s
Write datarate (SSE) 9024 Mb/s

I left the memory timings the same for each run & boot from bios. No memset etc.

auto auto fast fast auto auto auto 4-4-4-12-4-30-2-10-10-2-8-6-5-5-4-4