crazy Venice divider issue, please help!

[ErikaeanLogic]

I just received a Venice yesterday, week 0517 APJW. I plopped it into my DFI Ultra-D and let her rip, quite hopeful that I would hit the 2.7GHz mark with the supplied stock cooler. I left the default multi and cranked the fsb to 300, dropped the HTT multi to 3x. Nothing. Cranked voltage as high as 1.7V. Nada.

Actually, before I continue let me post some system spec's:

A64 3000+ 0517 APJW
DFI Ultra-D, bios 5/10-3
PCP&C 510SLI
2GB (2x1GB) CorsairXMS3200C2
x850xt
74gb raptor
various other drives

Okay, so I'm disappointed that I can't do 300fsb. Well, I can't do 290, 285, or 280 either! I've reduced the memory multiplier to 1/2 (ddr200) even though I intend to run the memory at or around 200fsb (the unfortunate side effect of running 2gb). I begin to smell a rat because the Winnie I just took out of here wouldn't budge over 266fsb no matter how many volts I threw at it. So I began to do some math and tried other dividers. Nothing got me into Windows at over 2.6GHz. Until, that is, I tried 8.5 x 318! Now I was about all out of options after tweaking every LDT/HTT/VID and memory tweak in the bios. I had always thought of the .5 dividers as being off limits. But the first thing I did (besides be happy that I was in Windows) is run CPU-Z to verify that I was running 2703MHz @ 1.5V, and that the memory was chugging along at 208fsb (2.5-3-3-8 timings). I ran SuperPi 1M and got 32 seconds, which is a new record for me. I ran two instances of Prine95 (using small FFT's) for over 6 hours with no instability or crash. I played Call of Duty for a while. Then, I go into Doom3 to bench my new framerate and I kept getting "memory could not be referenced" errors. So I loop MemTest test 5 for a while and get zero errors. I bumped the cpuVID to 1.6V but that didn't work. I tried playing with the dividers/fsb again but to no avail. Before I went to bed last night I set the system back to 8.5x318 and ran Prime95, SuperPi 32M, and looped 3DMark2001 test 4 (Nature). When I woke up this morning the SuperPi had completed, 3DMark was fine, and Prime95 is still running. Wtf?

Please help! I can't figure out why i can do 8.5x318 but not 9x300. What do these "memory could not be referenced" errors in Doom3 mean? Can the 1gb sticks be holding me back somehow? (I photoshop alot and can't stand when I run out of physical memory so I run 2gb) I have scoured all the Venice threads for hours but could not find any issues remotely similar to these but I'd love to hear from you guys, thanks.

Erik

[OjorisO]

I just received a Venice yesterday, week 0517 APJW. I plopped it into my DFI Ultra-D and let her rip, quite hopeful that I would hit the 2.7GHz mark with the supplied stock cooler. I left the default multi and cranked the fsb to 300, dropped the HTT multi to 3x. Nothing. Cranked voltage as high as 1.7V. Nada.

Actually, before I continue let me post some system spec's:

A64 3000+ 0517 APJW
DFI Ultra-D, bios 5/10-3
PCP&C 510SLI
2GB (2x1GB) CorsairXMS3200C2
x850xt
74gb raptor
various other drives

Okay, so I'm disappointed that I can't do 300fsb. Well, I can't do 290, 285, or 280 either! I've reduced the memory multiplier to 1/2 (ddr200) even though I intend to run the memory at or around 200fsb (the unfortunate side effect of running 2gb). I begin to smell a rat because the Winnie I just took out of here wouldn't budge over 266fsb no matter how many volts I threw at it. So I began to do some math and tried other dividers. Nothing got me into Windows at over 2.6GHz. Until, that is, I tried 8.5 x 318! Now I was about all out of options after tweaking every LDT/HTT/VID and memory tweak in the bios. I had always thought of the .5 dividers as being off limits. But the first thing I did (besides be happy that I was in Windows) is run CPU-Z to verify that I was running 2703MHz @ 1.5V, and that the memory was chugging along at 208fsb (2.5-3-3-8 timings). I ran SuperPi 1M and got 32 seconds, which is a new record for me. I ran two instances of Prine95 (using small FFT's) for over 6 hours with no instability or crash. I played Call of Duty for a while. Then, I go into Doom3 to bench my new framerate and I kept getting "memory could not be referenced" errors. So I loop MemTest test 5 for a while and get zero errors. I bumped the cpuVID to 1.6V but that didn't work. I tried playing with the dividers/fsb again but to no avail. Before I went to bed last night I set the system back to 8.5x318 and ran Prime95, SuperPi 32M, and looped 3DMark2001 test 4 (Nature). When I woke up this morning the SuperPi had completed, 3DMark was fine, and Prime95 is still running. Wtf?

Please help! I can't figure out why i can do 8.5x318 but not 9x300. What do these "memory could not be referenced" errors in Doom3 mean? Can the 1gb sticks be holding me back somehow? (I photoshop alot and can't stand when I run out of physical memory so I run 2gb) I have scoured all the Venice threads for hours but could not find any issues remotely similar to these but I'd love to hear from you guys, thanks.

Erik

Erhm you should know this board has trouble with *some* dividers.

I could loop memtest for hours with a 180 divider, yet any game I would try to play hung up after a while.
The .5 multi's don't have thesame memory bandwith as the x.0 multi's, perhaps that explains why.

Also your memory, erhm, isn't the best around

Try disabling Bank Interleaving, it should help stability with high HTT's and put the ldt multi @ auto, that's all I can think of right now except getting some TCCD and running it in sync

[Jimbo Mahoney]

Also, try editing your boot.ini to turn off DEP (I'm assuming you are running XP SP2):

change the bit where it says:

/noexecute=OptIn

to this:

/noexecute=AlwaysOff


Be careful of capitals.

Let me know if that solves any boot issues. I'm hoping it will because I seem to be one of the only people that this helped with! (OK, so there are at least two others I know of, but still!)

[ErikaeanLogic]

OjorisO: thx for the suggestions, tried 'em all and still the same thing. . .

Jimbo: ya, already got that one in play

update: got Doom3 to run the timedemo without crapping out now, will try to play later this evening. . .but I'd still like to figure out how to deal with this multi thing since I believe this cpu's got more in her and I'm close to the mem limit for the 8.5 multi I'm on now. . .

more suggestions, please?

[TheMeatFrog]

Divider problems here also, it's just part of this boards nature

[eva2000]

i partially solved my divider issues with 3000+ 0516 GPAW Venice cpu in 2nd half of the post at http://i4memory.com/showthread.php?t=573

max async latency = 8ns
read preamble = 5.5ns
TREF = 0648
drive strength = 3
drive data strength = 1

try it

[TheMeatFrog]

i partially solved my divider issues with 3000+ 0516 GPAW Venice cpu in 2nd half of the post at http://i4memory.com/showthread.php?t=573

max async latency = 8ns
read preamble = 5.5ns
TREF = 0648
drive strength = 3
drive data strength = 1

try it

Excellent! I just tried these settings and they seem to be helping! I'll report back later after some more testing

[eva2000]

Ah good to hear.. keep us informed

[TMM]

The .5 multi's don't have thesame memory bandwith as the x.0 multi's, perhaps that explains why.

seriously?

8.5x310 "166" mem divider ( = 239mhz mem), 2.5-3-3-7 i get 7020/6950~ in sandra,

yet with 10*239, "200" mem divider, 2.5-3-3-7 i can't break 6800/6700


is this because the increased CPU speed (or htt) or am i right in saying that 8.5x gives better bandwidth on my system?

[blackjok3r]

the .5 multies use a mem divider, if the cpu is clocked higher then the mem bandwidth will be higher but the ram will not be at 1:1 so to speak.

[TMM]

Yeah, they use the "set" of dividers from the "full" multi above them - so to speak.

e.g.
9x291 1:1 = 2626mhz, 291mhz memory (CPU/9)

8.5x309 1:1 = 2626mhz, 291mhz memory (CPU/9)

theres no loss in bandwidth, you just have to know what CPU/* memory divider it's going to use.

[ErikaeanLogic]

i partially solved my divider issues with 3000+ 0516 GPAW Venice cpu in 2nd half of the post at http://i4memory.com/showthread.php?t=573

max async latency = 8ns
read preamble = 5.5ns
TREF = 0648
drive strength = 3
drive data strength = 1

try it

hi eva, long time no see . may I ask what your methodology was for finding those settings and getting them to work for you? is there a thread you could link me to where you go through the process? thx

I tried your setting recommendations and did not get into Windows again, BUT. . .I did get past the "BACKUP BIOS...OK!" text, which I couldn't do before. So I changed the "drive data strength" to 3 and that almost worked so I went back into my 318x8.5 Windows boot and opened A64 Tweaker to see where all of the settings I had set to "auto" were resting in a Windows boot that worked. I then went back into the bios and changed the speed settings to 300x9 and into the memory screen and statically set all of the memory settings to what I had copied down out of A64 Tweaker. I noticed that there were still 2 settings that were set to "auto" after all was said and done: DRAM RESPONSE TIME and WRITE CAS LATENCY. After many cmos clearings I found that WRITE CAS LATENCY works when set to "1" and I got into Windows at 300x9! It was still unstable, though, so I went back into the bios and changed DRAM RESPONSE TIME from "fast" to "normal" and now I could Prime in Windows!!

New issue: performance was not up to par of the 318x8.5 (2703MHz) when I was in at 300x9 (2700MHz). The memory divider was still set to 2/3 (DDR266) so at 300fsb it ought to have been an even 200MHz (DDR400) on the memory, right? Wrong! It came out to about 192MHz (DDR384) and that pretty much meant the difference of about 1sec in a 1M run of SuperPi. I cranked up to 311x9 (2800MHz) but had to crank the voltage to 1.663Vcore to do so. I did get 31sec at my 1M run at SuperPi (besting my 32sec at 318x8.5) but nothing I did (more Vcore, etc.) could keep me Prime-ing for more than a few minutes. Not sure if it's a cpu limit or a chipset/memory issue, to many gaps in my data but it could be either. I've decided to go back to the 8.5x multi for now at 322x8.5 (2737MHz) @1.6Vcore, 2.5-3-2-7 mem @211MHz, and running 2 instances of Prime now as I type this. When Gigabyte releases that iDrive thingy next month I can go back to 1GB of system memory and my bottlenecks will widen so that I can narrow either the cpu, chipset, and/or memory as my main issues. Still, 2737MHz out of a 1800MHz cpu ain't too shabby

[groovetek]

some of u xs guys should join i4memory

we need some non-aussie members, haha...

[eva2000]

hi eva, long time no see . may I ask what your methodology was for finding those settings and getting them to work for you? is there a thread you could link me to where you go through the process? thx

hey there.. i did it through trial and error but used pcmark04 custom test i devised listed at ** yes i4memory.com http://i4memory.com/showthread.php?t=474

some settings in bios will BSOD on boot, but ones that get into windows at least i use this test..

of course i did all the outlined testing first in memtest as outlined at http://i4memory.com/showthread.php?t=327

[ErikaeanLogic]

thanks for the response, e

update2: as OjorisO predicted, I Prime'd for hours yesterday and immediately began having gaming stability issues with the divider set to DDR266. I went back to clock settings I knew to be possible with my old Winnie and still had stability. . .and realized that the 3 settings which had made using the 9x300 possible in the first place were the culprit! I reset WRITE CAS LATENCY, DRIVE STRENGTH, and DATA DRIVE STRENGTH back to "auto" and all is well (for 277fsb, that is).

Can anyone recommend good sub-3.2Vdimm memory that costs less than $200/gb? I'm going back to 512x2 for now so that I can run 1:1 to see if that can help remove the performance/stability bottleneck.

[perry_78]

I had the same issue. I had a 133 divider, and my venice just sucked. Then I found out why. The divider doesnt seem to work too well, and I had my trusty BH5 running 240 @ 2-2-2 wikth 2,87V

Now I'm trying out a better divider, pray for my sake

[ErikaeanLogic]

I had the same issue. I had a 133 divider, and my venice just sucked. Then I found out why. The divider doesnt seem to work too well, and I had my trusty BH5 running 240 @ 2-2-2 wikth 2,87V

Now I'm trying out a better divider, pray for my sake

My prayers are with you . I am going to order some TCCD memory today, hopefully will have it tomorrow. Just can't figure out which stuff to get. . .

[perry_78]

Thanks Doesn't seem to change the fact that my Venice sucks terribly, but I'll give it some serious testing before I complete the RMA.

[douirc]

ErikaeanLogic - so where you at with your OC? you get your new TCCD? what did you order? you able to run 300*9? I have an identical proc in a UT SLI-DR and am hitting a wall at 290*9. I'm gonna follow what you've written so far and see if I can get 300*9, but wanted to see where you're running.

thx!

[ErikaeanLogic]

okay, you're not gonna like this. . .I'm clocked way back because I got tired of throwing moneez at this problem. I'm getting an X2 anyway in a few weeks so this won't matter since I'll have the 11x multi to work with. The guys at OCZ tech support say that alot of Venice's have a 285-ish fsb cap, so the 3000+ can really be hit and miss. At one point I even had the system Prime95 stable for like 24 hours but as soon as I entered a game I would either reboot automatically, bluescreen, or both. I shoulda spent the extra $50 on the 3200+. Good luck, man!