Printable View
my 2x1gb crucial ballistix ddr2 800 2.2v wont scale as well as i would like.
best i have done so far is 446.4mhz 5:6 ratio 4-4-4-4-12-6-42-9-13-10-4-8-5-4-3120t-2t at 2.18v
super pi 1m 13.875s
is one of the subtimings too tight? maybe twtr which i left at 10? or trrd which i set to 4? or one of the other subtimings? or maybe d9s are incapable of scaling high on gigabyte boards, particularly mine? please help. btw if i posted this in the wrong part of the forum let me know :)
Whats your vNB and vPLL? Is the CPU set up to ensure the RAM is the only thing that could be holding you back?
Tell us everything about your settings :D
all voltages other than ddr2 are at stock, vmch is at stock i think thats the nb voltage, and i dont think my board even has a setting for pll voltage
cpu is at 3.35ghz im positive its the memory that caused the bluescreen when i tried to raise the fsb beyond 372 fsb
edit: ddr2 voltage set to 2.15v my board overvolts it to 2.18-2.19v
pci-e voltage at stock
fsb voltage at stock
mch voltage at stock
cpu voltage at stock
there are no other voltages for my motherboard with f12 bios
try vMCH to 1.35v and see what happens :) and vFSB to 1.26v (1.30v is generally the safe limit for P35)
my board has the option to increase voltage in 0.05 increments it doesnt tell me what the default voltage is, the exception is cpu voltage which has much smaller increments
edit: are you sure none of my subtimings are too tight? see first post
edit2:vmch goes up in 0.025 increments
edit3: i saw one guy used 4-4-4-4-8-7-39-9-13-14-7-9-7-8 on his crucial 10th anniversary sticks, those subtimings are way looser than mine
Double sided or single-sided?
if single-sided then 446 cl4 is about the max what theyd do , you should try them with 5-5-5 , should get 550+
if double-sided you can try another divider, 2:3 or 5:8 for instance , if they cant hit 500 4-4-4 2.2v even suicide then must be super bad batch
THats it! I knew there was something I should say about dual or single sided but I wasnt awake enough to follow the train of thought :p:
Single sided 1GB DDR2 sticks suck :p:
Thought FSB termination was CPU related in terms of harm which might come from using >1.35v ?Quote:
Originally Posted by K404
Anyhow,what does sticker on modules say?
If it ends with .16FD3,then you are doing something wrong.
If it ends with .8FES, than it this is not your lucky day,and they are scaling just fine with regards to their capabilities.
Like Žile says look at your sticker and find out what do you really have. If it's .16FD3 than you have Micron D9GMH or maybe GKX/GCT chips and they must go more than you get now. For example, I have .16FD3 with GMH chips (Ballistix Tracer 800) and they can go up to 1250+ MHz.
The situation doesnt make sense. Asus and Gigabyte revised their safe limits for P45 and most users noticed FSB scaling up to 1.6v+ for vFSB. P35, X38, X48 didnt behave that way at allQuote:
Originally Posted by žile
Maybe that's your memory, there isn't any guarantee that you memory has to OC well or mad ! or is there ? :D:D:D:D
If it is 16FD3 or 16FD5, it is. :)
i tried loosening the subtimings to whats at the end of post #5, made no difference whatsoever. still bluescreens just sitting on the windows desktop if i try going from 372 fsb to 374 fsb, its bl12864AA804.16fa
at 372 fsb its super pi 1m stable
i use setfsb to overclock atm since i cant set the subtimings in bios i need memset 4.0 for that as well
board used to scale up to 400fsb with an e6850 and i only stopped there due to cpu limitations on the stock cooler, so i cant imagine running into a fsb wall already, i guess ithe only thing to try is 5-5-5 timings, if it doesnt scale with that somethings wrong. btw i still havent changed any of the voltages yet, and ddr2 voltage still at 2.18v
btw this board actually scaled higher with powerchip ics in the past, 2x2gb ocz reaperX ddr2 800 and it scaled past 500 no problem at 5-5-5 timings, none of the subtimings were ever changed on those
16FA?! and no number behind the FA?
Just googled them but couldn't find any info on what chips are used. If they were the chips known as "D9" by most ( D9GMH/D9GKX ) they should react to voltage, or those were the first clock that bad.
Try to loosen subtimings, these are very likely some other Micron chips not capable of tight timings.
Can you post a picture of CPU-Z with the SPD tab open?
i believe .16fa is even older than .16fd , and what exactly are you looking for in cpuz spd?
and there is nothing after the .16fa
i already tried loosening subtimings and it made no difference, see the end of post #5 to see what i tried, except i used tras at 12 and trfc at 42, which is the default for my ram apparently
Okay, I looked into it a tad further, you're right, they are way older than FD sticks. Just forget what I told you.... :D
I think I know what chips they have, they should be using fatbody chips, the D9 chips that were famous before D9GMH. Damn, I really had to think about their code for a second, haven't used my kits for over a year. D9DQW or D9DQT it is.
Were there any other chips in between those two series?
Try raising voltages a tad to 2.20V, and are you memtesting before entering Windows?
if i set ddr2 voltage to 2.2v it overvolts to 2.26v and its currently set to 2.15 overvolts to 2.18 so im really worried about going overboard with voltage
so in theory these can probably do 3-2-2 timings but thats of little use to me, i want them to scale
bios settings are pretty much at stock except ddr2 voltage set to 2.15v and c1e, tm2 and eist disabled to avoid cpu throttling, all my overclocking is done in windows with setfsb and memset 4.0
any tips on what subtimings to use for fatbody chips?
perhaps these scale with timings such as 5-4-3 or am i mistaken?
That's some serious overvolting... let's wait for some more opinions, though I'm pretty certain there were no other chips that could run that speed at that time.Quote:
Originally Posted by RAMMAN
I looked into the fat body list and found the P/n, but without .xxx in the end. :(
You'd need an excess of voltage to do that, reasonable limits with 2.25V should be ~480-490MHz 4-4-4-12 and 530MHz CL5-5-5-15. 5-4-4-15 should yield better performance at slightly lower clocks.
even my value D9GKX are doing 4-4-4-12 490MHz 2.18V
nothing especial :)
Crucials are the very best for real
if you mean 24/7
you can have:
~850 3-3-3-8 scalable with 2.25-2.35
1066 4-4-4-12 scalable with 2.25-2.35
1300+ 5-5-5-18 scalable with 2.25-2.35
and everythign stays in green voltages
2.35 is nothing especial for D9GXX Chips
even 2.45 are green for them
somedays OCZ had ddr2 1400 ram on the market and they were with 2.45V but never released to public
btw these are the sticks that are known for not booting at spd settings, they default to 400 4-4-4-18 1.8v
they need 2.0v to get stable at stock settings
because thats wrong SPD programmed
ddr2 800 D9GXX Rams need 2.0+ if they run 4-4-4-12 and 2.2+ if they run Cas3
D9GKX are best if you stay under 2.5V or they are equal to handpicked D9GMH
but if you go over 2.5+ D9GMH are scaling even with 3.4V and GKX arent
im thinking of getting something that i know will scale, like team xtreem 10400 but i guess they are as rare as hens teeth
btw dont forget to check my earlier posts, i tend to edit them quite a bit
cpuz spd mentions the serial number, will that be of any use to anyone?
Sorry, I can't help you concerning subtimings, the last time I used them I was on my Striker Extreme with 680i. I never tested them on an Intel chipset.Quote:
Originally Posted by RAMMAN
You missed the fact that those sticks are most likely no D9Gxx chips.... :rolleyes:Quote:
Originally Posted by westsider
These sticks will scale too, just with even higher voltages than D9GMH.... ;)Quote:
Originally Posted by RAMMAN
what are these rams?
older then FD3 and FD5?
maybe Fatbody D9 :)
if so ,youll need more voltage
cpuz spd says serial number:840034c6
jedec #1 400mhz 4.0-4-4-18 trc 24 1.8v
please make a screen from spd table :)
Quote:
Originally Posted by westsider
I really don't mean to bother you, but did you actually take the time to read the thread?Quote:
Originally Posted by westsider
@RAMMAN: serial number is different on all sticks ;)
It's not P/N
my last post contains everything of interest in cpuz spd, all the other fields are greyed out
however, the stick in stick in slot # 3 has serial number 8a003d7d
only other stuff in cpuz spd that are not greyed out are the obvious, like crucial technology, 1024mb module size , pc26400, correction none, ddr2
spd timings table only has jedec #1 400mhz 4.0-4-4-18-24-1.8v
Yeah, as I said already, serial number is different on all sticks.Quote:
Originally Posted by RAMMAN
I'm pretty certain they use fat body D9, maybe some of the other old member that have such a kit will post and confirm my guess.
You can use MemSet to get a glance at the SPD with all timings and subtimings.
to save time can you tell me how to post screenshots in forums?
spd.bmp (471 KB)
When you post, there is a second part after the box where you write your message, it says "Additional Option", there you can find the button "Manage Attachments". If your screenshot is bigger than 200kB you'll have to upload to an imagehost service.
only persons i know of that have done extensive fatbody tweaking on intel systems are fugger and hipro5 i think, there maybe more that i dont know of
check post #32, it has link to entire spd table in memset, apparently there is cas2 in jedec#2, i thought tighter than cas3 is impossible with ddr2 lol
That SPD is funny indeed! :D
Subtimings at stock speed are quite tight, those must be using fat body chips, I can't think of anything else.
It might be worth to check out their old threads, here is another one I remembered that was posted not soooo long ago.
Just get some paper, a pen and Memtest and you'll figure your memory out! Have fun! :up:
so these respond to voltage, not loose timings?
the memory in your link are 5400ul 2x512, wont they behave differently to crucial ballistix 2x1024 even if they both use the same type of ics?
all d9 dont mind tight subtimings
and your fatties should do 1000mhz cas 4 easy just need volts
you gonna be pushin 2.3 v
my fatties actually prefered cas3
i did pump upto 2,7 into them and they still alive lol
its quite possible tou have d9gct and they need serious volts to scale
anything over 2.2v 24/7 for d9 is a risk;)
If it's the same chip, the bigger capacity sets will just stop scaling sooner than the smaller kits.Quote:
Originally Posted by RAMMAN
Yes, they do react to voltage, lots of voltage, if they scale any further at the same timings but with a notch more VDIMM, they should be some sort of D9, fat bodies are highly probable. Just remember to put a fan over them when pushing up voltage.
I'm off to do some work now, if you need more help I'll be online again tomorrow.
Or just listen to zsamz_, he's a good guy, his only problem is that he doesn't like Gigabyte boards... :p:
Don't you think those are fatties?
heres a link to what my mobo actually defaults my memory to, it might be of some use
memset.bmp (515 KB)
i guess later on i will try setting 2.25v in bios and hope my board doesnt overvolt it much past 2.3v, i already know 2.15 = 2.18 and 2.2 = 2.26
if the chips are big (19x13mm, take the most of pcb) then those are fatbody for sure
if you can't visually detect them - test them, fatbody hits roughly same clocks with 5-5-5 as with 5-3-3
450-500 with 4-3-3 (~2.3v) or with 5-3-3 (~2.2v) is the best daily fatbody setting
could it be my mobo just doesnt like my memory? i heard d9s especially fatties dont like gigabyte boards
i find it annoying to no end that cas latency can only be changed in bios :/
edit: i set ddr2 voltage to 2.2v, which overvolts to 2.26v and increased cas latency to 5 in bios. i set main timings to 5-3-3-12 with memset 4.0 didnt touch any subtimings and it runs just fine at 400mhz, now i just need to start raising the fsb.
so if it scales anywhere near 450 its definitaly fatbody?
check post #188 page 8
http://www.xtremesystems.org/forums/...t=77442&page=8
according to that .16fa is d9dqw thats fatbody right?
http://ramlist.i4memory.com/ddr2/
according to this d9dqw is one of the 4 possibilities
i bought them mid 2006 off ebay, cant remember for how much but it did cost me an arm and a leg at the time. i didnt think much of them as they didnt scale much at the rated voltage and timings, so i switched to some ocz reaperX with powerchip ics as soon as they became available but i kept the crucial ballistix around. recently i sold off the ocz reaperX and went back to the crucial ballistix
I picked up some 16FD5 that was in a HUGE batch of 8FE5, and it clocks horribly. It clocks like the single sided stuff +20mhz or so. Doesn't respond well to voltage either... My other set does 1200 easy w/2.25v
good news! my memory actually scaled higher with 5-3-3-12 and default subtimings 2.26v 450.5mhz qx9650 3378mhz super pi 1m 13.750s
super_pi_1m_13_750s.bmp (2.7 MB)
if it actually clocks higher at 5-3-3-12 and a slight voltage bump its got to be d9dqw right?
it could possibly have more in it, i didnt push the fsb until it bluescreened this time because i was so glad to achieve 450mhz at tight timings
this memory has been to this voltage in the past and never reached this speed, perhaps it just likes high volts, loose cas and everything else tight
after close inspection without taking off the heatspreaders the memory chips appear to take up about 2 thirds of the pcb. that and they seem to clock just as well with x-3-3-x timings im positive they are d9dqt/d9dqw aka fatbodies.
btw it has a black pcb making it hard to see, luckily i had a flashlight
edit: heres something from anandtech same memory from early 2006 they are d9dqt, if you go to page 2 they have .16fa on the heatspreaders same as mine
http://www.anandtech.com/memory/showdoc.aspx?i=2700&p=1
edit2: heres a review of .16fa sticks doing 3-2-2-8 timings but at low clocks
http://www.xbitlabs.com/articles/mem...dr2-800_3.html
Those pretty much confirms they are fat body chips, have fun with them and pump up the volts! :up:
With more volts you might be able to get them to 500MHz CL4-3-3-10, which is still a rather nice speed.
super_pi_sub_14_sec.bmp (2.7 MB)
4-3-3-10 is already a reality with tight subtimings (the same ones in the link you showed me)
http://www.xtremesystems.org/forums/...d.php?t=161861
only reason i didnt go further was because my cpu was getting too hot, i guess i will have to either lower the multiplier and say goodbye to any chance of beating my current super pi 1m time or change the fsb/memory ratio and keep the memory voltage up 24/7. or get a better cpu cooler. anyway i had my voltage set to 2.25v but the bios and everest were both reading 2.19v, same as if i set it to 2.15v. since it never scaled like that in the past i dont believe it for a second.
edit: just how hard is it to get fatbodies to work with 4-2-2-10 timings at 450+(900mhz effective) ? does it require excessive voltage?
I didn't notice you edited your post until I got your PM, sorry. :(
1M is not a good test for memory, it fits into the CPU's cache and your score will not change much from running looser timings. If you want to see the memory performance, run 32M or Everest bandwidth tests.
For 24/7 I would go for some more CPU speed to be honest, 3.4GHz is not really much. And getting into FSB or temp problems when testing memory is often the case when you run a quad, that's why I use a E8600. ;)
its summer over here and im running the qx9650 on the stock cooler for now, at 3.4ghz after the super pi 1m bench i tried running orthos, it never became unstable however i had realtemp running at the same time, after 2 minutes i stopped orthos and rebooted, setting everything except memory voltage back too stock, which i then set from 2.25 to 2.15 in bios. i just freaked out when i saw the qx9650's hottest core hit 85c!
i use a quad because i also like benching 3dmark06
I forgot you're having summer now at the other end of the world... but get a decent cooler man, such an expensive CPU on stock heatsink. :shakes: ;)Quote:
Originally Posted by RAMMAN
Get a TRUE or a Megahalems, both are great coolers! :)
i keep forgetting its an expensive cpu, since i got it for about the same price as a q9650 and most enthusiasts have moved on to i7 systems
edit: see the koolance pot in my avatar? i plan on doing some dry ice benching in the future but i must admit i do need something better than the stock cooler for 24/7
edit2: interesting, at 460mhz(920 effective) at 4-3-3-10 2.26v it locked up, not bluescreened thats usually caused by subtimings being too tight isnt it?
edit3: this is dissapointing, at 933 4-3-3-10 default subtimings 2.3v it bluescreened after super pi 1m finished but before i could save the result :(
edit4: keep in mind my 2x2gb ocz reaperX with powerchip ic's used to reach 1000 no problem with 5-5-5 timings, i dont have them anymore though
Yeah, you really need something better for 24/7!
These kits need voltage to shine, 2.3V is rather low, but the board might be the culprit here, P35 aren't known for clocking the FSB of quads up.
If you want high FSB on your quad you'd have to go with the Gigabyte P45 UD3 series.
im pretty sure its memory limitation not fsb limitation, i wouldnt be surprised if my memory clocked better on an asus board but i dont think its a fsb limitation here
one last option i could try is up the fsb and mch voltage by 0.1 but im not comfortable upping the ddr2 voltage beyond 2.3v
i pretty much pulled out all the stops in my last run, at least by my standards anyway :)
everest.bmp (789 KB)
+0.1v fsb
+0.1v mch
+0.50v ddr2 (overvolts to 2.35v, nice to see it overvolts by 0.05 consistently even this high)
i even lowered multiplier to 8 to eliminate any possibility of the cpu limiting the overclock
946 4-3-3-12 default subtimings
i got greedy and tried to go from 394 to 396 to 398 to 400fsb with setfsb before trying another everest bench, it bluescreened soon after i set 400fsb
edit: memory speed seems to scale about 6.5 (13mhz effective) every time ddr2 voltage gets bumped up 0.05v. while it is possible i might have run into a fsb wall i doubt it and i will be sorely dissapointed if my board stops just short of 400 fsb. btw +0.50 ddr2 voltage is where it starts to flash red in bios, but i guess it was inevitable :D
edit2: for when i get my team xtreem 1300 i wonder if my board can handle running a different fsb/mem ratio. so it goes to 1066 when the fsb is at 333, its currently at 2.4 i guess that means i will have to change it to 3.2 if i want to have any chance of the new memory reaching anywhere near its potential. the board is rated for 1066 ram so my board should be able to run that ratio right?
i changed the memory divider to 1:1 to test the fsb, its orthos stable at 400fsb so my board isnt limiting my fatbodies overclock, obviously team xtreem 1300s are a different story. only way to scale anywhere near that far is with 5:8 i think. im not game to raise the fsb further because even the highest fsb on hwbot with my board and cpu(or any quad and my board for that matter) is 413fsb, and when i get better cpu cooling i will just raise the cpu multi and change memory divider to go further. everest bandwidth will suffer due to low fsb but it cant be helped with this board :(
edit: heres a list of the best known results achieved on my board:
http://hwbot.org/hardware/motherboar...b=hall of fame
check #1 3dmark06, looks like my board+qx9650+dice has already been done, but theres still sub 10 sec super pi 1m(should be a cakewalk with my system+dice) and 100k 3dmark 01(very difficult, impossible with my current videocard i think.) pitty we dont get points for getting high scores on older boards.
You'll need a better board to push your quad to high FSB and high memory, look for Gigabyte's UD3 series! :up:
Or you can run your TG-sticks at low memory speed with tight timings, though high memory speed with looser timings will beat it bandwidth-wise.
another way of really pushing team xtreem 1300 ram with this board is 2:3 ratio with 4-4-4 timings i think
1000 4-4-4-12 should work without having to go past 2.2-2.3v right?
then i can raise the fsb and see how close i can get to 1200 4-4-4 with higher voltage
edit: according to the #1 3dmark vantage score with my mobo on hwbot 1280 memory speed is possible with 5:8 memory divider with a quad, its just that with a quad im limited to 400fsb while with a dual 500+fsb is possible(with great difficulty).im definitaly not changing cpu, i need a quad for 3dmark06 and i will be needing the unlocked multiplier if im to have any hope of scaling to 4.5-5ghz on dice in the future anyway and i cant afford a new board atm.
edit2: i think if i was to get a new board i would be looking for an asus board with p45 chipset
500Mhz CL4 is easy and should be doable with "low" volts, but 600MHz will require some serious increase in VDIMM I would not want to run 24/7.
Mate, you don't need another CPU, you need another board! :p:
Try to get your RAM to 650+MHz CL5 and run a low PL, that should give you the best performance!
i plan on dice benching early next year, im not exactly looking for 24/7 settings lol. though i do drop the ddr2 voltage back to 2.2v and fsb to the default 333 after every benchmarking attempt with my fatties atm
edit: so attempting to run ducks settings(but lower clocks obviously lol, and tighter performance level, and different memory divider and 2t due to my boards limitations) will just get me nowhere?
edit2: i figured if my board is going to stop me from reaching 1300 memory speed i might as well see what it can do with 4-4-4 timings, but i guess we will see what it can do when it arrives
Okay, if it's a bench setup increase your VDIMM! :D
Don't try to replicate someone else's settings, use them as guideline but try to find out what your specific kit can or cannot do!
Until next year there's still enough time for some nice Memtesting hours...
it is a bench setup on a budget, but a bench setup nonetheless. i pretty much spent it all between the qx9650 i got cheap and the team xtreem 1300. anyway i will be away for a couple of weeks.
edit: i can see where your coming from, i will have to agree running team xtreem 1300s with a 2 year old cheap gigabyte board is a travesty.
I will contribute it to this forum for the first time
Though there is not even a great record, it is reflected on AMD-40GP of Mr. tanuki
The tolerance of Ballistix of my possession is such a feeling by super pi mod time
Ballistix PC5300 D9GMH 1G*2=570MHz-4-4-4-5-2T 3.15v
http://ameblo.jp/tsubasa1998/image-1...314809234.html
Ballistix PC6400 D9GKX 1G*2=570MHz-4-4-4-8-2T 2.85v
http://ameblo.jp/tsubasa1998/image-1...312663750.html
MB=DFI LP DK 790GX M2RS(1012 β)
CPU=AMD Athlon Ⅱ X2 250@(1.55v)3.99GHz
CPUcooler=Thermaltake MaxOrb
VGA=Force3D RADEON HD3650-512MB
http://ameblo.jp/tsubasa1998/image-1...303019181.html
http://ameblo.jp/tsubasa1998/image-1...303019182.html
http://ameblo.jp/tsubasa1998/image-1...303019183.html
http://ameblo.jp/tsubasa1998/image-1...315248507.html
Oh, I try where I turn around to because I own Ballistix PC6400 1G*2 of the D9HNL deployment in few days
Then please reach the age that people are good for
thats a lot of voltage for 570mhz 4-4-4 :shocked: im hoping to get 600mhz 4-4-4-12 at no more than 2.5v havent tried yet though, 555mhz 4-4-4-12 2.3v and 533mhz 4-4-4-12 2.2v is already a reality though on my last board(ga-p35-ds3p), not stable however. i should be pushing my ram at 5-5-5-15 too but i will probably end up with lower everest bandwidth numbers due to lower fsb and different memory divider and strap.
edit: heres my best everest bandwidth results so far, as you can see im still struggling to get high fsb
vdimm 2.3v vtt 1.3v mch 1.24v pl8 3:4 400 strap
http://g.imagehost.org/0341/1248_5-5...468fsb_pl8.jpg
locked up almost instantly when i tried running memtest so not exactly stable lol. i suspect more mch voltage is required, perhaps a bit more mch reference voltage than the stock 0.76v, maybe 0.8v. and im not sure 3:4 memory divider 400 strap is the best for achieving high everest bandwidth numbers and certainly isnt helping achieve high fsb either.