My new system

[TheGeekMistress]

267X12 but I get massive video stutter playing bf2. Probably a setting thats off but I'm not gonna try to figure it out right now since it's working really well where it's at.

it must be your ram timings then. are you running 2T? at 266mhz i have to run 2.0-4-4-8-2T ...for some reason my ram refuses to run CL 2.5 ..but 10mhz higher on the ram won't make up for the drop in timings. i have to go from 235 2-2-2-5-1T to 255 2-3-3-5-1T to remain with the same performance, but the bandwidth increase makes my system feel faster. i've tried high mem busses with one stick but the spread from 255 using loose timings can't make up the performance. so for me 250~260 2-3-3-5-1T is my sweet spot. i'll just adjust the ht/mem bus around in there and use the multi to raise the cpu frequency. i'm going to try 254x13 to get 3300mhz cpu. that should get me a nice boost. from my current p-rating of 4900 to 5300 or more.

 

TGM

[NEOAethyr]

Nah I was talking about superpi and that was it.

Evrest of course is still effected by latency with higher mem speeds and lower timings.

However once you get to a certain point, superpi is'nt affected at all, not in the lower ms even.

However, if I change to single chan mode, my times go up, because I'm not pushing 4k in memory bandwith or more.

I'm around like 35 or something, I forgot(Mostly depends on the stick used in sc mode, if it's the adata I can go lower on most timings and especially tREXT, which in it's self has got me over 800megs a sec diff before, it gets me over 4k in single chan).

 

Wierd part is, in everest wise...

If I get to a certain speed, lowering tCL slows me down.

tCL lowering is'nt stable though.

Higher reads by 2megs a sec at 275 in single chan.

Higher latency though, by 0.2ns or something I forgot.

Tested multible times.

I would guess it's because I'm not stable though, so many people put there effects into cas...

 

 

I'm not saying lower the mem speed...

By all means keep going up.

But I am sayign that once you hit par, the times in superpi will not change not even by a ms.

From there it's all cpu mhz when it comes to that one program.

Meaning you are on par or better then any computer out there in the world, super computer bench wise.

 

As is btw,m I get over 4200megs a sec in reads in everest now.

4k in writes, or maybe it was 4.1.

Latency however is around 64ns : .

2.2ghz btw.

Gets worse as I up the cpu, because of the SIP ROM, your SIP ROM is identical on the lower half, or at least it was supposed to be, it's close anyways.

It gets worse as you go up in multiplier.

The lower the multiplier the lower the bandwith, as you shoudl know that higher cpu speed give more bandwith...

That is true without a doubt, but a higher multiplier decreases speed.

 

 

Small ?, do you guys have a timing called tWE?

I have 0 for both of my sticks, but there is no diff at all in evrest or sandra.

Windows and gaming wise there is, superpi in single chan there is too.

[Thraxz]

Weird, at 280x10 9:10 2.5-3-3-5-1T (sisandra- ~7100mb/s) my 170 will run super pi to constistent mid 30's (30.3xx-30.6xx). If I decide to try and push my cas to 2-3-3-5-1T I have to run 280x10 5/6 2-3-3-5-1T (sisandra - ~5200mb/s)my times increase into the mid to upper 31s range (31.6xx-31.8xx). I wonder what makes such a difference.

[TheGeekMistress]

Nah I was talking about superpi and that was it.

Evrest of course is still effected by latency with higher mem speeds and lower timings.

However once you get to a certain point, superpi is'nt affected at all, not in the lower ms even.

However, if I change to single chan mode, my times go up, because I'm not pushing 4k in memory bandwith or more.

I'm around like 35 or something, I forgot(Mostly depends on the stick used in sc mode, if it's the adata I can go lower on most timings and especially tREXT, which in it's self has got me over 800megs a sec diff before, it gets me over 4k in single chan).

 

Wierd part is, in everest wise...

If I get to a certain speed, lowering tCL slows me down.

tCL lowering is'nt stable though.

Higher reads by 2megs a sec at 275 in single chan.

Higher latency though, by 0.2ns or something I forgot.

Tested multible times.

I would guess it's because I'm not stable though, so many people put there effects into cas...

 

 

I'm not saying lower the mem speed...

By all means keep going up.

But I am sayign that once you hit par, the times in superpi will not change not even by a ms.

From there it's all cpu mhz when it comes to that one program.

Meaning you are on par or better then any computer out there in the world, super computer bench wise.

 

As is btw,m I get over 4200megs a sec in reads in everest now.

4k in writes, or maybe it was 4.1.

Latency however is around 64ns : .

2.2ghz btw.

Gets worse as I up the cpu, because of the SIP ROM, your SIP ROM is identical on the lower half, or at least it was supposed to be, it's close anyways.

It gets worse as you go up in multiplier.

The lower the multiplier the lower the bandwith, as you shoudl know that higher cpu speed give more bandwith...

That is true without a doubt, but a higher multiplier decreases speed.

 

 

Small ?, do you guys have a timing called tWE?

I have 0 for both of my sticks, but there is no diff at all in evrest or sandra.

Windows and gaming wise there is, superpi in single chan there is too.

well, with my old NF2/Barton one stick or two made no difference in SuperPI, it must be your modbin tweaking that has made something different.

 

best i got with the nf2/barton and my BH5 was 255x10 2-2-2-2-10-1T and i got 62ns and 4100mb/s ...altho with my TCCD i got 260x10 2-2-2-2-8-2T.

 

but thats the old system and you can't compare it to the A64 with it's internal memory controller. in fact, with this system, if i take out one stick i get higher superPI times. and i'm running the single channel NF3 mobo.

 

TGM

[TheGeekMistress]

Weird, at 280x10 9:10 2.5-3-3-5-1T (sisandra- ~7100mb/s) my 170 will run super pi to constistent mid 30's (30.3xx-30.6xx). If I decide to try and push my cas to 2-3-3-5-1T I have to run 280x10 5/6 2-3-3-5-1T (sisandra - ~5200mb/s)my times increase into the mid to upper 31s range (31.6xx-31.8xx). I wonder what makes such a difference.

you've got a different platform than i do. you're running the 939 cpu with dual channel memory. NEO is running the older NF2/Barton system. i'm running the DFI NF3-250 skt754 at 3100mhz, 259x12 2-3-3-5-1T single channel and i get 37ns and 4000mb/s in Everest. the tighter the mem timings and higher the internal memory controller frequency and core frequency, the lower the superPI scores, but that only takes you so far, the rest is up to the frequency of the memory controller. since it moved from the North Bridge in the NF2 platform to the "on die" design of the A64 cpu, the internal memory controller's latency perfromance and frequency is controlled by the frequency of the cpu itself.

 

i get 29s [1M] and 1m 7s [2M] in superPI ...i bet i could do better but i haven't much messed with the vastness of the settings in the mem page in the bios. i leave everything on AUTO and just change the basic timings. but i'm working on getting 3.2ghz outta the chip right now.

 

TGM

[Thraxz]

I am really starting to think my RAM choice sucked nads and that my poor 170 is totally starved for memory bandwidth. Everyone else seems to give mild .1-.2s decreases or sometimes less running a divider up or down and mine is responding in the .5s range to a divider change down and tighter timings.

 

Anyone wanna give away some BH-5? Anyone?

[TheGeekMistress]

I am really starting to think my RAM choice sucked nads and that my poor 170 is totally starved for memory bandwidth. Everyone else seems to give mild .1-.2s decreases or sometimes less running a divider up or down and mine is responding in the .5s range to a divider change down and tighter timings.

 

Anyone wanna give away some BH-5? Anyone?

mostly you're core only running at 2.8ghz makes a difference b/c your internal memory controller is only running at 2.8ghz and then your memory timings. if you ran the bus down lower so you could run 2-2-2-5-1T then your superPI time's will drop some... but the cpu/memory controller frequency of 2.8ghz slows you down too.

 

just try dropping your mem bus so you can run 2-2-2-5-1T at the fastest config you can. then see what your times are. you really don't need all that memory bandwidth for superPI ...it's pretty much useless for low superPI scores.

 

it's all about the cpu/memory controller frequency and the tight timings of your ram.

 

TGM

[NEOAethyr]

Nm TGM, you were'nt listening.

 

And no my modding is'nt what is doing it.

The diff is because I'm around 800megs a sec slower in single chan, because of timing diffs.

If I go dual chan it makes up for that.

 

If I chnage my timings to something tighter, the diff is min.

I cannot make up the diff completely though.,

I'm talking about 280 (560) mhz bus here...

 

 

 

@Thraxz

Those are running diff dividor's are they not?

Kinda hard for people to get a grasp of what is what there...

If superpi has'nt reached it's peak for mem bandwith needed, then lowering say tCL(cas) would help for sure.

 

 

@TGM

I don't care what platform it is.

I'm not comparing the 2 here.

I was talking genericly.

Get it through your head that I don't give a hoot about your platform, and I'm not trying to compare the diffs here.

I'm only mentioning what I know, since you asked some dude his superpi scors, I told you that once superpi reachs a certain amoutn of mem bandwith or latency, it does not get better, once you hit that mark, it's all cpu.

It doe'snt matter what platform you have, once you hit the mark, it's time to up the cpu to get better when it comes to superpi'ing.

That and maybe the small things, like bus latency's, such as you addon cards and hardrives, more memory may help make it more consistent through the loops, etc.

 

Also note that not all memory is created equal, even if it's the same type and batch.

The diffs, though related directly to memory bandwith and latency, can be controlled on the mobo.

If such has'nt been inplemented, it seems the de's at dfi don't write there own code, instead they tak the value off the SPD's and diretly use those, and it seems to be on a per stick basis (ie you can mix sticks and have diff timings for both at the same time).

So if you have'nt reached the peak in superpi, the diff in memory brand's and rev's of there spd may make up diff's, but so do the viewable timings on the board.

 

Lable me as a modder wannabe, I don't care.

I may use some tools that have'nt been written by me yeah, like taiphoon, winhex, modbin and cbrom.

But it does'nt mean I don't do alot of work, some of the work is done by hand wrtten batch scripts and asm code.

So wtf ever ^^.

Thanks, I'm wasting my time posting in the hardware comunity as a whole, I know this allready, you're just making it obvious yo ^^.

 

 

Tell the mod's here you you got a useless post in this thread and you want it removed lol.

Again I think I've mentioned this a few times, I dont know what I'm talking about right?

I won't be mad if you do, it's to the point.

The mods may think I'm getting to aggressive, either way, whatever they choose is all ok with me.

 

Keh, jeez...

[TheGeekMistress]

geez dude chillout

 

all i'm saying is that we all have different platforms and they all have their different tweaks. and i did mention that once you get the mem timings tight and and high mem bus that the rest is the frequency of the internal memory controller and cpu frequency. maybe i didn't state it clear enough for you.

 

i hope Thraxz got what i was trying to say.

 

TGM

[Thraxz]

I did, I put the memory down to 2-2-2-5 1T but I had to put it on a insanely low divider that put the memory down at 180mhz. My efficiency was 94% but I could only pull ~5400mb/s and my superpi went up to 33.0xxs.

[Sorrento]

Correct me if I am wrong GM, with the socket A platform as well as others... do you get higher overclocks with only 1 RAM module or not?

 

I remember something like this, never tried but knew someone... oh, I remember now, he was using two sticks of 256mb each, and in theory you can overclock more than with two sticks of 512mb. Is this true as well???

 

I wonder what the socket 939 has to say about this, since maybe loosing real dual channel will hurt a lot... but the memory controller should work well with just one stick.

 

I feel I should have read all the other messages... lol, well sorry... me no english

[Thraxz]

I shifted back to the 2.5-3-3-5 1T and came up with ~6800mb/s and my superpi dropped back to 31.5xx from 33.0xx. Doesn't that in itself speak volumes for bandwidth over latency? Of course my RAM is crappy and I really wish I had some BH-5 since I can pump in so many volts on this lovely DFI board so can more effective take the "crappy-at-low-frequency-tight-timings-RAM" factor out of my equation.