OMG, Made a break through! (sorta)

[charlie22911]

it does in fact cache to L@, an oldie but goodie, i love this page, lots of good info:

http://www.adriansrojakpot.com/Speed_Demon..._Caching_01.htm

 

while it in theory help your situation, i dont think 64k would help . BTW, i thought the exact same thing about that setting until about a month ago .

 

EDIT:

it may only steal that 64k during 3d accelerated operations... not sure but that should maybe tell us something. i can tell you that you wont notice a real world difference outside of sandra and super pi. know what they say, don't fix what ain't broke .

[NEOAethyr]

Actually I did a bit of extensive testing, excluding sandra's cache bench, does'nt matter to me now though.

 

In everest, video ram caching made no real diffrence.

In superpi, the same, no diff at all(non mod ver).

 

Stability and resposiveness was slightly up though with it disabled, onyl a tiny bit.

 

Bios rom caching.

No diff in everest I think.

But in superpi, a slight increase in performance.

 

Stability improved.

Responsiveness went up again with it disabled.

 

Disabling both yeilded a bigger increase in stability.

I was gaming stable at lower ambients at this setup I have now, but when testing, I was'nt gaming stable, matter of fact, in gtasa my forcing the weather(fog) came undone.

 

After disablign both, the fog no longer messed up and it kept running.

It gamed maybe a bit longer, but that weather thing was a defenite improvement.

 

Defently is faster and more stable, but only by a tiny tiny tiny bit.

I'll disable both by default in the next release .

 

The reason I had bios caching enabled is because it used to be something you did...

It was natural.

 

Vide ram caching, is somehting I used to disable, but after my buddy said it helped his oc on his vga memory, I took a lok at it.

It seemed at the time an increase in performance.

My system runs a bit diffrently now though so I dn if that was valid or not.

 

I mean, the diffrences were barely noticeable.

The only proof I've seen is superpi wit the bios caching.

Same resulting time, but the times mid way were slightly better.

I've ran it enough times at this setup and I doulbe checked with biso cachign disabled and both disabled.

It's defently a tiny bit faster.

 

It's either more stable because:

1: Would use l2 cache, when the cpu is getting stressed for some reason, it could cause stability issues.

2: Whenever it tries to read these roms, it reads them directly, instead of reading ram, the ram may be unstable...

 

Thanks for bringing it up .

And to tell the truth, I got high enough clocks on my 6600gt's memory without the video ram caching being enabled so...

 

 

Edit:

duh...

I think I know why he got the results he got.

 

VIA and ATI AGP 3.0 Chipsets

•Problem

The use of AGP-protocol cycles for coherent access to regular system

memory results in data corruption on systems based on VIA and ATI AGP

3.0-compatible chipsets.

AGP-protocol cycles to the AGP aperture are not affected.

• Workaround

To correct the data corruption problem, the Release 75 driver exclusively

uses PCI-protocol cycles to access regular system memory when it detects a

VIA or ATI AGP 3.0-compatible chipset.

 

Well, it does it on nvidia chipsets now too btw.

Makes it slower.

 

Anyways...

He moved a portion of his used rom into the l2 cach instead of ram, which in turn up'ed his stability with his peticuler driver on his via kt800 board.

The reason why I saw gains before was because I was using a driver that slowed my access down to system ram, l2 cache was all good.

 

After re-enabling system ram access using agp instead of pci, my gains were nil compare to using the l2 cache.

Maybe there's still some gains to be had by using video ram caching, but, even pages are loading up faster and my start menu is even faster with both of these disabled.

I just gonna leave them be as disabled.

 

I dn if I'm right on all of this though, it just struck me as a possiblity.

I remember he had a via board and that via boards have video probs.

So does ati, and I guess intel and sis maybe I dn about those 2 right now.

 

 

I'll have to test his rig out on some 66x drivers the next time I'm over there.

 

It'll give me a chance to check ths out as well:

 

HKR,,UseUncachedPCIMappings, %REG_BINARY%,01,00,00,00 ;Disables Cached I/O Addressing

;HKR,,UseUncachedPCIMappings, %REG_BINARY%,00,00,00,00 ;Enables Cached I/O Addressing

 

That's the generic chipset ver.

I use the oppiset for nv chipsets.

I notcied small gains, but then again I've never really sure about the small thigns wihtout alot of testing.

The best way is just to test on a rig that actually has a prob.

This is for thr nv vga drivers.

I was'nt sure if it was dword or binary, most say dword but that did'nt make sence, seen some result posts when using binary though so...

 

See what I mean though?

This may be why my buddy had better oc's...

[charlie22911]

bios caching is useless as well, there is no need as windows does not use it at all. way back in the day it did speed up DOS apps and other programs that accessed bios, but now the bios is used for booting only, its the main reason hot flashing is possible and im thankful for it . anyways, just something i spotted when testing it. something of interest, is there a hidden option for setting AGP aperture size to 0Mb? i seem to remember talking about it some time... anyways im wondering how it would work on a PC with 256Mb Vram...

[NEOAethyr]

Well video ram caching and rom caching is actually used in windows though...

 

 

Anyways that aperture size, I tried 0 long ago, hardlocks the system.

 

It's used though, the aperture, I forgot exactly what for, but you actually need to setup a certian size.

I allways use the min, but it's wrong.

 

I'll look up my stuff on it in a little while.

That's somehting I gotta do as well, setup my aperutre correctly and help others with that.

I got the src code on that one so...

I can do that, I belive it had somehting to do with some kind of buffer, the aperture size is used in an algo to setup that bufer.

 

I'll have to look it up again though, you can't find info on that on the web.

 

 

As for bios caching.

I am ona 50mhz pci bus and I have a bios chip that runs at 33mhz stock so I dn, maybe there's yields to be had at lower speeds and slower bios chips but whatever I dn.

Then again it may be totally useless in windows.

I don't exactly trust here say on the web if you know what I mean .

 

Hotflashing aught to work without bios caching too, I dn, I don't think I've ever tested it.

You can flash a video card's bios though without any prob at all with rom or ram caching on or off.

 

 

 

Edit:

found the code, but can't find my notes.

I will re-look in a min.

 

Aperture size determains the sizes of push buffer & staging buffer on a nvidia based video card.

I belive those were overridable though, I ill have to look, because smaller aperture sizes are actually better, but a smaller size of one of those buffers may not be good .

You have an ati card thoguh I think, I'm afraid I don't have any code for those.

I can however figuer out this code I got and give ref's for peep's with nv cards, you could probably get an idea of what to use form that.

I dn, if you wanted to anyways.

 

Lol, I gotta look up these "push buffer & staging buffer" things on google to see what the heck they even are lol.

[charlie22911]

okay, the PSU came today and it simply pwns. i have seen a MASSIVE increase in stability, no more background noise in the sound, max speeds have gone way up as i hit 270x9.5 right off the bat, doing 260x9 apparently stable as i did some benching and gaming no problems, will prime later, higher needs too much vcore. this thing is rock solid, but im guessing my cpu or mosfets may have been damaged by the old PSU, 2.5Ghz stable is a no go for now, 2.4Ghz seems to be the limit at the moment. BTW, all of this is in dual channel, no single channel runs here .

[charlie22911]

woa.. i did a 3dmark bench @ 270x9.5 and got through until the last test and it BSOD, arg! i may be getting some dry ice soon so i can go for some insano clock speed . ohh 3Ghz on a $60 board would pwn :drool:

[Clay]

okay, the PSU came today and it simply pwns. i have seen a MASSIVE increase in stability, no more background noise in the sound, max speeds have gone way up as i hit 270x9.5 right off the bat, doing 260x9 apparently stable as i did some benching and gaming no problems, will prime later, higher needs too much vcore. this thing is rock solid, but im guessing my cpu or mosfets may have been damaged by the old PSU, 2.5Ghz stable is a no go for now, 2.4Ghz seems to be the limit at the moment. BTW, all of this is in dual channel, no single channel runs here .

 

Looks like you gotta' pair of good sticks there "Charlie" Which sticks are those exacly. If they'll memtest stable 265 theyll do 262 in windeers.(T1)

Edit: Gettin' windoze stable past that in nf2@T1 is by no means an easy task. (I've tried!!Hehee)

[charlie22911]

these stick i have are G.Skill PC4400LE using Samsung TCCD (440) rated for 277Mhz (ive seen them do 300 easy in nF4). i will do a memtest before i leave tonight and post my results. going to try first for 270Mhz, may even post a pic of it, and my rebuilt rig if i can find that darn camera...

[charlie22911]

okay, i offically hate shin etsu thermal compound

my temps at 2.5Ghz can go as high as 60°C eventhough i have an SI-97 and a 120mm fan, grr. looks like it just dried up... anyways, ive come to the conclusion that any FSB OC above 210 demands single channel and one stick of ram to be 100% stable. go figure, looks like the orginal problem still exists to some extent. if i can get a hold of 1x1Gb then i will shoot with that, cause i need 1gb ram min for the games i play... so here is how its looking, 250x10 100% stable 1 stick of ram, 210x12 100% stable 2x512Mb dual channel. BTW, @ 270FSB, i pass memtest all the way into test5 before i get millions of errors . guess the chipset cant handle it. havnt tried for 260 yet, need some TIM

[Clay]

okay, i offically hate shin etsu thermal compound

my temps at 2.5Ghz can go as high as 60°C eventhough i have an SI-97 and a 120mm fan, grr. looks like it just dried up... anyways, ive come to the conclusion that any FSB OC above 210 demands single channel and one stick of ram to be 100% stable. go figure, looks like the orginal problem still exists to some extent. if i can get a hold of 1x1Gb then i will shoot with that, cause i need 1gb ram min for the games i play... so here is how its looking, 250x10 100% stable 1 stick of ram, 210x12 100% stable 2x512Mb dual channel. BTW, @ 270FSB, i pass memtest all the way into test5 before i get millions of errors . guess the chipset cant handle it. havnt tried for 260 yet, need some TIM

 

"Charlie" I'm not sure if I read that right. But problems with dual-ch like that seem only too familar. Except when I parked "Da Ho" I was having problems @200fsb D-C. with sticks that memtested prefect to 265 each. I wonder if maybe the same thing happened to your board?

To this day I havent a clue as to what happened to it. Except she had been ridden hard before I got er'. And rode er' even harder that the guy I got er' from. It was werkin' ok one day then suddenly it seems no more D-C at nearly any speed..

[Aeka]

@Clay, and others...

 

When you run in dc, can you pass memtest?

If you can pass memtest but nothing else, try letting it loop memtest for a half hour once using test #5.

See if that helps.

As long as you can passs at least 4 loops wihtout any errors, you aught to beable to break it in that way.

 

Recent testing on my own has shown that even if it does error out afterwards, it get's progressivly better over time while looping that.

For example say errors pop up at loop say 18, at 8m.

Then loop umm, 25(who cares), at 6m.

Then loop 47 at 5m, etc.

As long as it does'nt seem to be getting worse.

 

It aught to look something like this really:

 

Just an example

 

Fail, loop 10, 8m - 2 errors.

Fail, loop 25, 6m - 2 errors.

Fail, loop 58, 5m - 1 error.

Fail, loop 70, 4m - 2 errors.

Fail, loop 150, 2m - 2 errors.

 

Etc, something like that sorta.

Just showing ya.

 

It aught beable to pass the point where it failed the 1st time.

Say it fails loop 15, hopefulyl it'll pass up to at least 30 until it errors again, sometimes not though.

As long as it does'nt start erroring out loop after loop.

 

Superpi 32m works better, depends on the cercumstances I guess though.

 

It's worth a shot, got me 275x8 dc stable the other day by letting it run around 30 mins(memtest), no errors though.

I tried at 276 and I was getting something like the above, it was getting better.

 

I dn though, It's probably best to only let it loop if it won't error at all.

To be safe about it.

 

 

(Rant)

I need to get more fans and new memory...

The contacts on my memory apeprently are'nt doing a good job.

 

Also , after I lost that enermax psu due to an accident, I lost lots of stability.

It's very hard now to get s&m stable anything past 250.

I got 275 stable, but not 266 yet.

Though I did run prime for 18hrs a while back.

However, I reset the cmos, and it'll fail at 1hr and a half.

Reset again and it'll pass for over 8hrs...

 

It's ticking me off lol.

I burned an 80m dvd+rw yesterday, with nero.

Failed prime while doing so at around an hour and a half again.

But afterwards it primed for over 5 hrs before I reset my rig.

 

I've tried bios'es that have got me the speed, nothing.

I've made new bios'es, one of which(a test) did get me stable,. but I tossed the change(middle of testing).

So that sorta sucks...

 

Anyways something I do know.

When I got 280 stable, and most of the time when 275+ was stable in dc or whatever, my +3.33v rail did not fluctuate at all.

However, half the time I run on my rig, it fluctuates 0.03v on that rail when starting up.

Other times not at all.

 

The only rail level I'm unsure of is the +5v rail right now.

That's the only one to be adjusted.

Yet this aught to be good enough...

 

 

One thing that does make me happy though, is that 250, is going strong .

Err, except for today.

 

My room fan, I had it on my bed, no casing.

It fell while I was sleeping, burned out.

Anyways when to get online at 250, firefox exited on me instantly :.

 

Said screw it, kinda ticked like alwe man sorta thing.

Went to 266, priming now like it was nothing, no probs like that ^^.

Heh...

 

Anyways I may be moving out of my mom's house right after this next firday, I hope.

Depends, moving to a cousin's house, he said it was cool, I just need to order cable in my name and pay that bill and I can stay.

It's up to his wife though.

He's got dsl allready , I'm so tired of dialup lol .

 

 

When I del a bunch of old posts of mine, I found this, something I was gonna get anyways.

 

 

 

He could be right though about the odd vdimm thingy.

If it's designed anything like the nfii was anyways.

 

 

I did some of my own research this morning.

I figuered out before that my 1st stick of memory (both are 256 megs for ref), could not handle 3.3v.

Yet it can handle 3.2v.

 

In the manual and in the bios(slews and drive settings) the slots are said to be arranged this way from the cpu:

1

2

3

 

I noticed my other stick can handle 3.3v.

I noticed all this when I was testing them one by one, each by them self in single channel in "slot 1".

 

Now, When I have this setup:

 

1: Stick 1 (max 3.2v)

2: Nothing

3: Stick 2 (max 3.3v)

 

I get errors in the last 32meg chunk, 480-512 if I got them at 3.3v, otherwise no errors.

Showing me that the problem lies within the 2nd slot, yet the 1st stick is in slot 1.

 

I swaped the sticks around:

 

1: Stick 2 (max 3.3v)

2: Nothing

3: Stick 1 (max 3.2v)

 

Now I get no errors at all.

 

Also when I get errors from the 1st stick now(Was cas or cpc, man I can't remember now, been up all night), it's the 1st 256megs instead of 256+.

 

Meaning that slot 1 is probably really slot 3.

Then slot 3 would be slot 1.

 

Cpu

3

2

1

 

Instead of:

 

Cpu

1

2

3

 

The voltage however is probably setup like this:

 

cpu (for ref for location of the slots)

1

2

3

 

These are only guesses though...

I do know that my 1st stick when in the 3rd slot does come up as the 1st 256megs though ^^, and it no longer craps out at 3.3v when i the slot for some odd reason.

This is on a lanparty ultra b though, but the same could hold true for the nf4 I suppose.

 

I could be wrong about the slots being the oppiset way around, but the voltage thing is defently there.

3.3v would cause me probs in that 1st stick no matter what the timings were, now they don't...

 

 

 

 

Forget the mumbo jumbo about the slot order .

What it means is that slot closest to the cpu, gives me troubles voltage wise.

 

My adata shows me that.

 

I got a board with one of those smd caps on one of the fets, specifically the fet for that slot closest to the cpu.

 

That slot gives me horrible probs.

Very bad, I can't do really much of anything on it.

 

Some testing a while back showed me that if I slacked timings and setup in a diff way, it was the most stable slot.

But when I tighended it up, it was the worst slot.

I dn lol.

 

 

 

Anyways...

Like I said 250 was doing great.

With my adata dudes, I can do 3-2-2-2, 9-12, 0-0-2-2-0-5-3, cas2.5, cmd2.

I like that, $18 memory.

My samsung can't even get close , old 346 batch.

I want to get new memory so I can do what you guys can do, it would make 275+ alot easier for me as well, probably alot easier.

 

 

One last thing before I end my post.

I did some more timing comparisons recently.

 

266x8 SC mode:

 

10-3-2-3, 9-13,

0-0-2-2-0-2-3

cas3, cmd2

 

3483

3552

3501

74.4

 

 

11-3-2-3, 9-13,

0-0-2-2-0-0-3

cas3, cmd2

 

3461

3551

3527

74.9

 

 

3-3-2-3, 9-13,

0-0-2-2-0-5-3

cas3, cmd2

 

3550

3553

3490

74.4

 

 

The last one seems to be the best overall, but the hardest to do.

Would want more voltage at those higher speeds.

 

 

Err...

You know, I got 278 in dc, prime stable, at those timings, except the alpha's were 0-0-2-2-0-6-3.

Long ago, when I 1st got the panaflo's.

I can't even get close to that now.

I needed 3.1v vmem though, I forgot the vdd.

 

 

Laters anyways.

[Clay]

No it wouldn't pass memtest in D-C at nearly any speed when I parked it.("Infinity") I surmised that the dual channel controller was damaged. My LP-B is good though 252X10 T1 D-C last night went 8hrs 5mins of prime got an error @1.8vcore.(I usually make it do 10hrs.)

At any rate that's an improvement I didn't expect. A while ago it wanted 1.9 to do 210X12 Wonder what happened board burn in? Blaugghhhh!!! Hehee...????? Oh-well