8RDA6+ Pro Romsip Mod

[Wolf2000me]

Originally Posted by Runnel

Thanks for your compendium. Indeed, I'm very impressed! This is really scientific research what you guys in Poland are doing. Respect! Your explanation of the bios being responsible for differences in memory-performance is quite convincing to me.

Yet one basic thing maybe I didn't quite understand: You are all testing with fix fsb and variable multi. If we would talk about Via-chipset based motherboard I could understand. Since on Via-board you can easily change the multi in Windows (with cpu-msr from your southern neighbor) but not the fsb.

On nForce-boards - the 8RDA6+ Pro is my first one - I learned that it's not possible to adjust the multi in windows. So what is that testing with different multis good for? Tests for the bios-performance or to get informations for new bios-programming?

At least in term of optimizing one's system to me it still makes sense, to find the maximum for instance of memory-performance with a fixed multi and variable fsb - since it's the fsb we are able to change and not the multi???

The reason i'm focussing on multipliers is for several reasons of which some of them probably correspond with the other dudes like MietmaN, Naibot etc.

The fact that the memory performance increases or decreases while you only change your multiplier is very much stranger then the perfomance increase you get in upping your FSB. You could say it is because of the speed gain of the CPU but after the latest testing it seems that is not true, which could thus only be related to other variables, specified in the BIOS for example. I'm sure if you ask Merlin he could produce some details. Also the speedgain by the cpu is normally no real factor for your memory. That's the whole thing about the FSB being at a, or several dividers. But anyhow... By testing new alpha or beta bios binaries made by ex. Merlin may get in a longer term the Bios with the fastest stable and perfomant multiplier you want when you sort of found your FSB and memory setting. Or perhaps even overall better memory results because of a new tweak he discovered workes best for nforce2 and/or the 8RDA3/6 Pro. That's why i'm interested in it. I'm looking at it from a performance way of viewing. You are more viewing in a practical way of viewing, i think as i can tell from your intrest in a speedfan 12v mod.

I hope that makes sense, and of course all of the above may or may not stroke with your reality

edit: I also noticed i can't change my multiplier on the fly. I used to be able to do that with my VIA based board with cpumsr or central brain identifier. If anyone knows something about that it could be handy in a practical way of viewing

[Gizmo]

Originally Posted by Runnel

You said: "so there's not much chance of speedfan being able to control them, unless you make some modifications to the board!"
You mean it's possible to connect the fan pin to some other place of the board instead of the plain 12 Volt line? This would be a nice project to do I've got again some more time.

To do this, you'd have to have a FET that you could hook up to one of the PWM outputs from the WinBond SuperIO chip. This of course assumes that your board has a WinBond SuperIO chip or something similar (highly likely) which has a PWM output (probably) that isn't used (maybe). And then you have to get a FET and bias it properly to do what you want. Not really a trivial mod.

[MietmaN]

Originally Posted by Runnel

@ MietmaN

Thanks for your compendium. Indeed, I'm very impressed! This is really scientific research what you guys in Poland are doing. Respect! Your explanation of the bios being responsible for differences in memory-performance is quite convincing to me.

Yet one basic thing maybe I didn't quite understand: You are all testing with fix fsb and variable multi. If we would talk about Via-chipset based motherboard I could understand. Since on Via-board you can easily change the multi in Windows (with cpu-msr from your southern neighbor) but not the fsb.

On nForce-boards - the 8RDA6+ Pro is my first one - I learned that it's not possible to adjust the multi in windows. So what is that testing with different multis good for? Tests for the bios-performance or to get informations for new bios-programming?

At least in term of optimizing one's system to me it still makes sense, to find the maximum for instance of memory-performance with a fixed multi and variable fsb - since it's the fsb we are able to change and not the multi???

I'm already getting lost in all those details a little bit , so I'll try to describe the aim of such tests step by step :

1) We are all computers' freaks and geeks - a normal user wouldn't pay attention to such things nor he would notice any significant difference in memory performance (or performance of the whole system).

2) We're talking about overclocking all the time and getting the absolute maximum performance out of our PCs. It concerns both daily use and maximum OC (for benchmarks or just to know, what the max is ). In Poland we say : "to get the 7th sweat out of something" - in this case out of our computers.

3) Changing multipliers does afflict our memory performace - this is a stated fact, based on all the tests that have been made so far (on different motherboards) - it's true at least for our nForce2 chipset.

a) As a rule, lower multipliers seem to have some negative influence on memory bandwidth - many Abit NF7 motherboards just won't boot with a multi set to 9x or 9.5x while using stock Abit bios (I've experienced this myslef). This issue was the cause for which TicTac started modyfing bioses some time ago (he makes them for Abit NF7, DFI Lan Party and DFI Infinity). Even if a mobo can boot with such a multi, its memory performance is worse than expected.

b) Bad memory efficiency with low multipliers concerns only those motherboards, that can reach high FSB speeds of 250MHz and above. Imagine a geek, who wants to make some highscore and can set the FSB of his mobo up to 270Mhz (256MB of RAM, single channel) - providing the maximum CPU overclock is about 2.6GHz, he sets the 9.5x multi and suddenly his mouth drops open! He gets lower score than at 10x260 - what the hell is going on?!

c) There are some exceptions though - like our motherboards' stock bios, what you've found out Mutli 9.5x gives better memory efficiency than 11x - it's good for you while using high FSB in daily use and decent but not maximum CPU OC, but mark most of us don't usually run at high FSB speeds : typical 11x200 setting should be very efficient in terms of memory performance (multi 11 is well optimized in most of the stock bioses), but it isn't in this particular case. Also, high FSB isn't something common for Epox nFroce2 motherboards.

4) Gimme a break - I'll finish later!

[MietmaN]

4) /final version ;/ As a rule, higher FSB means better performance while overclocking 1:1 (CPU : Memory), but then the multiplier comes into the game and messes things up. When every single MHz of our FSB is important we can suddenly discover, that higher FSB with lower multi isn't as efficient as lower FSB with higher multi, although the differences are rather small (providing the same CPU overclock) and this all due to memory bandwidth. It's especially drastic, when the multi used for testing is some sort of a breaking point - e.g. decent bandwidth with 10x, but a sudden drop with 9.5x - such things happen.

5) Higher FSB = better performance? Not in all cases, because apart from the multiplier the memory timings come into game as well. In some cases, when "a bad multi" pairs up with loosened timings, it turns out that higher FSB isn't worth trying at all.
Examples : 11x - best multi, 10x - the bad multi.
11x230, 2.5-3-3-11 vs. 10x250, 3-4-4-11
It can happen, that the latter is worse in terms of performance due to sudden drop of bandwidth with the 10x multiplier and loosened memory timings, although an increase of 20MHz on the FSB is rather large.

5) The aim of all this testing is to find out which multi pairs with the best memory efficieny. Knowing this, we can then configure our system in the most efficient way, both for daily use and some overclocking experiments. It ain't easy, I must admit, because we have to think about everything at the same time : CPU OC in MHz, FSB speed that our memory and chipset can handle and all the other things like memory timings and all voltages.

6) The aim of modyfing bioses, apart from other reasons like e.g. alpha timings or L12 soft-mod, is to optimize memory bandwidth for the widest possible set of multipliers. When good optimized, we don't have to think about choosing the proper multiplier - knowing the max FSB and the max CPU OC we can set it in bios without any worries and achieve the max stable FSB with good memory bandwidth.

Huh, that will be it!

PS. I never overclock in Windows - used 8rdavcore a month ago, but I got some strange system instabilities.

[ConiKost]

Anybody know how can I mod a Bios myself ?

[Wolf2000me]

Hey

here you can find Tictacs homepage. He, together with Merlin and a few others are the most known Bios modders. Ask them and start your search there. Perhaps in the forums there they may be able to help you. There is already some software for modding bios binaries there however the page is still under construction.

[ConiKost]

Wolf2000me,
are you using Merlins Bios ?

I got the Problem, that when I use this Bios, my System ist not stable and MEMTEST86 freezes in Test 5 ... (FSB200)

In FSB166 and lower aren't any Problems!

But when I am using the original Bios, everything is fine, MEMTEST86 too!
On FSB200 too!


If you are using Merlins Bios, could you test your RAM with MEMTEST and tell me if it freezes?

[labtech1]

I get the same result with Merlin's modded bios. Tictac posted this ROMSIP Table Mod Guide in the PC Perspective forums.

[Wolf2000me]

Originally Posted by ConiKost

Wolf2000me,
are you using Merlins Bios ?

I got the Problem, that when I use this Bios, my System ist not stable and MEMTEST86 freezes in Test 5 ... (FSB200)

In FSB166 and lower aren't any Problems!

But when I am using the original Bios, everything is fine, MEMTEST86 too!
On FSB200 too!


If you are using Merlins Bios, could you test your RAM with MEMTEST and tell me if it freezes?

Your memory probably has a problem with the CPC1 setting. Try to flash to the CPC2 bios by Merin. I don't have any trouble with memtest though I use a different program which i uploaded.

[MietmaN]

@ ConiKost :
What about that SATA HDD of yours? Have you resolved the problem you had or should I report an issue to Merlin?

As to MemTest86 errors : it's a pity I'm not home now, since I would like to check my memory. The complicated thing is : sometimes one mobo works well with a specfic bios and the other (the same revision and everything) does not - I'm not kidding guys; I've experienced that myslef during my adventures with Abit NF7.

Other possibilities :
- the way we flash,
- some mem sticks need 2T, especially while in dual channel mode,
- some mem sticks can't handle even FSB200, while using a custom, aggressive and fast bios => hence the errors with Merlin's one and no errors with the stock one, I assume.

Every system can behave different - it's a pity only Merlin has done something for Epox users; the wider choice of modified binaries the better chance for us to pick the suitable one.

cheers

[Naibot]

Hi there,

Yesterday i opened up my pc's case, and took one memory module out.
So no dual channel and memory adressing problems anymore
Look at the results

http://www.xs4all.nl/~druite23/OC/250.png

And it's pretty stable too, getting 255 stable is harder.
It seems everything is shifted about 10 MHz. Not only i'm stable at a 10 Mhz higher FSB. I can set my timings a lot tighter at all FSB's.

Man i love this board

ps. i noticed that the cold boot issue is related to the identification of the memory. The onboard LED displays C1, when you try a warm boot.

[MietmaN]

Originally Posted by Naibot

And it's pretty stable too, getting 255 stable is harder.
It seems everything is shifted about 10 MHz. Not only i'm stable at a 10 Mhz higher FSB. I can set my timings a lot tighter at all FSB's.

Man i love this board

Makes me mad I'm not home now (I mean Poland) - just one more week and I'll torture my poor system without mercy.

ps. i noticed that the cold boot issue is related to the identification of the memory. The onboard LED displays C1, when you try a warm boot.

Let's hope Merlin will fix everything in the next release. Have you run memtest86? There're some people here with stability issues and memory errors, but I think their mem is to blame - works well with stock bios, but can't handle some tweaks from the Merlin's one.

[Naibot]

Originally Posted by MietmaN

Let's hope Merlin will fix everything in the next release. Have you run memtest86? There're some people here with stability issues and memory errors, but I think their mem is to blame - works well with stock bios, but can't handle some tweaks from the Merlin's one.

I had no problems running memtest. Maybe they're using an older version of the program? I'm using memtest86+, not sure what's the difference between the other memtest86 Anyone?
Get it here http://www.memtest.org/

[Barix]

Some time ago I tried the old memtest and it was giving some false positives (errors on pretty stable system, where prime95 and memtest+ can run for hours without any error)

So use memtest+ .. the other is buggy IMHO

[Naibot]

[offtopic]

Anyone still considering wich mem to buy? BH-5 or TCCD?
Look at this review
I'm sticking to my G-skill babies, that's for sure

[/offtopic]

[MietmaN]

Originally Posted by Naibot

[offtopic]

Anyone still considering wich mem to buy? BH-5 or TCCD?
Look at this review
I'm sticking to my G-skill babies, that's for sure

[/offtopic]

[offtopic]

I've already bought exactly the same sticks you have (2x512MB dual kit of G.Skill PC4400 sht...E 2.5-3-3-7) - they're waiting for me in my home city (in 6 days I'll grasp them and feed my computer ) - good review, anyway.

<G.Skill rulez>

The second best result in our little Super PI competition was done with G.Skill onboard

http://www.pcprojekt.pl/forum/viewtopic.php?t=7029

another G.Skill's score :

http://www.pcprojekt.pl/forum/viewtopic.php?t=10046

(FSB/HTT 300 with 2.5-3-3-7 1T timings ) - no software is capable to read 200:180 memory divider on A64 platform, but I trust this guy, I know him

[/offtopic]

[Wolf2000me]

Hey,

I bought some OCZ VX 3200 2x512mb and the booster. It was priced in the range of some G.skill LC memory and i bought the booster. When i get it i'll post some results as well