Alpha Timings and the NF7-S

[felinusz]

Introduction to the Alpha Timings

The best way to start this thread, is by explaining what we understand the Alpha Timings to be. Another's words are better suited for this than my own.

Samurai Jack, DFI forums

Alfa timings are like internal timings in the mobo that are ussaully dead set in a bios. These timings have a lot to do with why X board will do this speed & how Y board will get this score at that clock & how Z board will clock to this level with that ram.

The Alpha Timings

T(DOE)
T(RRD)
T(W2P)
T(W2R)
T(REXT)
T(R2P)
T(R2W)


That's about the extent of what I've been able to dig up on Alpha Timings. I still have no idea what they each specifically do, even after several (numbered in the tens) hours of research on the subject. If anyone has a better, in-depth and specific explanation, please do share. An explanation of each individual timing would really make my day.

Simply put, the Alpha Timings effect your Memory, Memory controller, and (to a much lesser extent) your AGP controller, and the overclockability/performance of all of the above.

After a glance at the bandwidth changes one will likely see with different sets of Alpha Timings, it is easy to quickly dismiss them as game for only the finicky super-hardcore tweaker. This is not the case.

I cannot stress enough that the Alpha Timings are the key to getting the most out of your memory overclock, and your memory bandwidth. Alpha Timings are one of the main reasons why the DFI Infinity/LanParty B motherboards are widely regarded as the best Socket A FSB overclocking motherboards on the market. These boards have the Alpha Timings "unlocked" in newer BIOS revisions, allowing manipulation of them from the motherboard's BIOS, an extremely useful feature.

I was overwhelmed and confused when deception`` told me how useful and influential manipulation of the Alpha Timings can be to your memory overclock. A week and a half of messing around with Alpha Timings on a DFI Infinity later, I realized how important they are, and wondered why on earth they weren't in the NF7-S BIOS.

Our motherboard, the NF7-S, does not allow Alpha Timing manipulations from within the BIOS. Instead, we need to use a program from within windows to maniplate the Alpha Timings.

The Incredible Timing Tweak Program

The program that we can use to manipulate the Alpha Timings, is called nForce2 Tweaker, made by Tictac the mighty BIOS Modder and CodeRed, the author of the popular A64 tweaker.

nForce2 Tweaker link: http://users.tpg.com.au/spark69/NF2Tweaker/nf2tweaker025b.zip

This program is simply incredible, and allows for modification of most of the "main" memory timings from within Windows. The only downside is that it doesn't seem to play well, when any hardware monitering software like MBM5 is running at the same time.

What Is In This Thread

All this aside, let's get into why exactly I'm making this thread so "late in the game", with everyone switching to A64, and Socket A quickly becoming obsolete.

Very few Abit NF7-S users know about the Alpha Timings, and fewer have taken the time to play around with them. I didn't know about them, and likely still wouldn't, if I hadn't taken up a DFI Infinity for two weeks.

What I've started to do on a simple level, is compare several (9) sets of Alpha Timings that I've either come across, or found on my own, with the NF7-S motherboard. These are all sets that I found gave me interesting results with both my DFI Infinity, and my NF7-S. The idea is to use these 9 sets as a good starting point, to make tweaking out your own system's Alpha Timings that much easier to start in on.

To start with, I've done a basic Memory performance test with my 9 sets of Alpha Timings, simply using the SiSoft Sandra Memory Bandwidth benchmark across the board.

To eliminate the margin of error, I've run each test 10 times, and just kept the highest results. I'm fairly confident that the margin of error with these results is around or less than ~1 MB/s.


I also wanted to find a way to compare how each set of timings affects my Memory system's overclockability if possible, but haven't figured out how best to do this, and also can't really justify doing it.

The problem with comparing Memory system overclockability as it is related to each Alpha Timings set, is that since we can only manipulate the timings from within Windows, memtest86 can't be used to quickly and thoroughly test RAM stability at different voltages. Any memory testing programs run from within Windows are pretty much useless from the get-go because of the limited amount of memory they can test. Running 3DMark to test 3D stability isn't a good comparison for everyone, as not all of us have AGP controller stability issues tied to high FSB speeds. Prime95 doesn't use all of the memory, and isn't stressful enough. I don't know of any other good memory stability testors, that run fro within windows.

The issue with justifying the purpose of such a test is that while Bandwidth changes as related to Alpha Timings sets should be consistant between all NF7-S motherboards, the effect of Alpha timings on memory system overclockability will certainly not be consistant, especially given that all RAM chips are different, and will react differently to different Alpha Timing sets. Comparing the overclockability of my BH-5 in relation to each Alpha timing set, is completely useless to someone using Hynix BTD43 Memory.

That said, some sets have noticeably improved stable overclockability with my NF7-S, and my BH-5 memory, and some have decreased it.

In the end, I have decided that as far as number-specific tests go, I should only stick to Alpha Timings as they relate to memory system performance (for now), and leave out Alpha Timings as they relate to memory system overclockability. We will leave this testing up to everyone who reads this, and decides to give the Alpha Timings a try.

You, the end user, are responsible for finding the best combination for both performance and overclockability, for your specific system, if you decide to spend the time at it.

When it comes down to it, that's what tweaking is all about anyways

Time to end all the explanations, and get on with the numbers

 

[felinusz]

The Alpha Timing Sets

These do not by any means represent all of the Alpha Timing Sets that one could conceivably use. There are thousands upon thousands of possible combinations with this many timings, and, as you're about to see, the individual timings make less of a difference than the entire set as a whole. The Alpha Timings seem to be largely dependant on each other, performance-wise and behaviour-wise - this may be incorrect, but certainly reflects my personal observations of the behaviour with different sets. These are the main 9 sets that I tried out with my DFI Infinity and NF7-S, that I got interesting and signifigant bandwidth/overclockability results with. Some of these are timings sets that I stumbled across other people using, some are sets that I found myself.

As I said in the above lengthy post, the idea is to use these 9 sets as a start point, to make tweaking your own machine's Alpha Timings less painful and time consuming.

Now is a good time to say, 0-0-3-3-0-0-3 is the absolute lowest set of Alpha Timings that my BH-5 and NF7-S @ 200 MHz CAS 2-2-2-11 is able to run without system crashes, lockups, random high-pitched sound distortion followed by a crash, and general bad things happening (regardless of voltage ). Unless you are feeling adventurous, don't bother trying timings lower than these.


All Alpha Timing Sets are in the order:

T(DOE) - T(RRD) - T(W2P) - T(W2R) - T(REXT) - T(R2P) - T(R2W)

2-2-5-3-2-2-3 (Stock Abit Timings)
0-0-3-3-1-0-3 (Tightest Timings Possible, except that T(REXT) can go to 0)
1-3-3-3-2-1-3
1-3-4-3-3-4-4
2-3-5-3-2-2-4
3-2-5-3-2-2-3
5-4-6-3-3-5-6
5-6-4-3-3-5-6
6-6-6-6-3-6-6 (Loosest Timings Possible)

The Results

I should say again, that these gains/losses look totally insignifigant and insubstantial on paper - you really do need to try the Alpha Timings out for yourself before you can really believe in them. And remember, any gains from Alpha TIming tweaks are totally free .

All benchmarks were done with 2 X 256 Megabytes of BH-5 memory, at 200 MHz, 2-2-2-11, with a 10X multiplier, RAM in Dual Channel, CPC ON, CPU Interface Enabled, with Tictac's D26 Mantarays XT modified NF7-S BIOS. The RAM sticks were installed in slots #1, and #3.

All Benchmarks were repeated 10 times, with the highest results from either INT or FLOAT shown, to help eliminate any margin of error. I'll say it again, I am confident that the margin of error with these numbers is less then, or near, ~1 MB/s. These numbers represent substantial and 100% consistant gains that I see with my RAM and motherboard.

Results are written: RAM BANDWIDTH INT (MB/s) / RAM BANDWIDTH FLOAT (MB/s)

2-2-5-3-2-2-3 - 3068/2879 A
0-0-3-3-1-0-3 - 3076/2888 T
1-3-3-3-2-1-3 - 3076/2887 *
1-3-4-3-3-4-4 - 3072/2883
2-3-5-3-2-2-4 - 3068/2879
3-2-5-3-2-2-3 - 3068/2879
5-4-6-3-3-5-6 - 3064/2875
5-6-4-3-3-5-6 - 3070/2881
6-6-6-6-3-6-6 - 3059/2867 L

* = the set of Alpha timings that I find gives me both the best memory bandwidth, and the highest stable memory system overclock with my BH-5 RAM, and my motherboard.

T = the tightest set of timings I tested

L = The loosest set of timings I tested

A = The stock Abit timings, that (most) NF7-S BIOSes have pre-set

Here's a pleasant enough looking MS Paint/Excell graph, to give an idea of how signifigant the differences are (the graph begins at 2750; the performance differences are relatively insignifigant, less than 1%):

 

[felinusz]

Reserved Space

 

[disk11]

Nicely done.

What were your max FSB overclocks with the timings you specified?

 

[felinusz]

Here's where things get complicated - which is why I left the third post blank for more updates as I learn/test more on my system, and push my 3D stable FSB wall higher .

Right now, my system can run pretty much any FSB speed my memory can run, with Prime95, and memtest86 stability. It's with 3D stability that I am having issues with my personal motherboard. This makes it both easy to test, and hard to make relevant to others, any fully stable FSB gains that I personally see.

With 1-3-3-3-2-1-3 timings, my motherboard seems to be maxing out with full stability at a 235 MHz FSB speed 1:1, at least as far as 3D stability, and Prime95 stability. As I explained above, it is currently impossible to run memtest86 with anything but the stock Abit Alpha Timings, due to our need of a program to alter them from inside Windows.

A hacked/altered version of memtest86 with manipulatable Alpha timings outside of Windows would be a godsend, even this "late in the game" as far as Socket A is concerned. If we NF7-S users want to hold out a little longer on an A64 upgrade, more overclocking options like these will make it a lot more fun

My BH-5 is being run at 3.3V 24/7 right now, but will do a 235 MHz FSB memtest86 stable at CAS 2-2-2-11, Abit Alpha 2-2-5-3-2-2-3 with 3V.

Through some more overvolts, and use of the NF2 Tweaker, I've conquered the previous wall my system saw at 225 MHz FSB, for 3D stability. Altering AGP controller latencies, and Alpha Timings, have both done wonders for my board.

In contrast, the stock Abit Alpha timings still leave me stuck at a 228 MHz FSB 3D stable (I've started overvolting my board's chipset and VAGP more aggresively, this has also aided my beating the old FSB wall of 225 MHz). This is surprising, because the 1-3-3-3-2-1-3 timings I'm using now are tighter than the stock Abit ones.

I still haven't spent much time thoroughly testing the following timing sets, which garnished large FSB maximums for me with my old DFI board:

5-4-6-3-3-5-6
5-6-4-3-3-5-6
3-2-5-3-2-2-3

I'm currently running 3DMark looped with 5-6-4-3-3-5-6, at 240 MHz FSB. So far so good, been going for half an hour so far.


I'll have more results with my hardware soon, this stuff is very time consuming


I also wanted to ad, check out how some timing sets alter INT more, and others alter FLOAT more.

 

[pershoot]

good job man some points to add:

a value of 4 for tW2P and tR2P seem to work best for agp and fsb stability by the same token. a matched value works best.

tREXT is most important in boosting throughput, but can also make or break your stability (depending on your fsb mark).

high agp controller latency (i.e. 255) helps in combating the agp 'stuffed up ness' that happens at higher fsb.

tRFC complements your throughput the most at any given fsb mark. this can also break your overall stability (may be stable in some areas but unstable in others, while going to complete instability across the board).

tRC helps balance out tRFC.

Auto Precharge of Disable adds to FSB stability with trivial loss in throughput.

Drivestrength of 6 and Slewrate of 7 work well for BH5.

 

[Tebore]

We gotta keep this thing alive, there's a bunch of people here who still don't know about the wonderful world of alpha tweaking.

 

[felinusz]

pershoot

good job man some points to add:

a value of 4 for tW2P and tR2P seem to work best for agp and fsb stability by the same token. a matched value works best.

tREXT is most important in boosting throughput, but can also make or break your stability (depending on your fsb mark).

high agp controller latency (i.e. 255) helps in combating the agp 'stuffed up ness' that happens at higher fsb.

tRFC complements your throughput the most at any given fsb mark. this can also break your overall stability (may be stable in some areas but unstable in others, while going to complete instability across the board).

tRC helps balance out tRFC.

Auto Precharge of Disable adds to FSB stability with trivial loss in throughput.

Drivestrength of 6 and Slewrate of 7 work well for BH5.

Thanks dude, that's all useful information . I haven't been messing with DS/SR with my NF7-S as of yet, but I'll be starting in on that next, once I iron out the high-FSB stability issues I'm still having in 3D .

This is whack, my machine will run 3DMark2001SE looped at a 235 MHz FSB for ~2 hours, but will ultimately crash before the golden four hour mark. This is purely anecdotal, but I'm finding that the Lobby Tests are when 3DMark2001SE closes every time. I've got the AGP 1 latency set to 255, the AGP 2 latency set to 32 (stock), and the PCI latency set to 32 (stock); I'm still trying for all this with Alphas of 1-3-3-3-2-1-3.

 

[pershoot]

bump your agp up a notch and maybe a slight jolt in Vagp if needed. try agp bus latency of 16 once u bump up your agp up a notch.

good luck

 

[Tebore]

Try 1.7 VAGP it's a magic number for me. And bump both of the AGP latency settings up. It allows the AGP card to hog up more cycles which means higher 3D scores, I bumped up the PCI up to 64 and noticed my RAID array to be a bit more snappy.

 

[suurballe]

Hi,

Alpha timing is very interesting indeed. (bh-5 memory here too).
My only working high-fsb BIOS (merlin 2.01 CPC-ON) has defenetly
tweaked values for alpha timing!

Now question is how do you make them permanent in the
BIOS when you have found optimal alues. Does anybody have the exact
location of the alpha-timings in the BIOS file ?

Secondly im a little puzzled, but where can you change the drivestrengt + slewrate ? (Are they part of the alpha timings ??)

 

[felinusz]

Alpha timing is very interesting indeed. (bh-5 memory here too).
My only working high-fsb BIOS (merlin 2.01 CPC-ON) has defenetly
tweaked values for alpha timing!

Now question is how do you make them permanent in the
BIOS when you have found optimal alues. Does anybody have the exact
location of the alpha-timings in the BIOS file ?

I have no clue, BIOS modding is way, way more complicated then just changing a number. I'm pretty sure that to get customized Alpha Timings and DS/SR settings in your BIOS, you would have to ask Tictac or -=Merlin=- (or another BIOS mod god) to do some custom work for you. Unfortunately, last time I checked, Tictac and -=Merlin=- don't do custom work.



Slightly related, I've almost gotten 235 MHz FSB 3D stable. Invariably however, 3DMark closes on me after about ~30-~35 repetitions of the Lobby Tests (the tests that my machine is now always crashing 3DMark on at high FSB speeds - so I'm testing with the Lobby tests exclusively now). Very weird. I'm still tweaking, and working on it though.

It is a credit that Alpha tweaks have gotten me from ~228 MHz 3D/Prime stable, to 235 MHz *almost* 3D stable.

 

[suurballe]

Hi,

AquaMark is the most unstable here.
Crashes almost instantly with any other bios than Merlin 2.01 CPC-on
with FSb>=220

I had also the "lobby" problem but that seemed to be heat related !

 

[Miguita]

Does anybody know which modded NF7-BIOS has the tightest "default" Alpha-timings?

 

[Flip-Mode]

Extreamly interesting!
Thank you VERY much.

 

[Tebore]

Does anybody know which modded NF7-BIOS has the tightest "default" Alpha-timings?

The tighest they go is with the Abit stock config. Any tighter and you get worse overclocking, than the stock bios. These guys know what they are doing, they tune the timings to get high FSB without loosing too much 3D performance or bandwidth.

Besides if you go with a really timing certain ram will fail at stock speed.

 

[Runner30]

Alpha timings are good to check on different systems .
It would be great if BIOS modders could make a utility to patch a bios with custom alpha timings.

I know about similar tool for video bios- you load a bios file, set memory timings , save ,and flash.

 

[felinusz]

Hi,

AquaMark is the most unstable here.
Crashes almost instantly with any other bios than Merlin 2.01 CPC-on
with FSb>=220

I had also the "lobby" problem but that seemed to be heat related !

Very interesting, thanks for your input. I'm pretty sure that heat isn't my problem, my temperatures are all very solid. Getting past 228 is seemingly impossible right now, everything that I've tried crashes on Lobby within four hours, although some Alpha combinations do run longer than others. Disabling APIC seemed to help a little, but didn't lend any visible number gains. It seems that my NF7-S is simply not a stable >228 MHz FSB board . So much for sharing my overclocking gains in "chart format"

Runner30

It would be great if BIOS modders could make a utility to patch a bios with custom alpha timings.

That's a really fantastic idea. Unfortunately, on part because the platform is quickly becoming obsolete, I doubt that such a thing will ever be made for our nForce 2 boards.

Runner30I know about similar tool for video bios- you load a bios file, set memory timings , save ,and flash.

Intrigued, I've PMed you

 

[Runner30]

Intrigued, I've PMed you

Are you sure?

I've got no PM.

 

[felinusz]

Hmmmmmmmmmmmmmm.

Well, I sent it again

So, where can I grab a copy of that neat little program?

I could really use that

Take it easy,