Found registers for performance at high FSBs...

[nfrenay]

I was testing stuff here to discover what is going on with the performance problem at 185 and I've tested the system at 175FSB.

I might try to discover at which FSB exactly EPoX changes these registers for lower performance (read: higher stability for lower quality components).

At 175FSB performance is on par compared to 166 so I took a WPCREDIT screenshot at 175 and then at 185 (using same CPU speed so that I could compare benchmarks).

I've discovered that among 5 or more registers changes, there are 2 of them that are the most effective (at least in the two tests I'm using mostly).

At 175MHz:
6B[6] = 1
95[4] = 0

And at 185MHz:
6B[6] = 0
95[4] = 1

When I booted at 185MHz and manually set those two registers like the ones at 175MHz (faster), the benchmarks reacted as expected. Performance increased a lot. The PCMark2002 memory benchmark are:
- 185MHz "original" = 3801
- 185MHz "+6B[6]=1" = 3986
- 185MHz "+95[4]=0" = 4002

SuperPi original is 1m03s, then 1m02s and finally 1m01s in the sequence as above.

I tested all this with CPU @ ~1670MHz.

You can get more info and some nice graphs at:
FSB comparision and tweaking (zipped excel file)

Apparently my system is rock solid with the registers tweaks at 185MHz...

Can anyone tell me what you think and could ET try to answer why they do this in BIOS and also if there will be a BIOS that fixes this ?

I'm using 2312 BIOS for 8k3a+...

PS: I didn't run 3dmark because it's nonsense with a voodoo3 3000 pci...

[Holst]

Very nice.

I very much doubt that there will be a new bios with the features you want.
But through using WPCRSET you can get most of your speed back.

you should appreciate that the number of users who run over 166fsb is extrememly small, and the number that understand that the bios slows down slightly as the FSB increases is even smaller.
Its not practical for Epox to make specific bios for small groups like us. They have better things to do with there time, there not just writing bios for the new boards but for old boards for new standsards (UDMA-133) and stuff.

[Áedán]

It would be interesting to know what bit 6 of register 6B actually does in the KT333 chipset. The KT133 and KT266 used it as part of the AGP arbitration parking policy (bits 6&7).

Not to cast too much doubt, but the result from bit 4 of register 95 is less than half of one percent, which might well be within the error margin. You did run the benchmarks more than once to get the figures, right?

AidanII

[Pinky]

Originally posted by AidanII
AGP arbitration parking policy (bits 6&7).

Not to cast too much doubt, but the result from bit 4 of register 95 is less than half of one percent, which might well be within the error margin. You did run the benchmarks more than once to get the figures, right?

Aidan, I'll pay for your plane ticket if you promise to split the Jeopardy winnings

(ie--I'll sponsor you for a trivia show! )

BTW, I have now way of verifying if anything you're saying makes any technical sense, but if you can BS that well you'd probably be equally good at a gameshow

[Furor]

At 175MHz:
6B[6] = 1
95[4] = 0

And at 185MHz:
6B[6] = 0
95[4] = 1

I don't get this..is that 95 = 40/41 or what?

I've sort of been doing the same thing, here's what i have come up with..

Possible values to change to >

80 0F
F0 00
46 00
66 00
7a 00
6c 60

Values already changed >

92 E9
66 00
71 48
93 FF
94 5F
61 FF
55 84
95 0E

[SteveI]

Originally posted by Pinky


Aidan, I'll pay for your plane ticket if you promise to split the Jeopardy winnings

(ie--I'll sponsor you for a trivia show! )

BTW, I have now way of verifying if anything you're saying makes any technical sense, but if you can BS that well you'd probably be equally good at a gameshow

Wanna split the airfare?? He'd make a killing in Burbank,CA. :-D :-D

[Furor]

oops

[Wolf]

A lot of people have noticed this especailly at high fsb. I guess we have been owned by the Epox bios engineers. This board is an ok board but it will be my last Epox for quite a while.

[Nico67]

Originally posted by nfrenay
I might try to discover at which FSB exactly EPoX changes these registers for lower performance (read: higher stability for lower quality components).

179 good, 180 bad!. I guess thats reasonable considering DDR333 has effectively got 13 mhz overhead, if your ram is good enough. It would be nice to see a bios that was aimed at DDR400 capable ram, even if the board doesn't officially support it. Maybe further down the track

[nfrenay]

Thanks for the info Nico...

Just FYI I've set the two registers I said in the first post in WPCRSET and I'm up and running with performance... I can't post at 200MHz here and I don't even want to as I'm trying to find a spot between performance and safety. So far my decision is on 180 or 185MHz FSB.

It's really sad that Epox does this kind of thing. As I said before, if I wanted MHz I would have bought a P4. I want the most out of my MHz so I hope the new BIOSes fixes the CPU command problem (lack of option) and also this problem.

I haven't even tried another BIOS other then 2312 (I've modded a 25?? and tried it but before I took my processor to unlock I flashed the 2312).

But in the other hand, the new BIOSes have the nice PCI/AGP clk display and also some people complain about stability with 2312. I seriously think that anyone with 2312 stability problems have weaker hardware that can't handle the "performance" settings.

Also I don't think running 200MHz FSB on a "slower" BIOS would be better then running 180MHz on 2312 (supposing the performance is the same) as you would only gain problems because of the too high PCI/AGP clock speeds.

Sorry if I exagerated in this post but I needed to tell what I think.

[Chong345]

200fsb with 23.12? How did you do it? What settings? Are the timings softened. I have xms3000 and at 192 or more I get problems.

[nfrenay]

I think my post is a bit confusing but I surely stated "I can't post at 200MHz". My point is that I don't care to run at 200MHz using the newer and slower BIOS if I can run faster in real-world using 2312 BIOS and a 185MHz FSB or something near that.

One more point in favour of running a lower FSB with higher performance/mhz is that your PCI/AGP cards would be less "overclocked" so we have less IDE problems and NIC problems. And no AGP corruption (garbage on screen).

[[N8]]

Depends what you want your PC to be faster at really

I was able to get the 2619 to produce better sandra memory scores @ FSB's above 200MHz than the 2312 @ 185...but I have yet to find a BIOS to beat the 2312's 3D performance.

[Chong345]

nfrenay, are you saying to use these wpcredit tweaks with 23.12? After 185 the rigster soften up the timings so I need to adjust them?

[nfrenay]

That's pretty much it... From 180MHz (as Nico told us) the 2312 BIOS soften things up... I would also recommend for experienced users to try these two tweaks with the "slower" BIOSes as they might increase performance.

[Chong345]

I'm gonna do some testing. I'm gonna try another bios to see if it can be my 23.12 3dmark score. I'll use the wpcredit tweaks. Thanks for the info.

[nfrenay]

Keep us posted...

I will try to learn some stuff this month and maybe I'll make a tweaking software for KT333 (with EPoX on mind as it's my PC)...

[Nico67]

Originally posted by [N8]
Depends what you want your PC to be faster at really

I was able to get the 2619 to produce better sandra memory scores @ FSB's above 200MHz than the 2312 @ 185...but I have yet to find a BIOS to beat the 2312's 3D performance.

Precisely!, it's nice having those high sandra mem scores but it's only an artifically measurement. 3Dmark is more of a real would test that is showing there's more to consider than just bandwidth. To get that bandwidth you are more often than not, sacrificing latency, which is why you get higher 3Dmark scores at lower fsb.
I run 2312 @ 179 and it gives me the best result without having to turn down the timings. I could probably achieve the same results with a later bios at 200+ fsb, but whats the point?

[[N8]]

My only problem is that 179FSB wouldn't work out for me CPU MHz-wise.....dam, I need .1 multiplier increments :P

[nfrenay]

Originally posted by [N8]
dam, I need .1 multiplier increments :P

LOL me too...:-D