overclocked PCI-Clk G450 or G550 better?

[snn47]

Having the choice between these cards and looking to incease my PCI-Clk for improved Video performance on my ABIT KT7A (de-encoding MPEG1 +2), which one would you recommand?

[Dr Mordrid]

I'd leave it alone, especially on that mainboard...which is not a paragon of stability to start with so OC'ing it isn't necessarily a good idea.

MPEG encoding/deconding is mainly a CPU/memory/MMX/SIMD function, but first you need stabilty.....

Dr. Mordrid

[The PIT]

I would agree with that advice on abit kt7a.

[snn47]

unless overheated it seems to be quite stable

I have to add that I have been running performance tests to map performance of my KT7A-Raid with various BIOS settings (YH) with ISA-Soundcardand IEEE-1394 drive.

I used 570MB AVI (PAL 720*570) to MPEG1 encoding (TMPEnc) for >2 weeks and didn't encounter any stability tests, unless the Northbridge (?) overheated.

So nowthat I have an idea which tweaksworkbest I am looking for the better PCI performance of the 2 cards.

[Greebe]

What is more fundemental would be the speed the Gcards core is clocked at. Upping the bus speed won't really add diddle to improving their performance.

So our suggestion is leave it be or get a faster cpu. But if you want to push the bus UP, UP and AWAY.. either are fairly stable. (tho some makes of HD's and other periphials are not!)

------------------------

note: tho some HD's and other periphials are not so tolerant so it's give it a whirl and find out firsthand kinda thing

[snn47]

If I don't find ...

a way/the right Bits in the BIOS to activate the Mem-Clk =FSB+PCI also above 122MHz FSB.

I'll replace the Clk-GenPCIout with an external 33MHz PCI-Clk. From my tests this should be the same as adding 100-200MHz to the CPU.

Cooler is better esspecially when you have the same performance arround 1000-1200MHz with hard & soft tweaks to the board.

[Wombat]

I'll replace the Clk-GenPCIout with an external 33MHz PCI-Clk. From my tests this should be the same as adding 100-200MHz to the CPU.

You're going to do WHAT? Oh sure, just replace the PCI clock. It's not like it's designed to only work synchronously with the north bridge.

Either I don't understand you, you don't know what you're talking about, or both.

[snn47]

I don't think so

Since I could not find any source stating that the clocks have to be synchronized/phase-locked I am just going to try.

The only reason I could find is that the PLL205 has FSB/Mem/PCI-ratios locked toeach other, which is the reason why the PCI will get to high once you increase the FSB. What kind of fixed phase relation is there with all the fine tuning in 1 MHz steps, then Mem with or without PCI added, if I am wrong please correct me.

If you were right then I should already have problems when I just insert a not phase-locked external 33MHz instead of the 33MHz generated by the PLL205 at the 133_133_33 setting and then we'll see if it works.

Otherwise I have to Phase-Lock the signal.

It wouldhave been easier if the 133 FSB 133+PCI=Mem 33MHz PCI would be accessableinthe BIOS.
Using the 122+11 FSB-setting and Mem+PCI works and give methe desired Mem>FSB,

but the PCI is then to high for my components

[Greebe]

Pardon me but, wouldn't it be cheaper and far less time consuming (and by far less headache) to slap in a 1.4gHz Tbird which is less than $100 and get better times than all your tweaking, moding and incompatability issues (Oh did I mention wasted time)?

[Greebe]

Actually it can be done... but this is the wrong forum to discuss such matters... sooo... Kenny!

[Kruzin]

Moved to GHW/SW

[Greebe]

I does require a PLL to make this mod work... such a "kit" can be had (just assemble with included pcb).. cost ~$100USD... but as I have said earlier it's cheaper to go with a 1.4gHz Tbird and nix all the potential problems in the bud... and still get faster times

[Technoid]

Is he atempting to replace Clock generators??

Herd of someone in 99' that claimed that he had replaced clockgenerators on a BX bord....

[snn47]

Unless I find a source with the knowledge

I started doing this tests because I want to understand how to estimate performance based on the the variation of the BIOS and bandwidth of FSB-/PCI- and Mem-Bus as well as the latencies of RAM and HDD influences my performance.

& I rather do it with my KT7A now before spending my money unwisely on a thouroghbred and real 333 CL2,5 RDRAM once available.

I am going to use it mostly for video so normal benchmarks don't help me estimate what gain I can expect. Most test just give us benchmark-number, and if they use encoding you can't reproduce it. Numbers are just a snappshot! Using the next/previousversion of the benchmark or varying the CPU/MEM/PCI-Clk will give different figures so what good are benchmarks. Time seems to be the only constant. I can even use a stopwatch if my hardware clock variations render the encoding-time information provided by programms useless.

While using the TMPEnc and varying the encoding setting ( CBR/VBR/2pass, AVI/MPEG1/2) I could also vary the load on the Mem/PCI/CPU and still have an result that I can reproduce results to +- 1 to 2 s.

No that I have at least a feeling how all is realted it is easier to do further tests. Just adding a 1400 doesn't increase stability
unless I also improve cooling for the rest of the board e.g. chipset + 3phase regulator

[snn47]

Back to the original question if I may even so the question is in GHW now

what limitatations do the Matrox cards have -
are the the card clock signals, locked to the AGP and increase with AGP-Clk, is that what you are saying?
What max AGP would the G450/550 accept?

@Greebe
if you know what frequency you want, a crystal or a small XTAL-Osc for around 2$ are cheaper then building a PLL.

@Technoid
I don't want to do it, I have been doing it on all PC since my first Apple ][+ in both ways (under-/Overclock), whatever was needed