Encoding speed+performance benefits

[The Therion]

Hi,

I just wanted to ask if anyone knows which offers more performance in video encoding : more cpu speed , or more memory bandwidth ?

Thanks

[mrlipring]

more than what?

your question can't be answered unless we know what you have at the mo. after so much memory, it won't matter. If you've got something daft like 64mb, the amount of swapping will be insane, and no matter which cpu you have, it'll be deathly slow, but once you get to a gig or more, it won't make much difference, the cpu will be the bottleneck.

It depends what you're doing, ie, how many passes at the same time, what you're doing etc, but aim for at least half a gig, and as fast a cpu as you can get.

I've got a pretty fast cpu, but only 256mb (cos i'm skint) so when encoding (and playing games, and...) the pc swaps quite a lot. When i get some cash, i'm going for AT LEAST another 256mb. I might hold off and go for a gig. It should be pretty sweet then.

[The Therion]

Oh,no, i mean one more than the other in the SAME system.

I have the following :

P4 2.4 HT @ 3300Mhz
2x256MB 3500 RAM
ABIT IS7 (865)

So i could run at 3Ghz with more memory bandwidth (but obviously less cpu speed).

That's why i'm asking.

Thanks

[jggimi]

Moved to PC HW/SW forum.

[mrlipring]

just realised you said memory BANDWIDTH, not plain old memory.

looks like you've got a half-decent overclock there. is that with the multiplier, or/and fsb? if it involves fsb, i'll assume you're running the memory in sync, so there should be quite a performance improvement there, more so than if you were running at the same overclocked total speed, but by using multipliers, not fsb. If you're using only multiplier overclocking at the mo, try and get the fsb up instead/as well, because that's where the real performance gains are.

of course, none of this matters if it ain't stable, but that's for another thread ;)

you're asking about only a 10% difference in overall cpu speed though, i'd be inclined to say that the difference between a high fsb and a lower fsb but higher multiplier at 10% difference ain't gonna be much. i'd do some benchmarking at both settings, and see how you do. ideally, you'd have as high an fsb as your memory, cpu and mobo can handle, THEN start mucking about with the multiplier. you'll have to take the fsb down a notch or 2 probably, but when you reach the point where the pc doesn't like the overclock, that's the time to benchmark, to see whether the lower multi and higher fsb, or vice versa, produces best results.

sisoft sandra should be half decent to benchmark, but real world benchmarks can vary, so do a bit of encoding.

and get prime95 if you don't already, every overclocker needs it. :)

[The Therion]

Uuuhhmmm....Intel cpus have locked multipliers...

This overclock is achieved by running the memory at 4/5 of the FSB.My FSB is 275 for now.So the memory is @ 220 (440Mhz).

But how i do it doesnt matter much since i can measure my memory bandwidth with sandra or other programms.

I guess noone has tried this with video encoding.Oh well...i will.



PS: i'n still overclcocking ,so i'm using about 15 related programms, including Prime95.Thanks :)

[mrlipring]

i really wouldn't run the memory out of sync. if it's pc3200 you've got (220mhz isn't too bad for quality ddr400) then bring down the fsb to match it. most of the time running the memory and fsb asynchronously will slow the machine down a bit.

If it's >pc3200 then bump up the memory if you can.

didn't know intels had locked multipliers. that's a bit poo. oh well. no way of unlocking them?

[The Therion]

No.The only way to have an unlocked Pentium 4 , is , if it is an engineering sample from the factory.

My memory is 3500,and right now i'm searching its limits and at what set up it delivers the most bandwidth (and it will probably be probably asynchronously).The differences of 1:1 and 5:4 (when the memory has the same speed) are small.

I have to test it.
Just got my Tv tuner today.