2.6c @ 3.0 (1:1) or 2.6c @ 3.2 (5:4)?

Demonfire

Weaksauce
Joined
Mar 17, 2004
Messages
120
I can get my processer to run at 3.0 using a 1.1 ratio (fsb=231) OR I can run it at 3.2 using a 5:4 ration (fsb=198)

Which is better. Seeing as teh Pentium 4's tastiest snack is memory bandwith, im leaning towards the first value.

Which one is better, and what is my pc3200 ram running at in the first one?
 
i'd take the second - you get more mhz and more fsb and the P4 doesn't need ALL the bandwidth 1:1 provides it gets plenty at 5:4

and what memory timings are we looking at when you run 5:4?
 
ahh default for a p4 is 200fsb why are you running at 198 at 5:4?
your crippling your machine.
 
I think he got confused.
198 would make perfect sense for the ram. At a 5:4 that would put FSB at 247.5.
247.5x13 =3217.5MHz

Considering that your ram runs at 231MHz, I would be supprised if you can't tighten your timings a bit at 198MHz.



Btw, 1:1 means FSB = Memory frequency.
 
YOu want the 5:4 ratio. The P4 craves FSB bandwidth the most, followed by low latency and memory bandwidth.

At 5:4 ratio your ram should be right about with in spec so you should be able to run with some tight memory timings.
I know the 5:4 ratio will be faster. (period)

+ 200Mhz of CPU power
+ 62Mhz extra QFSB (quad pumped FSB = X 4 )
+ lower timings (tighter)
-------------------------------------
= higher performance


at the same exact speed its often a toss up between the 2 ratios... but you're getting an extra 200Mhz CPU, so you definitely want that. Run some benchmarks!!! Test the apps/games that you use.

Asking people which will run better on YOUR computer is like calling up people and asking which artwork will look better in your house. Its in your house.... we dont know all of the exact specifics.... test it yourself. (or which food taste better to you)
 
I agree with the bench testing. try them both and see what you get. Personally I run my ram at 1:1, but right now it's actually below spec at 200Mhz (rated for 250 at 2.5, 4-4-7) so I can tighten the timings a bit to 2, 3-3-6 and set GAT to F1, E-5-E-E.


hey chrisf6969, I notice you have an Aeroflow. Did you ditch the tmd fan? If not, what are your temps?

I just got this AI7 and 3.0C (not o/cing yet...still dual priming 24/7 for a couple weeks to get a good burn in so everything is stable and I have a base to go off of) and my cpu temps is 55load, PWM is 63load. I was wondering if I made an adapter and put a 92mm fan w/shroud on the aeroflow hs if it would make a good bit of difference.
 
At the same timings and FSB, 1:1 is literally 1-3% faster at best, this gap closes as you tighten timings at 5:4 compared to 1:1. Go for the 5:4 ratio. For the record, 3:2 at the same timings is about 2-4% slower than 5:4 also.
 
5:4 will introduce additional latencies, and since the memory speed is slower, i would fully expect it to have craptacular performance in certain instances.
 
Originally posted by jeef
5:4 will introduce additional latencies, and since the memory speed is slower, i would fully expect it to have craptacular performance in certain instances.

YOu have absolutely no idea what the f*&k you are talking about!! Maybe with an old KT133a or KT266, etc... running at ratios other than 1:1 gave you lower performance.

But with 875/865 @ dual channel DDR the only performance hit is from running the ram slower.... but usually tighter timings completely offset that. Some people (like Jeef) are so clueless. I wish I saved all of the links that show proof in benchmarks. The only thing that really likes 1:1 better is Sisoft memory bench.

BTW, I've run ram at 4:5 (ram faster than FSB) with a 533FSB cpu and it improved performance like 1% over 1:1
 
jeef said:
5:4 will introduce additional latencies, and since the memory speed is slower, i would fully expect it to have craptacular performance in certain instances.

you really have no clue........

pretty much everything you just said is bs.
 
Run some benchmarks. The 1:1 would probably get better result in memory bandwidth tests, as long as the timings are halfway decent. And the possibility of getting 2-2-2-5 or 2-2-2-6 is unlikely unless you have BH-5/BH-6 memory.
 
definately the 5:4 with the higher fsb and higher cpu speed will beat out 1:1 at the lower speed hands down.
 
chrisf6969 said:
YOu have absolutely no idea what the f*&k you are talking about!! Maybe with an old KT133a or KT266, etc... running at ratios other than 1:1 gave you lower performance.

But with 875/865 @ dual channel DDR the only performance hit is from running the ram slower.... but usually tighter timings completely offset that. Some people (like Jeef) are so clueless. I wish I saved all of the links that show proof in benchmarks. The only thing that really likes 1:1 better is Sisoft memory bench.

BTW, I've run ram at 4:5 (ram faster than FSB) with a 533FSB cpu and it improved performance like 1% over 1:1
ok, it seems i struck a nerve, and although you went on the defensive you couldn't post some evidence? im not if you read the thread title, but we're talking about 4:5 or 1:1. i know you're just baiting me, but perhaps you thought i wouldn't respond.
 
computerpro3 said:
you really have no clue........

pretty much everything you just said is bs.
how incredibly insightful. thanks for livening up the debate! i mean, you made it pretty clear that what i said is wrong, and im sure others will read your post and become convinced as i have.
 
Fine.

In 3dmark 01 and 03, 5/4 wins. In superpi and pifast, 5/4 wins. In game benchmarks such as ut2003 and splinter cell, 5/4 wins. In everything except sisoft sandra memory bandwith benchmark, 5/4 wins.
 
those are all cpu tests, of course 200mhz will make a minor difference, enough for you to say they 'win'
 
and what friggin real world application relies SOLELY on memory performance? The VAST majority of tests, games, applications, etc. depend on a mix of cpu and memory. That is why 5/4 will consistently beat 1/1. and the differences aren't that minor, about 1000 points in 3dmark 01, 2 seconds in superpi (a ton for that benchmark), etc. etc. Its a pretty big gain.

And for that matter, most of those are NOT CPU tests. how stupid can you be...you called 3dmark 2003 a cpu test....wow... and 3dmark 2001 is a whole system bench. unreal...ever tried playing that with a geforce mx and a 4ghz chip? What, you still can't play at 1600x1200...hmmm, wonder why? Everything you are saying is baseless.

If you think latiencies do not matter, I have a question for you... how come there is NO ONE on the front page of the Online Result Browser for 3dmark 01, or NO ONE on xtremesystems.org/forums in the top 20 that uses high latency pc4000 or the like? How come ram that can do low latency is $400+ a gig while kingston pc4000 is around $280-300? Do you know more about computers than Macci, Oppainter, Fugger, JcViggen? I think not. You're wrong.
 
5:4 here also...with tight ram timings....it sure makes up for the higher clocks with loose timings at 1:1.....
 
Basically it comes down to CPU speed vs. Memory bandwith.

Ive gone with 3.2 at the moment, but when I get water cooling later in august, everything will be much better.
 
dont assume you will get any better overclock with water. I had my 2.6c at 3.25 with a vantec aeroflow and decided to reinstall my water cooler (external) temps dropped 10C under full load but still would not go any further. :rolleyes:
 
Really, the P4 has no use for all the bandwidth, particularly if you are in dual channel mode. Theoretically, the 1:1 ratio with dual channel would saturate precisely 100% of the P4's needs, though it would only be useful if the P4 needed all that bandwidth capacity. Thus in practice, the gains are minimal. From experience, I'd say go with 5:4 and nice tight timings.
 
kirbster said:
dont assume you will get any better overclock with water. I had my 2.6c at 3.25 with a vantec aeroflow and decided to reinstall my water cooler (external) temps dropped 10C under full load but still would not go any further. :rolleyes:


Well, I have good reason to believe that water cooling will help. My northwood IDLES (note: idles) at around 50 when oced (Im using a really crappy vantec cooler, but remember this is a northwood)

You dont even want to know the load temps (they often break 75c)

So i think its safe to assume that even though my proc can stand high heat unusally well (its been running at these temps for many months) while keeping an ok overclock, cooling it down some will help it oc like crazy.
 
As i said my current full load temps are 33C max 30 C average according to Asus Probe and MBM5. They were 44C max 41 average with the aeroflow. Thats is a whopping 11 C decrease and i cant overclock even 1 mHZ more than before. The 2.6C chips seem to top out around 3.25-3.3Ghz. I assumed the same thing you do and i was wrong. It used to be that way but as i am finding out from lots of reading and my own experience, sometimes they just hit a wall and thats it.
Good luck.
 
Back
Top