Norz

.....agree,who needs them

Attached Thumbnails

 

__________________
Phenom II X2 550 Testing in progress
Phenom II X4 965 Testing.........


OC team Norway Norz aka Nassuz

bigvaL

The best case scenario is 1:1 at a decently high speed running tight timings. People used to think that high ram mhz was all that mattered. Try it for yourself. Your bandwidth WILL go down but overall speed will increase.

Ropey can explain this very well if needed.

For SuperPI, high speed is all that matters. For real world performance, tight timings, 1:1. Unless of course that would mean very slow RAM.

__________________
FOLD FOR OCN - TEAM 37726
CS:S Server - GunGame 100tick Ping Boosted 32 Slot: overclock.nuclearfallout.net:27015
OC Settings: 3800mhz 181x21, 1.275vcore, 1.65vdimm.

Ropey

Given two systems with identical hardware and the only difference being the CPU multiplier, FSB MHz, and DRAM MHz/Timings:

200MHz (FSB) X 13 (Multiplier) = 2.6Ghz with 4:5 (CPU : DRAM = 400Mhz:500Mhz) Ratio.

236MHz (FSB) X 11 (Multiplier) = ~2.6Ghz with 1:1 (CPU : DRAM = 472Mhz:472Mhz Ratio.

Now, both system run at 2.6GHz but you will find System #2 quite a bit faster than System #1 although in some "Canned Cache and Memory Subsystem" benchmarks it might seem as though System #1 is faster since the DRAM is running higher. Now if both systems were identical except for only the DRAM bandwidth, then there would be an small but appreciable difference with System #1 winning out minimally against System #2. This is testing with a digital stopwatch and loading of games such as F.E.A.R. etc. Loading a 20MB graphic rendering in Photoshop and rendering it to Lightwave 3D, etc.

The closer that these two system's FSB get, and the further away that these two systems DRAM moves will only increase this above premise.

In other words, all other things being equal, 1:1 is not as fast as 1:2 in canned cache driven benchmarks, but in reality the difference is minimal and if you tighten the latencies of the DRAM at a lower and 1:1 ratio, you will find that often this is faster. You want the highest CPU tied to the highest bandwidth with the lowest latency. You do not want to drop your FSB heavily to gain a higher DRAM throughput and you do not want to loosen your DRAM latency at the expense of the DRAM throughput.

__________________
Please do NOT IM unless it is Important! PM is the way to go
http://valid.canardpc.com/show_oc.php?id=536554
How Do I Remove Unused Devices From Windows XP/2003/64bit?
My Heatware

Carni4

Well, i wasnt talking about 1:1 or 4:5 being faster, but that faster running memory with loser timings is faster than slower running memory with tighter timings.

I've also read that if your default multi is 8(in my case) en u set a multi of 7. Your northbridge speed is FSB x (default multi/set multi).

For instance 400 x (8/7) = 457.

__________________
e6400@3,4GHz | Team Xtreem DDR2-800 @ DDR2-1160 | Cheap way to cool your 8800 | My buildlog

MADMAX22

thats a true statement with the p965 chipsets, if you read up on the boot strap issue with these boards it also explains why the 400fsb can sometimes be a reason why it doesnt alow you to use that setting but 405 or like 390 will work, yeah you will get better performance out of your mobo if you use the next lower multi setting on these mobos,

with amd i know that speeds that are tighter in timings will get you better results, with my opti 170 at 2.9 (290x10) and my ram running at 200 fsb with 2-2-2-5 settings and 1T i would get quicker times in everything but if i ran 290 on the ram with 2.5-4-3-7 1T i would get great bandwith but the speeds would be lower (atleast with stuff like pi)

havnt had my conroe enough to test the differences out yet with this chip so i cant say with these guys

just my 2

__________________
E6400@3856E4300@4158E6600@4510,Opti148@3102,opti170@2918,Xeon3220(B3)@3430(490fsb)
UD3P@8400600FSB

F3t1sh

pun?

__________________

..:: Quebec Overclockers - Linskingdom's VMark - Canadian OCN Club ::..

31D8M4a1r1k

..:: My Quad ::..

Carni4

Ok, this is the best I can do on 2,45V vdimm and 4:5 divider:


Let's how's how well it does 1 on 1.

__________________
e6400@3,4GHz | Team Xtreem DDR2-800 @ DDR2-1160 | Cheap way to cool your 8800 | My buildlog

gonX

Drop it to 1:1 and 6x, and see what you'll get.

__________________
THE Mouse FAQ | 32-bit Resolution Fix | Important Information
64-Bit Driver Signing Fix | The Infraction and Warning System
My Anime Progress | The HoN Discussion Thread

Please direct all tech related questions to a thread in the respectable forums, and not to my PM inbox. Thank you

CWell1337

Did you not even READ what Ropey posted?! Running tighter timings at a 1:1 divider (meaning a slower speed because it would be running at your FSB speed) will perform better than looser timings and having speeds of in excess of 1000Mhz for example.

The benchmarks you are using to test this out (Everest, SuperPi, etc...) are all THEORETICAL tests. Not reality. They are synthetic tests. Meaning you will never see that kind of performance really.

I like to think of it like this:

Lets say you have a passenger train, and the speed of the train represents your RAM speed. The timings of the RAM are represented by how quickly the passengers can get on and get off at each stop (tighter timings being better). And lets say that the rate at which potential passengers get to each station is your CPU FSB.

With higher RAM speed and looser timings, then the train (speed representing RAM speed) gets to each station at 1000Mhz, but still has to wait for passengers from the CPU who only get to the stations at 400Mhz (FSB speed). Then because of the looser timings, the passengers take longer to get on and off the train before the train can leave for the next station.

With lower RAM speed (equal to your CPU FSB) and tighter timings, then the train gets to each station at the same moment that potential passengers arrive (400Mhz). But with the tighter timings, the passengers can get on and off quicker.

So with the higher train speed, the train gets to the station faster, but it still has to wait for passengers. But if the trains speed is equal to the speed at which passengers show up to the station, then it can start the passenger exchange at the same time as the other train, and with tighter timings, be done first and off to the next station first.

Thats just how I like to think of it, and it's actually a pretty accurate analogy. BUT bottom line... slower RAM speed (1:1 ratio) and tighter timings will perform better IN REALITY than blazing RAM speeds (which have to be set on a divider) and looser timings.

Hope that helps draw a picture for you.

__________________

Carni4

Thanks Cwell, that does draw a very clear picture.

__________________
e6400@3,4GHz | Team Xtreem DDR2-800 @ DDR2-1160 | Cheap way to cool your 8800 | My buildlog