Some data about high/low FSB comaprison...

[Docta_Z]

If I keep the ram at that lower FSB, I could tighten the timings on that biatch and get even higher scores...(dare I say it?) with a lower fsb.

Obviously tighter timings will increase 3dmark scores since it would allow faster processes as well.

Finding the right balance between timings and fsb is the way to go. That's why people shell out so much to get fast AND low latt ram.


What I mean by the "made for your board" comment is - I don't know much about epox but I know with Abit - they have memory companies and particular models of ram that are "optimized"/recommended and work the best in that particular motherboard. I thought perhaps this might be something to factor into account as well.

236 is still pretty high fsb - that with tight timings is pretty flashy. It's all about tweaking to get the best possible results, but regardless of how fast you can make your timings, it's given that fast timings at 230 will beat fast timings at 200 in the way of 3dmark scores.

 

[dreIU]

Obviously tighter timings will increase 3dmark scores since it would allow faster processes as well.

Finding the right balance between timings and fsb is the way to go. That's why people shell out so much to get fast AND low latt ram.


What I mean by the "made for your board" comment is - I don't know much about epox but I know with Abit - they have memory companies and particular models of ram that are "optimized"/recommended and work the best in that particular motherboard. I thought perhaps this might be something to factor into account as well.

236 is still pretty high fsb - that with tight timings is pretty flashy. It's all about tweaking to get the best possible results, but regardless of how fast you can make your timings, it's given that fast timings at 230 will beat fast timings at 200 in the way of 3dmark scores.

I see what you are saying by "made for your baord," but I have neither an Abit nor an Epox (DFI!).

I won't be sticking at 236, just testing... definitely leaning towards 250 with high voltage (and some decent timings, tho I found they do not make a lick of a difference in gameplay).

I am a little aggravated at my EB tho... damn things wont be stable at 250fsb anymore. 245 will pass memtest86 for 13straight hours, but after 43 minutes at 250, it gets error after error. I don't want to jump to 3.3v until I put some sort of active cooling on those biatches, BUT I will definitely be getting rid of them if they can't do 250 at that voltage... my friggen processor will go beyond 250, I am positive. oh well... more testing I suppose

 

[Docta_Z]

I see what you are saying by "made for your baord," but I have neither an Abit nor an Epox (DFI!).

I'm blind sorry.

Are you saying you used to get 250 on the ram?

The jump to 3.3 requires a volt mod eh?

 

[JigPu]

Performance is a delicate balance you must achieve between two factors, one of which is based on two to three factors as well You need to find an optimal tradeoff between CPU speed and memory bandwidth. Memory bandwidth is affected by FSB, timings, and ratios. Change any one of these things, and your performance will also change.

What you've shown is that just FSB makes some difference, but not enough to really merit better memory (in your oppinion). But FSB is only a third of half the equation Like you said, if you were to tighthen up timings at the same speed, you could probably pull out even more. Only OCing FSB, or only OCing CPU, or only lowering timings will all only make a small difference. You need to find the best combination you can find, which usually means getting good memory since having the CPU limit your performance SUCKS (my CPU is locked, and so most attempts to OC my box end up with only tiny performance increases )

JigPu

 

[Psykoikonov]

Psykoikonov, what exactly do you mean by "the fsb is waiting for the agp?" I understand what you mean by the percentage increase, BUT I still must argue against all the people that told me to lower the multiplier and up the FSB. Apparently, it doesnt make a valuable difference.

What I'm saying is that the fsb is so much faster than the AGP (or something in between) that you notice no gains. It's not the fsb that's holding you back, whatever it is starts to really show above 200MHz. If I run a prime95 bench @ 200x12.5 or 250x10 there is a huge difference in avg times taken. Even when crunching SETI the fsb alone gave me a good 25-30 minutes off a work unit from 166x15 (~avg 2:20) to 250x10 (~avg 1:55). However my frame rates in games only increased noticably from 166 to 200...after that from 200-250 is not really noticable. I would like to see a P4 perform the same sort of tests you did to either prove or disprove my theory. I've seen this on ALL the gaming systems I've built (all AMD, about 10 of them now). The other thing is +/-50 3DMark points is well within tolerance, so if your only performing 1 run then 120xx is the same as 125xx, same goes for Aquamark. I've booted in with 12 processes (XP Pro) running ran 3DMark and then did the exact same thing again 5 minutes later and got a different score by 100 points easy.

Psyko

 

[Vengance_01]

Ya but a 1000 3dmarks is what 5FPS max is most games. IMO Whats he saying for people who really don't care for high marks, PC3200 should be fine. Far Cry is a great example, I bet he would gain next to nothing. I mean if you have the money, I would say hell ya but if it means skimping on a vid card, cooling ect.. its just not worth it

 

[germanjulian]

nice one! makes me feel better about my 166Mhz bus

dam old ram wont oc by 2 Mhz and again

but hey dont need it

 

[PCGUY112887]

Is this running in DDR or what?

 

[germanjulian]

Is this running in DDR or what?

sorry but can you read the first post. what else runs at 166Mhz up to 250Mhz FSB?
and 3200 ram sticks and 3400 ram sticks etc only exist in DDR.

sometimes I wonder what the people with all those stars under their name acctually post and comment

AMD Mobile Barton 2600+ IQYHA 0351
DFI Lanparty Ultra B
OCZ PC3700EB @ stock timings -->8-2-3-3
ATI Radeon PRO->XT @ 425.25/375.25

 

[dreIU]

Docta_Z, yeah 250 on the ram. It was at 2.9volts prime stable for well over 12 hours. I played Call Of Duty for over 5 hours straight on night (when we got sent home from work ) and no crashes. After that, it started rebooting randomly, and after memtest86, it showd many many errors at 250fsb even after 45ish minutes of running it. I then tested at 233 (stock) to see if the parts were defect, but it passed for 12hours. I did not go farther than 245 on them at 3.1 volts for 13hours straight on memtest86. I will see what I can do with 3.3volts. Oh, and no it does not require a volt mod, that is the max the Lanparty goes, woot!

JigPu, you make an excellent point about all the factors involved. BUT, I must say that while running my memory at 233fsb with 5-2-2-3 (CAS3 bc from what I have been reading, these chips are made to run like that and not take a hit from it) and gained an insignificant amount of points while benching with Sandra (while comapred to stock timings of 8-2-3-3 @ 233fsb). Same goes for testing at 245 fsb with stock and tighter timings. I am not sure why the difference is not there (ar at least, not a noticeable difference).

So, I am not too worried about not pushing my CPU past 245 at the moment, since it seems not to bring any worthwhile results. I am a little mad that I cannot do it (actually, that my RAM cannot do it), but since the benefits are not there, I am not too worried about it.

Psykoikonov, thank you for your comments and the data you have provided. The jump from 166 to 200 is way bigger than the jump from 200 to anything else.

Vengance_01, exactly.

PCGUY112887 & germanjulian, yes it is in Dual Channel DDR memory.

Please feel free to post any more comments.

 

[Psykoikonov]

dreIU: What kind of Sandra Ram Bandwidth are you getting...you should notice a huge increase from 200->250 in that test. MIPS and MFLOPS have no bearing on fsb, only the CPU speed matters for that test.
BTW
@ 250x10=2500 I get 9547 MIPS 3945 MFLOPS
Int: 3729MB/s Float: 3423MB/s
@ 216x11=2376 I get 9058 MIPS 3747 MFLOPS
Int 3158 MB/s Float 2855 MB/s
@ 211x10.5=2215 I get 8466 MIPS 3501 MFLOPS
Int: 3016MB/s Float: 2802MB/s
@ 166x15=2490 I get 9574 MIPS 3949 MFLOPS
Int: 2275MB/s Float: 2103MB/s

using SISoft Sandra 2004 SP1


Psyko

 

[dreIU]

dreIU: What kind of Sandra Ram Bandwidth are you getting...you should notice a huge increase from 200->250 in that test. MIPS and MFLOPS have no bearing on fsb, only the CPU speed matters for that test.
BTW
@ 250x10=2500 I get 9547 MIPS 3945 MFLOPS
Int: 3729MB/s Float: 3423MB/s
@ 216x11=2376 I get 9058 MIPS 3747 MFLOPS
Int 3158 MB/s Float 2855 MB/s
@ 211x10.5=2215 I get 8466 MIPS 3501 MFLOPS
Int: 3016MB/s Float: 2802MB/s
@ 166x15=2490 I get 9574 MIPS 3949 MFLOPS
Int: 2275MB/s Float: 2103MB/s

using SISoft Sandra 2004 SP1


Psyko

@245*10 ---> 10161 MIPS 3857 MFLOPS

memory ---> IN 3745 mb/s, Float 3492 mb/s


Gimme a little bit to do the other ones...

edit: 8-2-3-3 tmings

 

[PCGUY112887]

I meant if it's running in dual channel!
I see you have 2 sticks... you can either run them in singal channel which is less stressful on the memory and you can get a higher FSB, or you can run them in DDR and get a lower FSB overclock but more memory bandwidth....

I'm not THAT stupid!

 

[dreIU]

I meant if it's running in dual channel!
I see you have 2 sticks... you can either run them in singal channel which is less stressful on the memory and you can get a higher FSB, or you can run them in DDR and get a lower FSB overclock but more memory bandwidth....

I'm not THAT stupid!

hehehehe, no one is calling you stupid

Psykoikonov, as far as those results go... it will be a little while, I accidentally tried to boot at 233x12, and my motherboard takes things SOOOO seriously... hehehe I must now wait a while with battery out and blah blah blah DFI blah.

EDIT:

OK, I wont do the CPU test anymore, just the memory bandwith with the following fsb's 200,233,245.

200 @ 8-2-3-3 ---->INT 3081, Float 2918

200 @ 6-2-2-3 ---->INT 3089, Float 2929

233 @ 8-2-3-3----->INT 3563, Float 3292

233 @ 6-2-2-3----->INT 3564, Float 3323

245 @ 8-2-3-3----->INT 3745, Float 3492

245 @ 6-2-2-3----->INT 3749, Float 3513


The jump from 200 to 245 is a good increase, IMO. The jump from stock to tighter timings is minimal, IMO... So it appears as though, with this particular setup, higher fsb (with stock timings) is to be wanted more than lowe fsb with TIGHT timings... comments?

 

[PCGUY112887]

Then I guess that means you are running it in DDR

 

[Docta_Z]

What I'm saying is that the fsb is so much faster than the AGP (or something in between) that you notice no gains.

AGP frequency is locked on nforce2 boards.

 

[dreIU]

AGP frequency is locked on nforce2 boards.

So... this was always a subject that confused me. I have my AGP freq. set to 66mhz in my bios, yet I have no idea why... just because people have been saying that it needs to be at 66.

Is it good or bad that it is locked?

PCGUY112887, yes, dual channel it is.

 

[DumpALump]

So... this was always a subject that confused me. I have my AGP freq. set to 66mhz in my bios, yet I have no idea why... just because people have been saying that it needs to be at 66.

Is it good or bad that it is locked?

PCGUY112887, yes, dual channel it is.

agp specs are to run at 66mhz. Running it out of spec might cause it to crash or lock up. Same with PCI spec is 33mhz. Back in the days there wasn't any locks so often you would have to buy certain modems/network cards/soundcards that could handle a higher PCI bus.

 

[Psykoikonov]

Then I guess that means you are running it in DDR

DDR = Double Data Rate in single channel
Dual DDR = Double Data Rate in dual channel

 

[Psykoikonov]

AGP frequency is locked on nforce2 boards.

Only locked where ever I set it (50Mhz-100Mhz I think in 1MHz intervals)

The point I'm trying to make is the limiting factor in 3D benches is not the fsb. It's probably the AGP slot itself or GART controller that's holding you back from getting a higher score with increased fsb. I have run out of spec AGP and I did get gains...but @ 250MHz fsb I can't get 3D stable without backing off to 66Mhz AGP.

dreUI: Nice RAM bandwidth with that EB, your matching me with lower fsb and looser (if that's a word) timings!! Have you tried 2.5-2-2-6??