LL vs High Bandwidth

[CaBoOse999]

I've been wondering what applications benefit more from high bandwidth and ones that benefit more from LL. Ive been overclocking my Athlon 64 3200+ Winchester which hits the wall at 2.55 Ghz. I can either run it at 300 x 8.5 = 2550 Mhz with my memory at 250 Mhz at 2.5-3-3-5 1T or 255 x 10 = 2550 Mhz with my memory at 212 Mhz at 2-3-2-5 1T. I havent noticed any great differences in gameplay or several other applications. I just need to know what benefits from LL and what benefits from high bandwidth. I'm leaning towards high bandwidth because of the higher fsb speed. Thanks.

 

[Gamingphreek]

Latency matters very little for the A64 due mostly to its on die memory controller.

Check out Zebos Memory Matrix thread. It will explain a lot to you.

-Kevin

 

[Duvie]

I think you have that backwards gamingpreek...or read it wrong...Bandwidth means very little, it is cas latency where you are going to see the differences

according to the numbers the PC3800 consistently beats the PC4700...The pc4700 would have the higher bandiwdth yet the pc3800 has cas 2.5,2,2, and thus latency wins....Also note how PC3000 (370ddr) a whole whopping 200+ddr slower but with cas 2,2,2 actually comes close to beating the pc4700 in a few apps....the proof is look at pc3000 versus pc3000 in several of the apps...cas 2 versus cas 3.....

It makes differencein superpi, it is just Zebo chose to do the 1mb where difference are not really seen (too short)...In 32mb it can be very huge...I have seen 14-16 seconds drops just going from 2.5,4,4,10 to 2.5,3,3,10....

 

[Wedge1]

Duvie, for a P4 system, would games (Doom 3, Half Life 2) benefit more from a high overclock using a divider, or more from a smaller overclock that is running 1:1?

Taking my PC, P4 2.8C I can get:
3.36GHz using a divider of 3:2
3.15GHz running 1:1

 

[THUGSROOK]

Originally posted by: Wedge1
Duvie, for a P4 system, would games (Doom 3, Half Life 2) benefit more from a high overclock using a divider, or more from a smaller overclock that is running 1:1?

Taking my PC, P4 2.8C I can get:
3.36GHz using a divider of 3:2
3.15GHz running 1:1

its close, damn close 😉

need more info ~ cas settings for each ratio and what motherboard it is. (makes a big difference)

🙂

 

[thanatos355]

Originally posted by: Duvie
I think you have that backwards gamingpreek...or read it wrong...Bandwidth means very little, it is cas latency where you are going to see the differences

according to the numbers the PC3800 consistently beats the PC4700...The pc4700 would have the higher bandiwdth yet the pc3800 has cas 2.5,2,2, and thus latency wins....Also note how PC3000 (370ddr) a whole whopping 200+ddr slower but with cas 2,2,2 actually comes close to beating the pc4700 in a few apps....the proof is look at pc3000 versus pc3000 in several of the apps...cas 2 versus cas 3.....

It makes differencein superpi, it is just Zebo chose to do the 1mb where difference are not really seen (too short)...In 32mb it can be very huge...I have seen 14-16 seconds drops just going from 2.5,4,4,10 to 2.5,3,3,10....

yeah, but when you can have BOTH, what do you get? that's what was missing from 'the matrix'. 😛 i want some high bandwidth with reasonaly low latency. say 2.5-3-3-8-1t @ ddr-533-550+. i guess once i get my darn xp-120, i can post some numbers for you guys. 😉

 

[miketheidiot]

Originally posted by: Duvie
I think you have that backwards gamingpreek...or read it wrong...Bandwidth means very little, it is cas latency where you are going to see the differences

according to the numbers the PC3800 consistently beats the PC4700...The pc4700 would have the higher bandiwdth yet the pc3800 has cas 2.5,2,2, and thus latency wins....Also note how PC3000 (370ddr) a whole whopping 200+ddr slower but with cas 2,2,2 actually comes close to beating the pc4700 in a few apps....the proof is look at pc3000 versus pc3000 in several of the apps...cas 2 versus cas 3.....

It makes differencein superpi, it is just Zebo chose to do the 1mb where difference are not really seen (too short)...In 32mb it can be very huge...I have seen 14-16 seconds drops just going from 2.5,4,4,10 to 2.5,3,3,10....

i did some benchies and at 2.4 ghz it was faster in general at 240X10 with 2.5-3-3-10 than with 2-3-2-10 and 200X12

 

[KDKPSJ]

Originally posted by: miketheidiot
i did some benchies and at 2.4 ghz it was faster in general at 240X10 with 2.5-3-3-10 than with 2-3-2-10 and 200X12

Isn't it because it has higher FSB (HTT)? Why don't you try 240x10 with 1:1 divider and 2.5-3-3-10 vs. 240x10 with 5:4 divider and 2-3-2-10? If you try that, I believe you will have similar result with Duvie.

 

[Duvie]

Originally posted by: miketheidiot

Originally posted by: Duvie
I think you have that backwards gamingpreek...or read it wrong...Bandwidth means very little, it is cas latency where you are going to see the differences

according to the numbers the PC3800 consistently beats the PC4700...The pc4700 would have the higher bandiwdth yet the pc3800 has cas 2.5,2,2, and thus latency wins....Also note how PC3000 (370ddr) a whole whopping 200+ddr slower but with cas 2,2,2 actually comes close to beating the pc4700 in a few apps....the proof is look at pc3000 versus pc3000 in several of the apps...cas 2 versus cas 3.....

It makes differencein superpi, it is just Zebo chose to do the 1mb where difference are not really seen (too short)...In 32mb it can be very huge...I have seen 14-16 seconds drops just going from 2.5,4,4,10 to 2.5,3,3,10....

i did some benchies and at 2.4 ghz it was faster in general at 240X10 with 2.5-3-3-10 than with 2-3-2-10 and 200X12

I did benches as well and ram at or near 400ddr with cas 2,2,2,6 was as fast as 480ddr cas 2.5,3,3,5.....Ram at 533ddr at cas 2.5,3,3,8 is slightly faster.....BUt when one considers the stress on the controller the vdimm it takes and even possible the notch extra a vcore that may be needed it may not be worth it....

 

[THUGSROOK]

i ran enough benchies on enough P4 systems to tell you this...
youll need an extra 10fsb (using the same or better cas settings) to make up for going 5:4 or it aint worth it.

P4 also has the ability to OC the NB latency which does tend to confuse things quite a bit. the best way to describe it, is that P4 NB OCing is worth 45fsb for your memory. neat stuff 🙂


AMD however is compleely different due to no NB 😉

 

[imported_whatever]

I hope you realize that you are not actually running it at 300*8.5 when it is set to it. It takes the end product (2550) and divides it by the next highest multiplier (9) to get 283.3MHz FSB. It then takes the 6:5 memory ratio (If it is that way as opposed to a CPU/something) and determines the RAM speed.

 

[CaBoOse999]

Originally posted by: whatever
I hope you realize that you are not actually running it at 300*8.5 when it is set to it. It takes the end product (2550) and divides it by the next highest multiplier (9) to get 283.3MHz FSB. It then takes the 6:5 memory ratio (If it is that way as opposed to a CPU/something) and determines the RAM speed.

Theres a huge argument about this existing or not. According to AMD it isn't supported but there is evidence going both ways. Interesting subject.

 

[Wedge1]

Originally posted by: THUGSROOK

Originally posted by: Wedge1
Duvie, for a P4 system, would games (Doom 3, Half Life 2) benefit more from a high overclock using a divider, or more from a smaller overclock that is running 1:1?

Taking my PC, P4 2.8C I can get:
3.36GHz using a divider of 3:2
3.15GHz running 1:1

its close, damn close 😉

need more info ~ cas settings for each ratio and what motherboard it is. (makes a big difference)

Well, actually I meant which *should* be best for gaming (Doom3 and HL2) . But assume the timings are 2.5-4-4-8 on both (because they are) even though I know I could lower them when using the divider; not sure if I can with it 1:1.

I'm running the P4P800-E Deluxe and Corsair XMS RAM.

I guess really what I'm asking is whether or not I would benefit most from the highest FSB I can acheive, or the highest speed I can acheive while running 1:1, even though it will certainly be lower than what I might reach using a RAM/FSB divider. I'm thinking the guidelines to follow are different from that of an AMD64 system.

 

[imported_whatever]

Originally posted by: Wedge1

Originally posted by: THUGSROOK

Originally posted by: Wedge1
Duvie, for a P4 system, would games (Doom 3, Half Life 2) benefit more from a high overclock using a divider, or more from a smaller overclock that is running 1:1?

Taking my PC, P4 2.8C I can get:
3.36GHz using a divider of 3:2
3.15GHz running 1:1

its close, damn close 😉

need more info ~ cas settings for each ratio and what motherboard it is. (makes a big difference)

Well, actually I meant which *should* be best for gaming (Doom3 and HL2) . But assume the timings are 2.5-4-4-8 on both (because they are) even though I know I could lower them when using the divider; not sure if I can with it 1:1.

I'm running the P4P800-E Deluxe and Corsair XMS RAM.

I guess really what I'm asking is whether or not I would benefit most from the highest FSB I can acheive, or the highest speed I can acheive while running 1:1, even though it will certainly be lower than what I might reach using a RAM/FSB divider. I'm thinking the guidelines to follow are different from that of an AMD64 system.

are there any ratios available between 1:1 and 3:2?

 

[Wedge1]

Originally posted by: whatever

are there any ratios available between 1:1 and 3:2?

Yes I think a 10% overclock is 5:4 (will double check but I'm pretty sure).

 

[THUGSROOK]

Originally posted by: Wedge1
Well, actually I meant which *should* be best for gaming (Doom3 and HL2) . But assume the timings are 2.5-4-4-8 on both (because they are) even though I know I could lower them when using the divider; not sure if I can with it 1:1.

I'm running the P4P800-E Deluxe and Corsair XMS RAM.

I guess really what I'm asking is whether or not I would benefit most from the highest FSB I can acheive, or the highest speed I can acheive while running 1:1, even though it will certainly be lower than what I might reach using a RAM/FSB divider. I'm thinking the guidelines to follow are different from that of an AMD64 system.

my previous reply still applies ~ 10fsb (using the same cas settings) and its worth it to cut a ratio. if you can gain better cas settings with the ratio then it might be even lower, 6 or 7fsb.

🙂