1:1 VS 5:4...How much difference does it really make??? ***added test***

[Duvie]

I agree with your methodology but it appears going back to an fsb to run cas and GAT enabled is too far down at 1:1 to make up the detrimental effect of lower clock speed of the cpu....


Fact is as Oldfart pointed out this is a P4 system and does not follow same logic as amd if you are thinking of t through your experience with your amds...He also states that 90-95% benefit comes form cpu and only 5-0% comes from memory, hence why we run ou cpu out rst until the end in coing and not vice versa.....

I wish I could run more gaming apps for you to see if this holds true everywhere, but it appears in all of these apps I have been running CPU SPEED IS GOD!!!! It comes first and ram timings are merely what they usualy are a tweak to gain a few more points....

 

[oldfart]

He also states that 90-95% benefit comes form cpu and only 5-0% comes from memory

I pulled those numbers outa my arse. .

Dont know what they really are, but if anyone spends any time benching like Duvie has, they will see that CPU MHz is BY FAR the most important factor. Memory helps...some....but it is a very minor factor in comparison. Some websites (that are interestingly enough sponsored by memory manufacturers), push the idea that the big $$ DDR makes this huge difference. They test with artificial benches like SiSoft since real applications dont show any big gains. Once you go to a DC DDR P4 setup, they all run pretty well. 1:1, 5:4, timing setings are all pretty small tweaks in the overall scope.

I cringe when I see people spending more $$ for mem than CPU and mobo combined! The performance/$$ ratio is horrible.

 

[Duvie]

Originally posted by: oldfart

He also states that 90-95% benefit comes form cpu and only 5-0% comes from memory

I pulled those numbers outa my arse. .

Dont know what they really are, but if anyone spends any time benching like Duvie has, they will see that CPU MHz is BY FAR the most important factor. Memory helps...some....but it is a very minor factor in comparison. Some websites (that are interestingly enough sponsored by memory manufacturers), push the idea that the big $$ DDR makes this huge difference. They test with artificial benches like SiSoft since real applications dont show any big gains. Once you go to a DC DDR P4 setup, they all run pretty well. 1:1, 5:4, timing setings are all pretty small tweaks in the overall scope.

I cringe when I see people spending more $$ for mem than CPU and mobo combined! The performance/$$ ratio is horrible.

LOL!!! Well testing seems to show what you pull out of your arse is pure gems!!!!

Bottom line if ocing get as far as you can with you cpu to the limits of vcore and heat you feel safe with, all the while the memory is set to real conservative timings and ratio....then see what the memory will do at that speed..If you can go down less then 5fsb to gain stability and able to garner 1:1 versus 5:4 or 5:4 versus 3:2 cas 2 versus 2.5 or cas 2.5 versus cas 3 and maybe some Gat features (i875 mobos) then it may be beneficial.....Otherwise I don't see it being beneficial

 

[RaynorWolfcastle]

I agree, look at the results with my board in the few tests that I've run. Running 5:4@3.3 GHz beats 1:1@ 3.2 GHz! By comparison a computer with 3% more CPU speed but a whopping 28% less memory speed is faster than my baseline (3.2 GHz 1:1) in the tests! IMHO, the current P4s have all the bandwidth they can use, if anything they need lower latencies.

 

[orion7144]

Duvie, I take it you got the disk today?

 

[Duvie]

Originally posted by: orion7144
Duvie, I take it you got the disk today?

I got it saturday, but installed sunday...Thanks for the bonuses...Better not let the wife see those ones....

 

[orion7144]

Originally posted by: Duvie

Originally posted by: orion7144
Duvie, I take it you got the disk today?

I got it saturday, but installed sunday...Thanks for the bonuses...Better not let the wife see those ones....

I guess I am lucky my wife enjoys them

 

[Tom]

I've been running at 3:2 ratio and 275 FSB since I got my IS7, this saved me from having to buy new memory, I just still use some old Samsun pc2700.

And it runs plenty fast.

 

[Big Lar]

On my own system, I found that 1:1 yields about 400 more in bandwidth than 5:4

 

[Duvie]

Originally posted by: Big Lar
On my own system, I found that 1:1 yields about 400 more in bandwidth than 5:4

Yeah and so does mine...1:1 at 3.033ghz vs 5:4 at 3.2ghz....400+ points in mem score yet the test show it doesn't mean a better performing system...Unless same or near same overal clock speed it appears 1:1 will not be the fastest for the system...

 

[oldfart]

Originally posted by: Duvie

Originally posted by: Big Lar
On my own system, I found that 1:1 yields about 400 more in bandwidth than 5:4

Yeah and so does mine...1:1 at 3.033ghz vs 5:4 at 3.2ghz....400+ points in mem score yet the test show it doesn't mean a better performing system...Unless same or near same overal clock speed it appears 1:1 will not be the fastest for the system...

Exactly. People have to start looking at what real system performance is rather than a synthetic benchmark that does not represent real world performance. A PC is a device you own to run applications games, etc. It is not a mem BW benchmarking tool.

 

[FPSguy]

Originally posted by: Duvie
* * * Bottom line if ocing get as far as you can with you cpu to the limits of vcore and heat you feel safe with, all the while the memory is set to real conservative timings and ratio....then see what the memory will do at that speed..If you can go down less then 5fsb to gain stability and able to garner 1:1 versus 5:4 or 5:4 versus 3:2 cas 2 versus 2.5 or cas 2.5 versus cas 3 and maybe some Gat features (i875 mobos) then it may be beneficial.....Otherwise I don't see it being beneficial

According to the XbitLabs tests that I cited above (link), "at 275MHz FSB it is often better to use 336MHz memory with aggressive timings rather than 440MHz memory with RAS# to CAS# Delay and RAS# Precharge set to 3." In other words, according to the XbitLabs tests at the same FSB it is better to use fast memory at a 1:1 ratio even if the memory timings are not good. However, oddly, at the same FSB it is better to use a 3:2 memory ratio with fast timings rather than a 5:4 ratio with mediocre timings. This is not true in all applications under all circumstances, but according to the XbitLabs tests it appears to be true the vast majority of the time. As with the conclusion about the 1:1 ratio, the difference in performance is slight, but it is measurable.

 

[MikeMantho]

Oldfart,

I have just purchased a new P4 system similar to yours and I was wondering if you could point me in the right direction regarding memory timings and overclocking. My old PIII had no where near the multitude of BIOS settings relating to memory, much of the information I have read so far is contradictory or confusing. Is there a document I can read that gets to the heart of it? I tried to OC my system to 210 FSB w/ the Game Settings on Turbo and the machine would not POST, all voltages were stock and my memory was set to "by SPD". This really suprised me as the POST error related to BIOS being corrupt.

My specs are:

Abit IS7
P4 2.6c retail
Alpha 8942 w/ 80mm fan
Corsair XMS 3200 (2 x 512MB sticks)
Abit GeForce 4 4200 OTES

Any help would be much appreciated and I apologize in advance if this issue has been beat to death.
P.S. love your icon.

 

[RussianSensation]

***Same fsb and timings....no argument 1:1 is better****

Thats NOT true:


250 1:1 3-8-4-4 vs. 250 5:4 2-5-2-2
Same thing proven by a different website

In fact NOT ONLY IS 5:4/3:2 with tight timings better if the system originally has higher MHZ than than 1:1 system, but the system with tight timings OF THE SAME CLOCK SPEED AS 1:1 IS STILL FASTER

cheers

and by the way guys you can pick up 2x256 for 150-160 not $300 so the price difference is minimal between good 3500 and good 4000 memory

 

[RussianSensation]

Actually Duvie you are right my mistake i read that again same fsb and timings 1:1 is better but when are we ever gonna get same tight timings like 2-5-2-2 at 280fsb????? probably not for like 6 months

 

[Duvie]

Originally posted by: RussianSensation
Actually Duvie you are right my mistake i read that again same fsb and timings 1:1 is better but when are we ever gonna get same tight timings like 2-5-2-2 at 280fsb????? probably not for like 6 months

NO problem!! Thanks for the link...Always good to have more data to look at. This is a tough often combative topic here and other forums...I like the data to speak for itself...

 

[FPSguy]

Originally posted by: RussianSensation

***Same fsb and timings....no argument 1:1 is better****

Thats NOT true: 250 1:1 3-8-4-4 vs. 250 5:4 2-5-2-2 Same thing proven by a different website In fact NOT ONLY IS 5:4/3:2 with tight timings better if the system originally has higher MHZ than than 1:1 system, but the system with tight timings OF THE SAME CLOCK SPEED AS 1:1 IS STILL FASTER cheers and by the way guys you can pick up 2x256 for 150-160 not $300 so the price difference is minimal between good 3500 and good 4000 memory

I can't read Dutch or Russian, but as best I understand those articles the first one says fast timings are better even at the same bus speed, and the second one says it depends on the application. This article finds that 1:1 is usually faster (though not by much), even with slow timings.

 

[Duvie]

BUmp....For some of the newbies to read...

Got some good issues I see many still struggle with...

 

[SinfulWeeper]

Try it all @ 3.033ghz then divide the memory multiplier and CAS settings using all your same tests...
I'll decide if 1:1 is worth it. If there was conversation about this already, sorry. I do not have much time on the computer. I can only say that running the same core speed, dividers hurt a systems performance. Giving it a higher clock rate does not say anything about performance increases/degregation with multipliers for obvious reasons.

 

[Duvie]

Originally posted by: SinfulWeeper
Try it all @ 3.033ghz then divide the memory multiplier and CAS settings using all your same tests...
I'll decide if 1:1 is worth it. If there was conversation about this already, sorry. I do not have much time on the computer. I can only say that running the same core speed, dividers hurt a systems performance. Giving it a higher clock rate does not say anything about performance increases/degregation with multipliers for obvious reasons.

HUh??? I already did that...first post at the bottom...

I ran 3.033ghz with 1:1 then 3.033ghz with 5:4 and same timings as 1:1 then 3.033ghz with 5:4 and better timings due to the lower ram speed allowed this....

No one will argue that in almost all cases that at smae speed and same timings the 1:1 ratio is the best performance....HOwever this was to show that to achieve these sometimes very high ddr speeds we do this with a major sacrifice to timings and sometimes limit our potential higher overall clock speed....

 

[SinfulWeeper]

Originally posted by: Duvie

Originally posted by: SinfulWeeper
Try it all @ 3.033ghz then divide the memory multiplier and CAS settings using all your same tests...
I'll decide if 1:1 is worth it. If there was conversation about this already, sorry. I do not have much time on the computer. I can only say that running the same core speed, dividers hurt a systems performance. Giving it a higher clock rate does not say anything about performance increases/degregation with multipliers for obvious reasons.

HUh??? I already did that...first post at the bottom...

I ran 3.033ghz with 1:1 then 3.033ghz with 5:4 and same timings as 1:1 then 3.033ghz with 5:4 and better timings due to the lower ram speed allowed this....

No one will argue that in almost all cases that at smae speed and same timings the 1:1 ratio is the best performance....HOwever this was to show that to achieve these sometimes very high ddr speeds we do this with a major sacrifice to timings and sometimes limit our potential higher overall clock speed....

My bad, I missed that one .
How about on the same CAS settings? I am going to build a computer for someone who is... 'stubborn'. While I mostly agree with the tests and have respect for your work, I am affraid that he'll spend a good deal more needlessly buying PC4000 or higher when some good PC3500 will do nicely. Too bad memory fast memory is so 'slow' so we can not have the cake and eat it.

 

[Insane3D]

Excellent work Duvie!

 

[XeoBllaze]

My A64's 3Dmark2001 scores:

5:4 (166MHz DDR) scored 19.3k **ALL STOCK**

1:1 (200MHz DDR) scored 19.7k **ALL STOCK**

Quite a jump with just 33MHz faster memory, but I think it's due to PC2700 speeds actually bottneck the A64 quite a bit

 

[Duvie]

Originally posted by: XeoBllaze
My A64's 3Dmark2001 scores:

5:4 (166MHz DDR) scored 19.3k **ALL STOCK**

1:1 (200MHz DDR) scored 19.7k **ALL STOCK**

Quite a jump with just 33MHz faster memory, but I think it's due to PC2700 speeds actually bottneck the A64 quite a bit

Actaully 166mhzx2 for 333ddr vs 200mhzx2 for 400ddr...It is actually 66mhz faster overall.....You need to tell us the cas timings cause if they were both rather similar then it really shows the 66mhz or in this case 20% increase did not equate anywhere near that in 3dmark....400pts in the 19000's means like 2.2%...not much and pretty neglible.

Run other apps cause I don't trust 3dmark or any of those futuremark apps to give true representations of real world performance....