High fsb vs tight timings on 875P

[WizardB]

High fsb vs tight timings on 875P

Hardware used:

Abit IC7-G
Intel 2.6C @ 3250 Mhz (13*250)
OCZ 3700 Gold
9700 Pro


GAT as follows for both 1:1 and 5:4 tests

Game Accelerator: AUTO
Refresh Cycle Time: Normal
Read Delay: AUTO
Read Delay Adjust: Disabled
Command Per Clock: Disabled

Setting all at AUTO didn’t gain any performance

Programs used:

Scisoft Sandra 2003
SuperPI
3DMark01
PCMark2002
Quake3

250 fsb is 1:1 and 200 fsb 5:4 setting on the memory

Settings Sandra Buffered Int
200 fsb - 2 6 3 3 5435
200 fsb - 2.5 7 4 3 5372
200 fsb - 3 8 4 4 5349
250 fsb - 2.5 7 4 3 5947
250 fsb - 3 8 4 4 5899


Settings Sandra Buffered Float
200 fsb - 2 6 3 3 5419
200 fsb - 2.5 7 4 3 5370
200 fsb - 3 8 4 4 5371
250 fsb - 2.5 7 4 3 6035
250 fsb - 3 8 4 4 5935

Settings Sandra Unbuffered Int
200 fsb - 2 6 3 3 2645
200 fsb - 2.5 7 4 3 2626
200 fsb - 3 8 4 4 2491
250 fsb - 2.5 7 4 3 3285
250 fsb - 3 8 4 4 3097

Settings Sandra Unbuffered Float
200 fsb - 2 6 3 3 2707
200 fsb - 2.5 7 4 3 2634
200 fsb - 3 8 4 4 2482
250 fsb - 2.5 7 4 3 3286
250 fsb - 3 8 4 4 3154

Settings SuperPI 1M
200 fsb - 2 6 3 3 43 s
200 fsb - 2.5 7 4 3 44 s
200 fsb - 3 8 4 4 44 s
250 fsb - 2.5 7 4 3 42 s
250 fsb - 3 8 4 4 42 s

Settings 3DMark01
200 fsb - 2 6 3 3 16593
200 fsb - 2.5 7 4 3 16321
200 fsb - 3 8 4 4 16296
250 fsb - 2.5 7 4 3 16732
250 fsb - 3 8 4 4 16718

Settings PCMARK2002 CPU
200 fsb - 2 6 3 3 8073
200 fsb - 2.5 7 4 3 8053
200 fsb - 3 8 4 4 8047
250 fsb - 2.5 7 4 3 8082
250 fsb - 3 8 4 4 8074

Settings PCMARK2002 MEM
200 fsb - 2 6 3 3 10148
200 fsb - 2.5 7 4 3 10108
200 fsb - 3 8 4 4 9844
250 fsb - 2.5 7 4 3 10800
250 fsb - 3 8 4 4 10712

Settings Quake3 Timedemo
200 fsb - 2 6 3 3 270,4
200 fsb - 2.5 7 4 3 264,3
200 fsb - 3 8 4 4 262,1
250 fsb - 2.5 7 4 3 276,4
250 fsb - 3 8 4 4 274,9





Conclusion

It’s quite obvious that you can use as tight timings as you want and still not be able
to win over high memory bus in overall performance.

So 285 fsb 1:1 at 2.5 7 4 4 for an example will outperform 285 fsb 5:4 at 2 6 3 3,
even 285 fsb 1:1 at 3 8 4 4 will outperform 285 fsb 5:4 at 2 6 3 3.

I can’t say anything about 2 5 2 2 timings since OCZ 3700 Gold doesn’t like them very much, but I doubt they will be so much faster then 2 6 3 3 timings that they will change this conclusion.

So if you plan to run 270-290 fsb use pc4000 memory and run them somewhere between 2.5 7 4 4 and 3 8 4 4 at 1:1, and if you can go above 300 mhz fsb use 5:4 divider with the best timings you can get since the CPU speed that you gained will be worth running 5:4 and loose some memory performance there, overall performance will be better due to the higher cpu clock.

The cpu is the bottleneck when it comes to memory performance so therefore it’s not worth lowering your fsb more then 5-10 mhz or something like that to be able to run 1:1.

[retrospooty]

Originally posted by WizardB

It’s quite obvious that you can use as tight timings as you want and still not be able
to win over high memory bus in overall performance.

Its not too obvious, that same test at 2-2-2 at 5:4 would easily beat 1:1 at 3-4-4 , even at 5:4 2-3-3 I am able to run the same test and its equal or beats 1:1 at 3-4-4. I have the same mobo as you... Its possible your ram doesnt like 2-3-3 very much either, or the the IC7 doesn't like your particular ram at 5:4.

If you ever get your hands on some bh5 chip ram try it again, youll see.

[daddy_fizz]

Originally posted by retrospooty
Its not too obvious, that same test at 2-2-2 at 5:4 would easily beat 1:1 at 3-4-4 , even at 5:4 2-3-3 I am able to run the same test and its equal or beats 1:1 at 3-4-4. I have the same mobo as you... Its possible your ram doesnt like 2-3-3 very much either, or the the IC7 doesn't like your particular ram at 5:4.

If you ever get your hands on some bh5 chip ram try it again, youll see.

exactly correct. i proved this myself, even though not through sandra memory bandwidth benchmark. In a real life situation (UT2003)...

I had a-data PC4000 3-4-4-8 running at 1:1 260FSB vs. Mushkin PC3500 running at same FSB with 5:4 2-2-2-5

my sandra scores were a little worse with the mushkin but in gaming performance the mushkin beat out that a-data by a couple of FPS.

~Fizz

[macci]

So 285 fsb 1:1 at 2.5 7 4 4 for an example will outperform 285 fsb 5:4 at 2 6 3 3,
even 285 fsb 1:1 at 3 8 4 4 will outperform 285 fsb 5:4 at 2 6 3 3.

2.5-7-4-4 is just as fast as 3-8-4-4.

I can’t say anything about 2 5 2 2 timings since OCZ 3700 Gold doesn’t like them very much, but I doubt they will be so much faster then 2 6 3 3 timings that they will change this conclusion.

Going from 2-6-3-3 to 2-5-2-2 gives you quite similar boost than going from 3-8-4-4 to 2-6-3-3. So yes it would have made quite a difference if you had run those tests at 2-5-2-2 instead of 2-6-3-3.

[WizardB]

Hmm ok, I guess I have to try with some bh5 memory on my system then

I'm planning to run fsb around 310 mhz so I will have to use 5:4 divider so the memory will run at 248 mhz.
It would be really nice here with some memory that can handle
2 2 2 5 at 248 mhz

Any recommendations?

[daddy_fizz]

bh-5 can but you have to pump some serious volts through it. the closest you will get without high voltage is ocz pc37000 gold at 2-3-3-6, if you get some good sticks


~Fizz

[WizardB]

OK I understand, now I just have to choose between this to setups...
I don't have any nice LL memorys that I can test now so if anyone have tried something like the two following setups please posts some results or comments about which setup that is the best one...

1: 300-310 fsb at 5:4 with 3700 gold running at 240-248 fsb 2.5 3 3 7
2: 300-310 fsb at 3:2 with xms 3500 c2 running at 200-207 fsb 2 2 2 5

[MeowChow]

Originally posted by WizardB
1: 300-310 fsb at 5:4 with 3700 gold running at 240-248 fsb 2.5 3 3 7
2: 300-310 fsb at 3:2 with xms 3500 c2 running at 200-207 fsb 2 2 2 5

I suspect that the difference in performance between 3:2 low-latency and 5:4 high-latency would favor 5:4. My reasoning is that the bus would not be saturated in the 5:4 case, as it would be in 1:1, so there would be no diminishing returns from increased DDR speed.

[toymaker]

I just compared my Corsair XMS3500C2(winbond BH-5) to a OCZ PC4000EL kit. With a P4 2.4C at 275 Mhz, I ran the Corsair @ 2,2,2,5/5:4/440 Mhz, 3.1v. The OCZ was run at it's best timings 3,4,4,7/1:1/550 Mhz and wouldn't take any more than 2.85v.

PC Mark 2002:
Corsair = 11357
OCZ = 10587

Sandra Mem:
Corsair = 5545/5547
OCZ = 6014/6017

UT2003:
Corsair = Flyby-273.6/ Botmatch-93.4
OCZ = Flyby-267.7/ Botmatch 87.6

By the way......the OCZ's best timings at 450 Mhz were 2.5,3,4,7. The Corsair is now running at 2,2,2,5/454.9 Mhz stable. Keep in mind though that the Corsair is using old Winbond BH-5 which is getting more and more difficult to find.

Don't mean to bash OCZ. Because for a new purchaser it's the best out there IMHO.

In my case which would you keep? (check my sig)

[retrospooty]

Originally posted by toymaker
I just compared my Corsair XMS3500C2(winbond BH-5) to a OCZ PC4000EL kit. With a P4 2.4C at 275 Mhz, I ran the Corsair @ 2,2,2,5/5:4/440 Mhz, 3.1v. The OCZ was run at it's best timings 3,4,4,7/1:1/550 Mhz and wouldn't take any more than 2.85v.

PC Mark 2002:
Corsair = 11357
OCZ = 10587

Sandra Mem:
Corsair = 5545/5547
OCZ = 6014/6017

UT2003:
Corsair = Flyby-273.6/ Botmatch-93.4
OCZ = Flyby-267.7/ Botmatch 87.6

By the way......the OCZ's best timings at 450 Mhz were 2.5,3,4,7. The Corsair is now running at 2,2,2,5/454.9 Mhz stable. Keep in mind though that the Corsair is using old Winbond BH-5 which is getting more and more difficult to find.

Don't mean to bash OCZ. Because for a new purchaser it's the best out there IMHO.

Which would you keep? (check my sig)

I'd say its obvious... If you're goal is the highest possible synthetic memory scores (Sandra memory bench), go with PC4000.

If your goal is the fastest performance in real applications, go with PC3500LL ram.

Its that simple !

[eva2000]

Originally posted by toymaker
I just compared my Corsair XMS3500C2(winbond BH-5) to a OCZ PC4000EL kit. With a P4 2.4C at 275 Mhz, I ran the Corsair @ 2,2,2,5/5:4/440 Mhz, 3.1v. The OCZ was run at it's best timings 3,4,4,7/1:1/550 Mhz and wouldn't take any more than 2.85v.

PC Mark 2002:
Corsair = 11357
OCZ = 10587

Sandra Mem:
Corsair = 5545/5547
OCZ = 6014/6017

UT2003:
Corsair = Flyby-273.6/ Botmatch-93.4
OCZ = Flyby-267.7/ Botmatch 87.6

By the way......the OCZ's best timings at 450 Mhz were 2.5,3,4,7. The Corsair is now running at 2,2,2,5/454.9 Mhz stable. Keep in mind though that the Corsair is using old Winbond BH-5 which is getting more and more difficult to find.

Don't mean to bash OCZ. Because for a new purchaser it's the best out there IMHO.

In my case which would you keep? (check my sig)

personally i'd keep both LOL

i have 2x 512MB XMS3500C2 and 2x 512MB OCZ PC4000 Gold (lent to me by a friend for extensive testing )

on 4PCA3+ i could get my XMS3500C2 to 240mhz 2-2-2-6 at 3.2-3.3v vdimm stable unfortunately 4PCA3+ bios isn't optimised and benchmarks don't reflect in the timings used

now i just got a P4C800-E and the bios are alot more optimised for bandwidth my OCZ PC4000 Gold i have here are doing very nicely

[hedge]

Very good thread. I think this will be helpful in deciding what to get when buying my intel stuff.

[o27]

So far I´ve managed to get 295fsb 1:1 with my A-data pc4000 - aiming for 300fsb 1:1
Here´s my Sandra for 295 1:1 cas 3/8/4/4 3.2vdimm

[Major]

Originally posted by rro
So far I´ve managed to get 295fsb 1:1 with my A-data pc4000 - aiming for 300fsb 1:1
Here´s my Sandra for 295 1:1 cas 3/8/4/4 3.2vdimm

300 is a nice round number !

[o27]

yes indeed any suggestions why I can´t run SuperPi at 295 1:1? could it be my gfx that´s the bottleneck?