Dividers, Sandra, Perf., are they related....lets see

[Radelon]

I came across something really weird on the Abit IC7........

1) 2x256 GeIL PC3200 Ultra with CPU FSB at 255 and the mem with 5:4 divider Performance mode Fast = 18,233 in 3dmark2001 (2700 unbuffered Sandra) 2-6-3-3

2) 2x256 GeIL PC3200 Ultra with CPU FSB at 255 and the mem with 5:4 divider Performance mode Disabled = 17,489 in 3dmark2001 (2600 unbuffered Sandra) 2-6-3-3

3) 2x256 OCZ PC3700 Gold with CPU FSB at 255 and the mem with 1:1 divider Performance mode Disabled = 17,983 (3400 unbuffered Sandra) 2-7-3-3

4) 2x256 OCZ PC3700 Gold with CPU FSB at 255 and the mem with 1:1 divider Performance mode Fast = No boot (will not boot at any timings)

5) 2x256 OCZ PC3700 Gold with CPU FSB at 255 and the mem with 5:4 divider Performance mode Disabled = No 3dmark (2900 unbuffered Sandra) 2-6-3-3 (will not run 3dmark at any timings)

6) 2x256 OCZ PC3700 Gold with CPU FSB at 255 and the mem with 5:4 divider Performance mode Fast = No Boot (will not boot at any timings)


I'm sure you noticed that Config #1 produced the Highest Score in 3dmark2001 and that's with about 700mb/sec less Sandra Mem bandwidth than the #3 config. Only different is that Config #1 has Performance mode set to Fast and I couldn't set it even with the slackest timings in config #3.

I'll definitely do a more thorough round of tests once I get my replacement sticks from OCZ.

*note, my OCZ Gold may have some SPD issues, and the issues are being noted by OCZ as the shipping sticks have been fixed.

But still, comparing #1 to #2 you see that just enabling Performance mode to Fast gave me an 800 point increase in 3dmark2001, but goin' from 5:4 (#2) to 1:1 (#3) only gave me 500 points. So technically if my OCZ does have these problems, then my replacements should not and I should be able to run Performance Mode Fast in the 1:1 config and gain another ~800 points in 3dmark2001 for around 18.7-18.8k score.

I'll have more updates as they come around.

(I added mem timings to all of the configs that were applicable. Sorry I just forgot about them since the differences were only the tRAS. I also did a lot of copying and pasting and realized that the #5 config had the wrong data. Sorry for any misconceptions, I hope it's all fixed now. Also, remember that my OCZ chips are not the norm and most likely nobody else with this ram is having the problems I'm having. OCZ is looking hard into this and are going to ship me some new chips that do not have this issue. So at the time I receive the new chips, I will run the same tests and see if different results occur)

[Duvie]

I am more perplexed about configs #3 and #6!!! How does the one with the faster memory speed work and not the slower one??? Timing issues aside if it is the same as #3 there should be no issues. After the no boots do you reset the cmos again???

[oldfart]

One thing that is missing from your testing is cas values. You need to state what they were set to for each of these tests. If you are set to disabled, what cas values were you using? Same with "fast".

I've been saying this for awhile. You are better off running 5:4 with tight ram timings than 1:1 with slow timings. I saw pretty much the same thing with my IC7 / 2.6C testing (http://www.bleedinedge.com/forum/showthread.php3?s=&threadid=132).

You can run 5:4 with fast timings, run a higher CPU speed (typically) and use cheaper memory and get ~ the same performance as 1:1 with slow timings and $$ memory.

[pastorjay]

I saw similiar things with my IC7. The only thing that scored better at 1:1 was Sandra unbuffered... Have not even tried 1:1 yet on this p4p800...

PJ

[Radelon]

I just finished editing the post. It now has more detailed information, sorry for overlooking that information guys.

[oldfart]

I dont know about you guys, but I dont really play much "Sandra Unbuffered". Those numbers dont mean squat.:rolleyes:

[Radelon]

Originally posted by oldfart
I dont know about you guys, but I dont really play much "Sandra Unbuffered". Those numbers dont mean squat.:rolleyes:

Technically, no, they don't...... but I could give you the Memtest read-outs, but Sandra just reflects what Memtest says. At any rate, using Sandra Buffered tells you nothing. Especially when using the same fsb, the difference is so little you can't explain it.

By all this, I'm showing that setting Performance mode to Fast will give you a better boost than going from 5:4 to 1:1. Now if you could have both, the system would scream.

[ThugsRook]

i dont mean to rain on your parade but.....

i personally feel that Sandra does not work correctly and your scores are incorrect.

not only that, but that whole world is basing 800fsb/springdale/canterwood performance on sandra ~ and its WRONG!

i strongly suggest you stop using sandra ~ its just confusing ppl.

:wave:

if you wanna know what your bandwidth is ~ bootup to Memtest86

[oldfart]

Who's looking at Sandra???

How about the 3DMark scores?? :scratch:

Do you think they are valid??

[ThugsRook]

yea sure ~ 3dmark scores look fine.
i have no problem believing that at all.

3dmark2001 does play favorites, but for tweaking your own system its extreamly valid.

why?

[ThugsRook]

Radelon ~ can you redo these tests using Q3A in 1024res or higher? would be nice to have a 2nd gaming opinion.

:wave:

[oldfart]

Overclockers.com and Sandra.... (http://www.overclockers.com/articles756/)

[Tony]

I sent Ed an email asking him to use the unbuffered bench, it more acurate than the buffered one but not as good as it could be.

Lets see what he come's back with.

[ThugsRook]

bigtoe ~ i dont think something is wrong is youre unbuffered test, its been a great tool in the past.
but right now i think something is wrong with sandra itself ~ its gone springdale happy!

got a i865/875 board? ....heres some extra points for you!

;)

[oldfart]

I just ran a few quick Q3 benches.

First, my "normal" setup to compare fast and disabled performance mode.

FSB | Ratio | cas settings | performance mode | Q3 FPS
257 FSB | 5:4 | 2-2-2-5 | disabled | 390.6
257 FSB | 5:4 | 2-2-2-5 | fast | 391.3

Next, 213 FSB since this is my max 1:1 FSB setting due to the limit of my ram. To get there I have to use slow timings:

FSB | Ratio | cas settings | performance mode | Q3 FPS
213 FSB | 1:1 | 2.5-3-3-7 | disabled | 335.8
213 FSB | 5:4 | 2-2-2-5 | disabled | 336.2
213 FSB | 5:4 | 2-2-2-5 | fast | 338.3

For me, "fast" Vs "disabled" is a very tiny increase. Not really enough to bother with.

What I want to point out again is you can run 5:4 with tight cas timings and get the same performance as 1:1 with slow timings.

This may allow a higher FSB if your memory cant handle the high FSB speeds (very common) and save you a bunch of $$ by not having to buy the high end ram.

You dont have to run 1:1 for max performance. Once you can run 1:1 AND 2-2-2-5 or so, then you are getting somewhere. 1:1 with slow timings doesn't get you anything.

I'd like to see this same type of test on a board other than an IC7 to see the results.

As far as the Sandra problem goes, I solved that long ago. Its quite easy really.

Start/settings/control panel/add remove programs....

[ThugsRook]

Originally posted by oldfart
I just ran a few quick Q3 benches.

You dont have to run 1:1 for max performance. Once you can run 1:1 AND 2-2-2-5 or so, then you are getting somewhere. 1:1 with slow timings doesn't get you anything.
same old story, different ratios ;)

its always been hard to beat aggressive timings, especially for games.
eventually mem quality will catch up tho, it always does.

:wave:

[ThugsRook]

oldfart

let me put it this way.....

if you had an unlocked cpu right now ~ what multiplier/fsb/mem ratio do think would be "the best"?

[Tony]

You guys are not using ram that is optimised for 1:1 usage on these boards.I can run my TwinmosBH5 at 2-5-3-2 ddr433 and my OCZ gold at 2-5-3-3 1:1 and the 3dmark scores are very diferent.Even Radelon has found the OCZ gives much better bandwidth at the same clockspeed even when comparing running async or sync..he just has modules that are like mine theat have a faulty spd that doesn't allow async operation.

We really do need to run bandwidth hungry tests to see whats better or not.

[oldfart]

Unlocked?? That would be cool. Let's assume an "average" 3.4 GHz over**** and very good ram that could do ~ DDR450+ 2-2-2-6??

15 x 227 1:1 2-2-2-6 would be great! You can do that now with a 3.0. But that CPU and that ram will cost ya plenty.

I wonder if it would actually be faster than a setup like mine with the slower FSB? I dont think it would all that much different. Just much more $$.

Since most cheapo overclocker types like me aren't using 3.0s or unlocked CPU's, we have to make do with the 2.4 - 2.6s and "normal" ram.

I'm talking about what to use for the actual setups people are putting together now, not "what ifs". This whole "must run 1:1 is hogwash from what I see. For these setups a 5:4 ratio is the way to go.
:rock: :rotflmao:

[oldfart]

Originally posted by bigtoe
You guys are not using ram that is optimised for 1:1 usage on these boards.I can run my TwinmosBH5 at 2-5-3-2 ddr433 and my OCZ gold at 2-5-3-3 1:1 and the 3dmark scores are very diferent.Even Radelon has found the OCZ gives much better bandwidth at the same clockspeed even when comparing running async or sync..he just has modules that are like mine theat have a faulty spd that doesn't allow async operation.

We really do need to run bandwidth hungry tests to see whats better or not.
Anyone want to send me some to "test"?? ;)

[Tony]

Just done a back to back with the P4P800D at 250fsb.
1:1 with OCZ cas2-6-3-3 and
5:4 with TwinMOS BH5 cas2-5-3-2

3dmark 1:1=17468
3dmark 5:4=17106

I think I'll be running 1:1 especially if i get a 3C to play with.

[Radelon]

Hmmm, haven't run Q3A in a while, heck, dont' even know where it is......

Fact is..... 1:1 should be faster than 5:4 at the same CPU FSB and only a change from 6 to 7 in tRAS....... which it is..... but with Performance set to FAST on the 5:4 config, the 1:1 config gets left in the dust. If I can find Q3A I sure will fire it up and test it as long as you tell me the exact things you want tested and with what patch, etc.

I think you are right about Sandra and it being 865/875 friendly, but damn, when 3dmark is saying that running basically 410ddr (5:4 config) with Fast Performance enabled 2-6-3-3 is faster than 510ddr (1:1 config) with Disabled Peformance 2-7-3-3, you just start poppin' questions. Especially since with Fast Performance enabled over a non-enabled 5:4 config gives you a 800 point increase, and 1:1 non-enabled config is only a 500 point increase over the 5:4 non-enabled config....... definitely makes you think twice about going 1:1 and running some wicked fast ram. This is so confusing cause is running 1:1 with blazing fast ram is slower than 5:4 with ok ram, really makes you wonder whats goin' on. Hopefully Abit finally will get on their horse and get this board running like it should.

[oldfart]

Is it that the IC7 does better with 5:4 performance than other boards??