IMO, good 1:1 high timing vs 5:4 low timing thread

[orion7144]

Originally posted by: oldfart
1:1 with slow timings and 5:4 with fast timings are really close. You will see some applications prefer one or the other.

I agree there but I still say that even if you took your above statement and use 1:1 @ 200FSB and 5:4 @ 250FSB same timings with both tests the difference in memory performance is not that significant for the current speed of computers (1-2%). However the overall system speed is significantly increased and I would and most people would rather run @ 3gig instead of 2.4gig and give a little in memory performance to achieve a better overal system performance.

 

[oldfart]

Originally posted by: orion7144

Originally posted by: oldfart
1:1 with slow timings and 5:4 with fast timings are really close. You will see some applications prefer one or the other.

I agree there but I still say that even if you took your above statement and use 1:1 @ 200FSB and 5:4 @ 250FSB same timings with both tests the difference in memory performance is not that significant for the current speed of computers (1-2%). However the overall system speed is significantly increased and I would and most people would rather run @ 3gig instead of 2.4gig and give a little in memory performance to achieve a better overal system performance.

Oh that goes without even saying! 600 MHz in CPU speed is a huge difference. I'm saying that even with the SAME CPU speed such as:

2.6C example:

13 x 250 FSB = 3.25 GHz
1:1 ratio = DDR500
slow timings (3-4-4-8)

13 x 250 = 3.25 GHz
5:4 ratio = DDR400
fast timings (2-2-2-6)

These two systems would perform similarly. The 2nd system will actually be a little faster due to the lower latency.

 

[pallmall]

Originally posted by: oldfart
Dude, I dont understand you at all. You are making WAY TOO MUCH OF THIS. What are you some scientific method test freak or something??? .

The tests have been conclusive. 1:1 ratio, high DDR speed, high latency performs ~ the same as 5:4 ratio, lower DDR speed, ow latency with a slight edge going to the latter. That is what all the tests all over the net have shown. Each has shown very similar results. I dont see what is so difficult to understand here.

Please excuse me for not testing each of the 10's of Thousands of CPU/mobo/ram/FSB/timing options for you. You are welcome to do so yourself. No one gets paid anything here. We just share our knowledge of what we have tested and seen around. You needn't micro-analyze this any longer.

Very good use of language.
I agree, everything about the computer performance is similar, small thing, so & so...
So let's just stay where we are, Big dude.

 

[THUGSROOK]

What are you some scientific method test freak or something???

hey! thats my job

oldfart is 100% correct ~ youd need a P4 D1 step ES and 2 different pairs of ram to prove this thoroughly.

w/o that we are left with comparing each other results on various moboards with various ram using various cas settings. not entirely scientific, but we are seeing a pattern.

 

[pallmall]

I am glad I am not told 100% incorrect.
Generally, I agree with the consensus in this thread, whcih is what I have experienced too.
But do you remember what I asked first?. Is it answered or even considered?
Tell me just if it is useless or not, which will be most helpful to me.

 

[orion7144]

Originally posted by: THUGSROOK

What are you some scientific method test freak or something???

hey! thats my job

oldfart is 100% correct ~ youd need a P4 D1 step ES and 2 different pairs of ram to prove this thoroughly.

w/o that we are left with comparing each other results on various moboards with various ram using various cas settings. not entirely scientific, but we are seeing a pattern.

Yes, but I still say that we are not comparing apples to apples. If we were to throw in an Nforce Dual DDR board in to the equation I would say that the CAS means a big difference in performance on that platform however, on the 865/875 I do not believe it to be true as in my situation with testing on the IC7-G and the IS7 produced similar results. Just tell me what to run and I will do it at the end of the week on an IC7-G. I have, at my disposal the OCZ PC3700 GOLD and 2 sets of the OCZ PC3500EL both 2X256. I will do all testes at 250fsb 1:1 and 5:4 as well as 200fsb 1:1.

 

[THUGSROOK]

here is what cas is worth on my system... link

there is also a comparison to 466 and 500ddr in that thread.
ill let you judge for yourself.

ill bet oldfart has his results posted somewhere on that forum too.

 

[orion7144]

Originally posted by: THUGSROOK
here is what cas is worth on my system... link

there is also a comparison to 466 and 500ddr in that thread.
ill let you judge for yourself.

ill bet oldfart has his results posted somewhere on that forum too.

See we are not talking the same language. I am talking system performance. Take the last page and add the sandra scores and the non memory results would show my point that the higher FSB is better overall. I am not saying that memory benchmarks don't suffer a little but that If I had to chose between 250 1:1 @ 2.5-3-3-7 or 250 5:4 2-2-2-6 I would chose the 250 1:1 since the system overall would be faster.

BTW I totally forgot about the GAT options and that was a quality indepth review.

 

[oldfart]

A few cut - n paste of 1:1 vs 5:4 @ same CPU speed.

By using 3DMark03 to judge what is my maximum FSB/memory/video setting for my P4 2.8c cpu on my Asus P4C800, I obtained the following results:

1) no name PC3200 (1G); FSB=235; 5:4 ratio; 2-3-3-6 SPD; result = 4375, CPU=730

2) Corsair XMS3700 (1G); FSB=235; 1:1 ratio; 3-4-4-8 SPD; result = 4316, CPU=759

3) Corsair XMS3700 (1G); FSB=250: 1:1 ratio; 3-4-4-8 SPD; result = 4328, CPU=810

My questions are:

1) why does my no name PC3200 ram at FSB=235 have a higher score than my Corsair XMS3700 ram at FSB=250?

2) why are my scores so low as I see people in the forum scoring 5000 to 6000 with similar config?

3Dmark03 was slower @ 1:1 due to high latency

XMS3200LL vs XMS3700 (5:4/1:1) benchmarks & comparison. Somewhat surprising results..

Ok, there has been a lot of talk about 5:4 vs 1:1 and XMS3200LL vs XMS3700. I decided to benchmark it. See the attached gif for the results. The most interesting are E & H.

Some general conclusions, will probably come back and edit this later as its 4am and I'm not at my most saliant atm.

Conclusion 1 - in the less synthetic tests, the advantage of the 3700 was utterly marginal at its rated timings. No improvement in Q3Bench at all. Just goes to show that synthetic memory performance makes not a lot of difference in the real world.

When I tightened them up a bit it did improve a tad, and right now I'm Prime95 testing 3.0-3-4-7. Will update if its stable. Just goes to show how important timings are.

Conclusion 2 - 5:4 or 1:1 you can get bloody similar performance. I don't think people should worry about 5:4 at all, its not crippleing your system despite what some places have said.

He didn't post bench scores, but they can be seen here
Q3 1:1 = 258.28 FPS,
5:4 = 258.83.
3DMark01 1:1 = 11051,
5:4 = 10992.
Basically, same performace. The gain of the 1:1 DDR speed is lost by the high latency.


This test:

XMS4000 DDR500 1:1 3-4-4-8 340.8 FPS
XMS3200 DDR400 5:4 2-3-2-6 338.9 FPS
Numbers are very close. 2-2-2-5 would have been faster than 1:1 if run that way.


This test:

XMS4000 DDR500 1:1 3-8-4-4 320 FPS
XMS4000 DDR500 1:1 2.5-7-4-4 338 FPS
XMS3200 DDR400 5:4 2-5-2-2 340.5 FPS
In this test, the PC3200 low latency is a bit faster than the PC4000 with medium timings, quite a bit faster than the slowest timings.

Octools article
3DMark2K1
250 FSB | 1:1 | DDR500 | 3-4-4-8
17979

250 FSB | 5:4 | DDR400 | 2-3-3-5
17964

The difference? Less than 1/10 of 1%.

This one:
250/250 (2.5-4-3-7)
SuperPi 1M / 16M seconds 45/1102
Prime95 Bench seconds 66.435
3Dmark 2001 / 2003 12197/3467

250/200 (2-3-3-6)
SuperPi 1M / 16M seconds 46/1096
Prime95 Bench seconds 66.836
3Dmark 2001 / 2003 12304/3461

Not a lot of difference there. SiperPi and Prime95 are tied. 3DMark2K1 is better on the 5:4 setup, 3DMark 2K3 is tied.

So, latency DOES matter. Low latency timings equal out the difference between 1:1 and 5:4 which is 20% raw BW speed difference.

 

[orion7144]

Originally posted by: oldfart
A few cut - n paste of 1:1 vs 5:4 @ same CPU speed.

By using 3DMark03 to judge what is my maximum FSB/memory/video setting for my P4 2.8c cpu on my Asus P4C800, I obtained the following results:

1) no name PC3200 (1G); FSB=235; 5:4 ratio; 2-3-3-6 SPD; result = 4375, CPU=730

2) Corsair XMS3700 (1G); FSB=235; 1:1 ratio; 3-4-4-8 SPD; result = 4316, CPU=759

3) Corsair XMS3700 (1G); FSB=250: 1:1 ratio; 3-4-4-8 SPD; result = 4328, CPU=810

My questions are:

1) why does my no name PC3200 ram at FSB=235 have a higher score than my Corsair XMS3700 ram at FSB=250?

2) why are my scores so low as I see people in the forum scoring 5000 to 6000 with similar config?

3Dmark03 was slower @ 1:1 due to high latency

XMS3200LL vs XMS3700 (5:4/1:1) benchmarks & comparison. Somewhat surprising results..

Ok, there has been a lot of talk about 5:4 vs 1:1 and XMS3200LL vs XMS3700. I decided to benchmark it. See the attached gif for the results. The most interesting are E & H.

Some general conclusions, will probably come back and edit this later as its 4am and I'm not at my most saliant atm.

Conclusion 1 - in the less synthetic tests, the advantage of the 3700 was utterly marginal at its rated timings. No improvement in Q3Bench at all. Just goes to show that synthetic memory performance makes not a lot of difference in the real world.

When I tightened them up a bit it did improve a tad, and right now I'm Prime95 testing 3.0-3-4-7. Will update if its stable. Just goes to show how important timings are.

Conclusion 2 - 5:4 or 1:1 you can get bloody similar performance. I don't think people should worry about 5:4 at all, its not crippleing your system despite what some places have said.

He didn't post bench scores, but they can be seen here
Q3 1:1 = 258.28 FPS,
5:4 = 258.83.
3DMark01 1:1 = 11051,
5:4 = 10992.
Basically, same performace. The gain of the 1:1 DDR speed is lost by the high latency.


This test:

XMS4000 DDR500 1:1 3-4-4-8 340.8 FPS
XMS3200 DDR400 5:4 2-3-2-6 338.9 FPS
Numbers are very close. 2-2-2-5 would have been faster than 1:1 if run that way.


This test:

XMS4000 DDR500 1:1 3-8-4-4 320 FPS
XMS4000 DDR500 1:1 2.5-7-4-4 338 FPS
XMS3200 DDR400 5:4 2-5-2-2 340.5 FPS
In this test, the PC3200 low latency is a bit faster than the PC4000 with medium timings, quite a bit faster than the slowest timings.

Octools article
3DMark2K1
250 FSB | 1:1 | DDR500 | 3-4-4-8
17979

250 FSB | 5:4 | DDR400 | 2-3-3-5
17964

The difference? Less than 1/10 of 1%.

This one:
250/250 (2.5-4-3-7)
SuperPi 1M / 16M seconds 45/1102
Prime95 Bench seconds 66.435
3Dmark 2001 / 2003 12197/3467

250/200 (2-3-3-6)
SuperPi 1M / 16M seconds 46/1096
Prime95 Bench seconds 66.836
3Dmark 2001 / 2003 12304/3461

Not a lot of difference there. SiperPi and Prime95 are tied. 3DMark2K1 is better on the 5:4 setup, 3DMark 2K3 is tied.

So, latency DOES matter. Low latency timings equal out the difference between 1:1 and 5:4 which is 20% raw BW speed difference.

So, you do agree that you should run at the maximum FSB the CPU will alow even though you may have to run the 5:4 multiplier?

 

[oldfart]

Yes!

CPU speed is WAY more important than any memory tweak.

 

[pallmall]

Originally posted by: orion7144<brSo, you do agree that you should run at the maximum FSB the CPU will alow even though you may have to run the 5:4 multiplier?

I think nobody can conclude it yet. Some benchmarks even ignore the major factor(cpu speed=fsb*multi).
Look at the other timings changes when they change the cas from 3 to 2.5 and 2.
The tradeoffs of these changes overall produces similar marks.
Thanks for anybody providing these real test data. But when analizing them, we need special care.
I am sleepy.

[edit]
Here is a little basis on why it is so hard to compare or analyze the performances.