INTEL - Synchronous is not important

[stevejst]

i think im gonna invest in some 4000ddr later on. i didnt get as lucky as you did, but i wish i had.

The funny thing my memory and my CPU are both refurbished, motherboard is new though. I guess the GEIL memory has been returned by some AMD user.
2x$43 = $86 for prime memory, $129.xx for the processor, I know it is hard to believe.

There is a new memory out there, it is called A-Data, and some guys have been able to get it running at 550 MHz reaching 7,000 on Sandra using 1:1.

 

[Slappy00]

I have my 2.4c runnin at 3.0 at 1:1 with a gig of bandwidth, man it smokes I tell you... sync is the way to go but to each his own, some like the CPU mhz up there I like massive bandwith to compliment the added CPU speed. 4200 ram helps also

 

[stevejst]

4200 ram helps also

Of course, that is the trick, no doubt.

 

[calfpheen]

i need money, and a new vid card, this gf4 mx 440 is killing me.

 

[stevejst]

i need money, and a new vid card, this gf4 mx 440 is killing me.

Abit Siluro 8x GF4 Ti 4200 128 MB was available the past 5 days on newegg refurbished for $79. Now it is gone. I do not need new graphic card but I was amazed that it lasted that long. Ti 4200 8x 128 would be a serious improvement for you.

 

[Jeff7181]

Shortly AMD CPU cannot gain anything from a fast memory.

THIS is how you run benchmarks... I've adjusted the multiplier (It's an AMD thing, you wouldn't understand) to keep the clock speed as close to 2.0 Ghz as possible and haven't changed the timing of the RAM at all so that we're only testing the difference that memory bandwidth makes.
(The following scores will be given with 166x12 first, and 200x10 below.
*EDIT* 200x11 scores are highlighted.
**EDIT** 200/166 @ 2.0Ghz scores are given next to the 200x10 sync results
***EDIT*** 221/166 @ 2.21 Ghz in Italics)

Sandra CPU Arithmetic
7435/2997
7452/2999 7429/2999
8165/3301
8204/3313

Sandra CPU Multimedia
10724/11716
10725/11752 10723/11746
11800/12875
11858/13001

Sandra Memory Bandwidth
2566/2429
3018/2793 2575/2391
3049/2840
2662/2484

PCMark2002
6145/5395
6229/5956 6198/5413
6814/6128
6825/5796

3DMark2001
13324
14032 13284
14512
13705

How does this translate to more real world performance?

Unreal Tournament 2003 Demo HardOCP CPU Benchmark @ 640x480
dm-antalus (low/average/high)
27/131/567
28/140/612 19/132/567
30/152/669
19/140/613

Unreal Tournament 2003 Demo HardOCP High Quality Benchmark @ 1024x768
dm-antalus (low/average/high)
64/133/480
82/139/482 72/137/546
85/140/582
65/137/454

dm-asbestos (low/average/high)
91/143/500
93/199/546 89/192/491
93/208/570
94/199/517

ctf-citadel (low/average/high)
18/132/339
59/143/371 56/131/323
60/148/387
27/139/350


So AthlonXP's don't benefit from increased memory bandwidth huh?

For sh!ts n giggles, I'll run the same tests at 200x11 (2.2 Ghz) a little later.

*EDIT* Edited for 2.2 Ghz scores.

**EDIT** Edited for 200/166 async @ 2.0 Ghz

***EDIT*** Edited for P4 style overclocking... 221/166 async @ 2.21 Ghz

 

[Slappy00]

Quote:"There is a new memory out there, it is called A-Data, and some guys have been able to get it running at 550 MHz reaching 7,000 on Sandra using 1:1. "

yah its not available in the states for some reason though I was looking for it b4 buying some GEiL 4200... If I had the money I woulda bought Mushkin that is the shizzle for rizzle ...

 

[Tom]

"..and haven't changed the timing of the RAM at all so that we're only testing the difference that memory bandwidth makes."


Your test is testing the affect of higher FSB speed on the cpu and motherboard, not just memory bandwidth.


And earlier you said I was missing your point. I'm not missing your point, you're just wrong about the significance of using 1:1. If you were right about a P4 system being starved for memory bandwidth then why does running memory 20% slower only create a 3% drop in performance ?

If there was a penalty for using a ratio the drop in performance should be greater than the reduction in memory speed, so a drop in memory speed of 20% should create a greater than 20% drop in performance, but that isn't what happens.

edit- three factors affecting overall system speed in a P4 Canterwood/Springdale system..

1. cpu speed
2. FSB speed
3. memory speed

Of the three memory speed is the most expensive, most likely to cause stability problems, and gives the smallest performance increase.

No one is arguing that faster memory is slower than slower memory, or that less bandwidth is better than more bandwidth, this thread is about whether or not it's important.

If money is not a consideration then sure a person should get the best memory available, but if money is a consideration then expensive memory is not anywhere near as important as getting a cpu that can be overclocked to a high FSB and a good graphics card, assuming we are talking about a typical gamer, enthusiast, computer.

 

[Jeff7181]

Take a P4C and stick some PC2700 RAM in it and run it at SPD settings and see if you only get a 3% performance drop.

 

[Tom]

Originally posted by: Jeff7181
Take a P4C and stick some PC2700 RAM in it and run it at SPD settings and see if you only get a 3% performance drop.

If you read the post by ZROC that I linked to you would see that is pretty much what he did, and the result was actually a 3% drop for using a 5:4 ratio, which is pc2700. All other settings he used were the same.


edit-corrected 2% to 3%

 

[Tom]

Here is a quote from that article, minus the chart. You'll have to click on the link if you want to see the numbers.

"Memory dividers...


Wow. I found this one really interesting...and breaking it out into percentages really made it hit home. The first percentage row shows the individual differences on each benchmark when dropping from 1:1 to 5:4. The Composite figure is simply those seven percentages divided by seven.
That's right, folks...the drop from 1:1 to 5:4 is an average of 2%. Overclockers everywhere should be smiling about right now.
In the third score column, I dropped to 3:2, and the following percentage columns are the percentage drop of 3:2 from 1:1 and 5:4, respectively.
3:2 is a 4.91% drop from 1:1. And 3:2 is a 2.85% drop from 5:4."

link

 

[Jeff7181]

Ummm... 4:5 is not the same as 5:4

 

[Tom]

Who said 4:5 ? There isn't a 4:5 ratio at 200fsb. If I said 4:5 somewhere with regards to 200fsb I meant 5:4.

 

[Jeff7181]

Originally posted by: Dead Parrot Sketch
Who said 4:5 ? There isn't a 4:5 ratio at 200fsb. If I said 4:5 somewhere with regards to 200fsb I meant 5:4.

The chart on the thread you linked to says 4:5

BTW... new benchmarks added to my list, since you didn't like that I changed the FSB with the RAM frequency... next I'll keep the multiplier at 10, and overclock the FSB using RAM dividers to keep the RAM at 166... which is the same method you have to use for Intel processors since you can't adjust the multiplier.

 

[Tom]

There are a bunch of charts in that link. He tested a whole bunch of stuff.

Scoll down to the chart that has the words " memory divider.." just above it..

 

[Tom]

"BTW... new benchmarks added to my list, since you didn't like that I changed the FSB with the RAM frequency... next I'll keep the multiplier at 10, and overclock the FSB using RAM dividers to keep the RAM at 166... which is the same method you have to use for Intel processors since you can't adjust the multiplier. "

It's not that I don't like it, I was just stating the facts. It's a pretty well established principla, for the last 10 years or so, that faster FSB is one of the best ways to increase performance. My point was that you were wrong in saying you were only testing memory bandwidth.

 

[Jeff7181]

There's some "P4 style" overclocking via the FSB only and using RAM dividers.

 

[THUGSROOK]

ok guys ~ time i posted my benchies for you to compare to...

rig in sig/ 1:1 full GAT enabled 100% stable at everything

sandra unbuffered
3362 / 3197 mb/s

pcmark2002
cpu = 8420
mem = 10078
hd = 1424

3dmark2001
16453

aida32
5298 / 2314 mb/s

memtest86
2865 mb/s

can anyone even come close to my bandwidth scores at only 382ddr? i didnt think so

 

[stevejst]

Thugsrock, you have some real strange math going on. How do you get to 3.45 GHz with 2.4 GHz processor (multiplier 12) using 191 FSB???

My math says 2.292 GHz.

 

[stevejst]

Beside I am easily beating your Sandra unbuffered scores using 183x2 = 366 MHz memory.
I get 3430/3427.
Of course this is at 3:2. Meaning my processor is at "only" 3.3 GHz. And I don't even have GAT enabled.

 

[Technonut]

Thugsrock, you have some real strange math going on. How do you get to 3.45 GHz with 2.4 GHz processor (multiplier 12) using 191 FSB???

THUGSROOK is running the 2.4B. (Multiplier 18)....

 

[stevejst]

Got you, didn't notice he wrote B.

Another one, mine unbuffered 220x2=440, usinf 5:4, I am trashing your scores with 3887/3900.

 

[stevejst]

Another one, using 5:4 at 225.6x2 memory, unbuffered 4088/3983, you cannot touch that!

And another one, 5:4 at 228x2 memory, unbuffered 4102/4063.

 

[Technonut]

Originally posted by: stevejst
Got you, didn't notice he wrote B.

Another one, mine unbuffered 220x2=440, usinf 5:4, I am trashing your scores with 3887/3900.

What timings? With Street Racer (2-3-2-5) settings @ 1:1 450MHz, My Sandra unbuffered scores come in @ 3443/3431

 

[stevejst]

What timings? With Street Racer (2-3-2-5) settings @ 1:1 450MHz, My Sandra unbuffered scores come in @ 3443/3431

I am sure you have better memory than me. Mine 228 runs at 2.5-4-4-8, just auto for GAT.