Sign in with
Sign up | Sign in
Your question

P4 Memory Crippling

Last response: in CPUs
Share
February 2, 2004 3:17:32 AM

I've heard that populating all four sockets on the 865pe/875p boards with 800mhz procs cripples memory and CPU performance. I hate imply that Tom’s benchmarking methodology is flawed, but if you look at the test setups from the current Prescott article you will notice that Tom is using 4 dimms to reach his 1Gb memory mark on the P4 board and only 2 dimms on the Athlon boards.

Look Here
http://www20.tomshardware.com/cpu/20040201/prescott-09....

If you look at the Open GL test that follows for Quake 3 you will see that the P4 2.6c and the XP 2600+ are pretty close, 190.9 and 184.9 respectively. About 3% difference.
You will also notice that the XP 3200+ beats the P4 3.2c by a fair amount, 259.1 to 223.3, or about 14%.

Look Here:
http://www20.tomshardware.com/cpu/20040201/prescott-10....

However, if you go back to older tests like the one in the Bidding Adieu to P4 3.2 vs. Athlon XP 3200+ you will see some startlingly different results. In this case Tom was testing with only 2 dimms installed on the Athlon and P4 boards.

Look Here:
http://www20.tomshardware.com/cpu/20030623/p4_3200-05.h...

Take a look at the resulting Open GL benchmarks for Quake 3 that follow. Here you see that a P4 2.6c beats the XP 2700+ by a much wider margin. 390.1 to 291 or 25%.
You will also see a complete reversal here with the P4 3.2c beating the XP 3200+, 450 to 347.3 or by about 23%.

Look Here:
http://www20.tomshardware.com/cpu/20030623/p4_3200-07.h...

Now I will be the first to admit that I am comparing the results from two totally different benchmarking tests, Demo001 and Timedemo THG3.
Also the earlier test uses 512mb memory and the second uses 1gb.
I know that these are graphically intensive benchmarks and that the graphics card is often a bigger factor, but even so, a 20% difference between the two tests. Is a little hard to understand considering that they run on the same program.

This leads me to the to conclude that the tests differ for one or more of the following reasons.

1.) Populating all four dimm sockets on a 865/875 based P4 board is detrimental to performance.
2.) The performance of the P4 and XP processors is somehow affected by the amount of memory in the system. Though I believe this to be unlikely, if true it means that XP processors can gain a 20% performance advantage by switching from 512mb to 1gb of memory.
3.) The CPU utilization is somehow significantly different in Timedemo THG3 from Demo001. If this is the case I would like to know what differences account for the wide discrepancy in the benchmarks.

Anyway, this is just something I noticed and I thought I’d put it out there. I’d like to rely on Toms and similar websites to give me an accurate picture of a CPU performance but when a site's results vary so widely from article to article without explanation it’s hard for me to trust what I read. I think someone should look into this.

More about : memory crippling

February 2, 2004 3:26:28 AM

woah, you work for intel by any chance? lol
February 2, 2004 3:47:24 AM

Good post. I have heard the same (about lowering performance if all 4 filled), which is why I am sticking to two 512mb sticks in my system.

----------------------
PIV 2.4c @ 2.89ghz
1gig PC3200 (512mbx2)
ASUS P4P800
GF3 Ti200 64mb (soon to be replaced)
WinXP Pro
3DMark2001SE: 6309
3DMark2003: 729
Related resources
February 2, 2004 11:14:28 AM

Nope....I'm just a dude looking for the strait story. I can't help it if I notice inconsistencies.
February 2, 2004 11:29:38 AM

Very interesting. I've noticed that most other sites test with 2x512, not 4x256... But I don't see how anyone can do anything about that right now. In any case, because of other reviews, I don't really think that this would make that much of a difference...

I could be wrong, though.

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
Anonymous
a b à CPUs
a b } Memory
February 2, 2004 11:35:39 AM

>'ve heard that populating all four sockets on the
>865pe/875p boards with 800mhz procs cripples memory and CPU
>performance

Do some googling. i think there was a (french ?) website that claimed this a while ago, with some spectacular benchmarks to prove it, but then I think Anand did an in depth investigation (can't find the link), and concluded the opposite; overall the 875 performs *best* with 4 Dimms, just like what intel tells us. Maybe someone else can give you the link ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
a b } Memory
February 2, 2004 11:39:20 AM

Have a look here (not what i was looking for btw):

"Despite the slightly lower overclock possible with four DIMMs, the best 1:1 performance is still with four DIMMs. For example, DDR516 using four DIMMs out-performs DDR530 with two DIMMs – since the four DIMMs are about 7% to 10% higher in UNBuffered memory performance."

<A HREF="http://www.anandtech.com/memory/showdoc.html?i=1839&p=7" target="_new">http://www.anandtech.com/memory/showdoc.html?i=1839&p=7...;/A>

God knows I don't always agree with THG and its methodology (or lack off), but they the right decission here it seems. (though testing 2 DIMM is also quite understandable, as this will be the most common configuration)

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 2, 2004 11:43:24 AM

<A HREF="http://www.anandtech.com/memory/showdoc.html?i=1839" target="_new">There you go.</A>

However, this poses the question... Why don't the usual review sites use 4 populated sockets instead of 2? Anandtech used only 2, and they "know" 4 gives a few percentage points of extra performance?...

I know I'm bitching around, but isn't that a reasonable question to be asked?...

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
February 2, 2004 11:45:25 AM

You beat me to it by 4 minutes, bbaeyens! :lol: 

I must have been reading the article myself during those 4 minutes. And I kept the message form open and posted later...

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
Anonymous
a b à CPUs
a b } Memory
February 2, 2004 11:58:06 AM

You got the more relevant article though (part 1), i could only quote the results since I linked part 2.

However, I seem to recall that Anand (re-?)did a specific article in a reaction to the "4 dimms sucks ass" claims made by another website.. but I could be wrong on that. either way, your link should give him all the answers he could hope for.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 2, 2004 3:55:05 PM

Thank you bbaeyens and Mephistopheles for the links. I think I may have seen this article before.

However, if we assume for the moment that Tom is in fact using the best memory configuration as the Anandtech article suggests. It still doesn't account for the wide difference in the two sets of benchmarks.

My question remains, Why such a big difference in the two tests?

If it's not the memory then is it a difference in the Demo Tests or is it the result of a the different amounts of memory in the system?
February 2, 2004 4:06:46 PM

The setup is not 100% clear in the first test...

<b>wooooow <font color=red> Killer Klowns </font color=red> ... from <font color=blue>outer space</font color=blue>... HOLY SH¡T!</b>
February 4, 2004 1:53:00 AM

Does anybody else have any comments on this? I'm surprised more people aren't interested.
February 4, 2004 3:19:44 AM

" Tom's" benchmarks seem to be hand picked by intel. Thier results show the Northwood 800 fsb chips in a much better light than any other sight. Most feel that if numbers have been scued, it is in Intel's favor.
!