Weird benchmark memory problem

Post by **Iceblade** » 19 Feb 2005, 20:39

My friend Madnax has the following system:

Mobo = GIGABYTE AMD K8NNXP DPS
CPU = AMD 3200+ (2.0GHZ) CLAWHAMMER
RAM = 2X512MB = 1GB TRANSCEND
GPU = GIGABYTE FX5700 ULTA NVIDIA
PSU = 400W PROCASE
HDD = 80GIG SATA SEAGATE 7200RPM
O/S = WIN XP SP1A

Now the problem is as follows:
He had 2x256 Transcend RAM in his sytem, the he added a 512 module and his benchmark score dropped by about 1500 points,
so I told him to get another 512 module and get rid of the 2x256 modules.
So he did, but the problem was the same.

BENCH MARKING AS FOLLOWS :

3D MARK 2001SE
512MB RAM = 15700 PTS
1GB RAM = 14430 PTS

3D MARK 2005
512MB RAM = 717 PTS
1GB RAM = 709 PTS

I don't get it,
why does a benchmark score drop when more RAM is added?
If he benches with both he gets a low score,
take out one and he gets a better score.

This is to weird for me!!!

Post by **Iceblade** » 26 Feb 2005, 17:05

Seeing no-one on this forum could answer this, I went to a internasional PC forum,
this is what I got:

Yep the exact same thing happened to me.

Got 1Gb RAM (2x512) worked well got good scores.

Added another (2x512) thinking more is better.

WRONG scores dropped by 1500-2000 points on all benches.

It may have been the Administrator's post i saw that said its to do with what you said about how much RAM the system has to control, apparently 512Mb is optimal for XP from what people have told me.

Boy i was pissed after forking out big bucks for RAM that slowed it down.

and

having more RAM than what is needed will slightly slow down the system, as you can see, it didn't slow it down much, but the MCH has more memory to keep track of so having 1gb will be slower than 512mb as long as you don't use over 512mb thats' why systems with 2gb of RAM will be slower than a system with 1gb, unless you're using over 1gb, i remember the administrator talking about that a LONG time ago when he had 2gb and actually got better benchmark scores with 1gb, i wouldn't really worry about it too much

So 1gig memory is not that good then

Post by **Iceblade** » 26 Feb 2005, 20:34

Another answer,
this guy has 2gig memory in his PC and this is what he said:

I get my best scores with only 256MB or ram. Why? not 100% sure but it can majorly depend on the brand and latency of the ram. normally the more ram you have the worse the ram latency which a benchmark program like 01 can way heavily on latency.

8O

Rayne · Post by **Rayne** » 27 Feb 2005, 14:45

Hmmm, so I should chuck 2 of my DIMM's then... 8O

You know I always wondered why reviewers, incl our beloved PCF team, always bench @ 1Gb of RAM. I really thought more is better, as I know that running 2x DIMM's dual-channel beats the single DIMM simply because instructions are now written 2 both DIMM's at once instead of waiting in line with the single so my presumption at time of buying 2 more DIMM's was that my pc would now be even faster as it would now be writing 2 4x DIMM's...

But now I learn it's slower. How so?

I would love 2 see an in-depth technical answer on this 1.

solo · Post by **solo** » 27 Feb 2005, 14:58

Download CPU-Z and check the Revision of his CPU. If it is a CG stepping it could mean that the BIOS is automatically setting his memory to use a 2T timing. This is a feature of the newer revision Athlon64's i.e. it results in better compatibilty when using 512 MB modules.

The drawback though is that it seriously reduces your memory bandwidth. Resulting in lower scores in benchmarks. If there is a option in the BIOS to set memory to a 1T timing, then do that and see what happens.

It MIGHT just be that.

goku · Post by **goku** » 27 Feb 2005, 15:35

This is why I stopped with this obsession of benching scores( no offends to the guys that do

, it's just that tweaking your rig for benching, gaming and other application don't run well and some games even plays worse.

Post by **Iceblade** » 27 Feb 2005, 17:20

solo wrote:Download CPU-Z and check the Revision of his CPU. If it is a CG stepping it could mean that the BIOS is automatically setting his memory to use a 2T timing. This is a feature of the newer revision Athlon64's i.e. it results in better compatibilty when using 512 MB modules.

The drawback though is that it seriously reduces your memory bandwidth. Resulting in lower scores in benchmarks. If there is a option in the BIOS to set memory to a 1T timing, then do that and see what happens.

It MIGHT just be that.

Thanks Solo,
you are the first one with any sort of advice

Will ask him to check it out

It got me worried as I have 2x512 DDR400 XMS corsair extra low latency modules waiting for my K8N neo Platinum board to arive

Bobendren · Post by **Bobendren** » 27 Feb 2005, 17:34

having more ram would improve system performance overall though wouldn't it? or otherwise what's the point?

Post by **Iceblade** » 27 Feb 2005, 17:37

Bobendren wrote:having more ram would improve system performance overall though wouldn't it? or otherwise what's the point?

That's what I was thinking,
but this guy said something different

having more RAM than what is needed will slightly slow down the system, as you can see, it didn't slow it down much, but the MCH has more memory to keep track of so having 1gb will be slower than 512mb as long as you don't use over 512mb thats' why systems with 2gb of RAM will be slower than a system with 1gb, unless you're using over 1gb, i remember the administrator talking about that a LONG time ago when he had 2gb and actually got better benchmark scores with 1gb, i wouldn't really worry about it too much

pSychoSis · Post by **pSychoSis** » 27 Feb 2005, 17:43

I also noticed this when i had 512Mb installed, and when i ran it with a small RAM increase to 768MB, my 3dMark01 Score dropped by 100-200 Marks.

Now I know why....

Bobendren · Post by **Bobendren** » 27 Feb 2005, 17:45

so what kind of programs use up more than 512?

photoshop or anykind of graphics editing or video editing software should, right?

in games like doom3 1gb must be more beneficial than 512, surely?

i was thinking of buying 2x512 ram modules but after reading this 2x 256 sounds like a better deal, considering the price.

solo · Post by **solo** » 27 Feb 2005, 18:06

I've benchmarked with 2x256 MB modules and 2x512MB modules and seen no difference. When overclocking though, 2x256 MB will usually overclock higher.

Modern PC's should have at least a gig of RAM in my opinion. Getting 512 MB instead, because you're worried about a huge drop in performance is just silly. Modern games like Doom3 and Far Cry do benefit from the increased RAM for sure.

Post by **Iceblade** » 27 Feb 2005, 18:15

solo wrote:I've benchmarked with 2x256 MB modules and 2x512MB modules and seen no difference. When overclocking though, 2x256 MB will usually overclock higher.

8O 8O 8O

no difference
8O 8O 8O
2x256 overclock higher

solo · Post by **solo** » 27 Feb 2005, 18:22

You'll usually get the highest overclocks with 1 x 256 MB stick though (if it's a single channel board) Two sticks no good for overclocking in that situation.

Bobendren · Post by **Bobendren** » 27 Feb 2005, 18:22

hey check out these 2 links:

http://forums.anandtech.com/messageview ... TMP=Linear

http://www.gamepc.com/labs/view_content ... e%5Ftest=1

i'm more keen on getting 2x512 again

Post by **Iceblade** » 27 Feb 2005, 18:25

solo wrote:You'll usually get the highest overclocks with 1 x 256 MB stick though (if it's a single channel board) Two sticks no good for overclocking in that situation.

Solo just of the topic a little

What specs are you running now and what benchmark 3D mark2001SE record is your best to date

solo · Post by **solo** » 28 Feb 2005, 06:32

Iceblade wrote: Solo just of the topic a little
What specs are you running now and what benchmark 3D mark2001SE record is your best to date

My current highest official score is 35448.
http://service.futuremark.com/compare?2k1=8277231
It was done a while back and could have been 36K with the little tweaks I have learned since then.

That setup was:

AMD FX55 @ 287x11 = 3.16 GHz :: MSI K8N Neo2 Platinum :: Gigabyte X800XT Platinum @ 555/590 :: PQI Turbo (Samsung TCCD) @ 287 MHz [3-3-3] with 2.75 volts 1:1 :: Prometeia Mach 2 cooling.

Myself, flytek / allbrain / bullfrog / gouhan / juju did do a little LN2 experiment in December and we got 37406.
http://service.futuremark.com/compare?2k1=8319986

I'm currently using my AMD Athlon64 3700+ :: DFI LanpartyUT NF3 :: 1 Gig OCZ3700EB RAM and 9800 Pro as a daily rig.

I sold my X800XT Platinum and got a X850XT Platinum. I'm just still waiting for my DFI NF4 Ultra to arrive now then that with my FX55 will be my main rig. I ordered the DFI like 12 Feb and it's only due 7 March (damn slow suppliers) so next Monday should be fun and games

Post by **Iceblade** » 04 Mar 2005, 22:33

solo wrote:
Iceblade wrote: Solo just of the topic a little
What specs are you running now and what benchmark 3D mark2001SE record is your best to date
My current highest official score is 35448.
http://service.futuremark.com/compare?2k1=8277231
It was done a while back and could have been 36K with the little tweaks I have learned since then.

That setup was:

AMD FX55 @ 287x11 = 3.16 GHz :: MSI K8N Neo2 Platinum :: Gigabyte X800XT Platinum @ 555/590 :: PQI Turbo (Samsung TCCD) @ 287 MHz [3-3-3] with 2.75 volts 1:1 :: Prometeia Mach 2 cooling.

Myself, flytek / allbrain / bullfrog / gouhan / juju did do a little LN2 experiment in December and we got 37406.
http://service.futuremark.com/compare?2k1=8319986

I'm currently using my AMD Athlon64 3700+ :: DFI LanpartyUT NF3 :: 1 Gig OCZ3700EB RAM and 9800 Pro as a daily rig.

I sold my X800XT Platinum and got a X850XT Platinum. I'm just still waiting for my DFI NF4 Ultra to arrive now then that with my FX55 will be my main rig. I ordered the DFI like 12 Feb and it's only due 7 March (damn slow suppliers) so next Monday should be fun and games

8O
You rule man

Slasher · Post by **Slasher** » 04 Mar 2005, 22:37

Nah, Mr Bean rules

j/k... You both are amazing at what you do, whish I had your talent to do it, or at least just the money... o for the money...