Athlon XP Memory Matrix

[JMONKEY1]

Zebo posted a great article about how "Overclocker" memory does not yield much performance for the price on Athlon 64(939) systems. Does anyone have a similar memory performance matrix for Athlon XPs?

I have 512MB Crucial PC133(2100) 2.5 CAS memory and wanted to upgrade to 1GB. Does anyone have benchmarks comparing higher FSB speeds at same processor speed on an AthlonXP system? Should I get PC3200 value ram or should I pay twice as much and get PC4400 memory?

I have a mobile 2500XP and plan on running the same processor speed regardless of the memory speed. I will start benchmarking my system at 100mhz and 133mhz FSB and post the results but since my memory doesn't go much higher I am not sure what will happen at 200, 250 or 300mhz. If you have numbers please post. I want to know what kind of performance increase you saw when clocking your FSB over 200mhz. Thanks.

 

[sangyup81]

In his post, bandwith DOES matter for the lower end.

I can't give you a matrix but unlike Athlon 64 systems, your memory needs to be timed like your CPU's FSB (synchronous). PC4400 memory is no good unless you jack up the FSB and you'll need a reliable motherboard for that. XP Motherboards are not reliable at high FSBs the way Athlon 64s are. Get the value ram but if you really want to spend money, get some low latency memory.

 

[JMONKEY1]

I have a DFI NFII Ultra Infinity which I have heard can get up to 300mhz FSB. If someone else has this motherboard and has experience with memory overclocking on this board, please send a heads up.

System:
FSP 350w Aurora
DFI NFII Ultra Infinity
Mobile AthlonXP 2500
Sapphire Radeon 9500 pro oc'd 340.62/311.54
512MB Crucial PC2100

I know my system is old school but here is what I found out so far. I tested my system at 1600mhz with a FSB of 100mhz and 133mhz. I don't know how this would translate to 200mhz versus 266mhz, so again, if someone has some benchmarks on that please let me know.

MEMTEST 86
712MB/s - 2.5,3,3,8 @ 100MHZ
963Mb/s - 2.5,3,3,8 @ 133MHZ
728MB/s - 2,2,2,8 @ 100MHZ
978Mb/s - 2,2,2,8 @ 133MHZ
An increase in bandwidth of 33% yielded an increase of 35% in throughput.
Lowering the latencies added about 2% increase in memory throughput.

3dmark01
10827 2.5,3,3,8 @ 100MHZ
12033 2.5,3,3,8 @ 133MHZ
Extra bandwith yielded a 11% increase in the score.

3dmark03
4144 2.5,3,3,8 @ 100MHZ
4401 2.5,3,3,8 @ 133MHZ
Extra bandwith yielded a 6% increase in the score.

HALO 800*600
1600mhz 12*133mhz @ 2,2,2,8 = 49.02fps
1600mhz 12*133mhz @ 2.5,3,3,8 = 48.87fps
1600mhz 16*100mhz @ 2.5,3,3,8 = 45.51fps

HALO 1024*768
1600mhz 12*133mhz @ 2,2,2,8 = 42.25fps
1600mhz 12*133mhz @ 2.5,3,3,8 = 41.68fps
1600mhz 16*100mhz @ 2.5,3,3,8 = 40.05fps

HALO 1024*768
1600mhz 12*133mhz @ 2,2,2,8 = 30.93fps
1600mhz 12*133mhz @ 2.5,3,3,8 = 30.79fps
1600mhz 16*100mhz @ 2.5,3,3,8 = 30.35fps

From above, it is obvious that this test is mostly limited by my graphics card so I could use some more info from other users with a high end card.
That aside, the 33% increase in bandwidth only gave me an increase of 7%. The lower latencies gave me no perceptible increase.


**CONCLUSION**
On my system which is bound by the graphics card, I get no benefit from high end memory. Neither the extra bandwidth nor lower latencies did much.
I plan on upgrading my graphics card eventually though so the help with the memory choice would be appreciated.

I will run benchmarks against Call of Duty to see if I can get more accurate results. I will also try Super PI.

If anyone has already done these tests for 200mhz FSB and above please post!

UPDATED
SuperPi 1M
72s - 2.5,3,3,8 @ 16*100MHZ
69s - 2,2,2,8 @ 16*100MHZ
65s - 2.5,3,3,8 @ 12*133MHZ
63s - 2,2,2,8 @ 12*133MHZ
52s - 2,2,2,8 @ 16*133MHZ
49s - 2,2,2,11 @ 16*140mhz

In these tests it looks like latency is in step with increased bandwidth. The lower latency is equal to roughly 16~25 mhz in bandwidth. That last run just show's that an increase in cpu speed has the greatest effect on time.