Memory Benchmark puzzling me

[GeoffW]

Hi

I have ust built a new PC built based on a P35 chipset motherboard. It all went pretty smoothly but I am puzzled about how it benchmarks against my older PC. I was expecting it to be markedly faster on memory bandwidth than it is. I am probably misunderstanding something, maybe someone here can explain what is going on ? The Specs below.

OLD PC
Prescott P4 3.5 GHz
FSB 200 MHz
Ratio FSBRAM 3:4 Giving Ram Freq= 266.7 Mhz (from CPUZ)
Setting 4-4-4-10
measured read speed = 6,074 Mbytes/s (Everest)
RAM Bandwidth = 4,536 Mbytes/s (Sisoft)


New PC
Core 2 Duo 2140 Overclocked to 2.2 GHz
FSB 275 MHz
Ratio FSBRAM 1:2 Giving Ram Freq= 550 Mhz (from CPUZ)
Setting 5-5-5-15

measured read speed = 6,938 Mbytes/s (Everest)
RAM Bandwidth = 5,626 Mbytes/s (Sisoft)

Both PC's running 2 DDR2 sticks, DUAL interleaved.

As the RAM frequency has doubled shouldnt that be much more significant than the slightly slower memory timings of 5-5-5-15 ?

Thanks for any ideas

Regards Geoff

 

[systemlord]

Did you know that theres isn't a huge difference between DDR2 800 & DDR2 1250MHz. There are some difference's: DDR2 1250MHz overcomes the high latencies with more MHz. Its a MHz to latency balance. Some people like running DDR2 800Mhz @3-3-3-8, or 5-5-5-15 @1250Mhz. The real fact is for gaming high Mhz is better as games like Mhz more. I just upgraded from DDR2 533Mhz to DDR2 1200Mhz and saw a huge diff. Try tightening up your timings a little and I bet you score higher. DDR2 1250Mhz starts to show some incredible memory speeds passed 400 FSB.

 

[GeoffW]

Hi

Thanks for the response

I wasnt expecting to see a doubling of benchmark even though I doubled the clock to my RAM, but I was expecting better results than I am seeing. The latency and 5-5-15 timing must have way more impact than I guessed.

These memory spec. figures get confusing, you quoted you had gone from 533 Mhz to 1200 Mhz. These cant be the raw input frequencies ? You are quoting after doubling the real frequency due to the 2 transfers per clock I think ?

If I quote my figures in the same way that means I have gone from effective 533 Mhz to 1100Mhz. Just trying to get these figures clear in my head.

Incidentally, my new memory is Crucial Ballistix, seems good so far, I should be able to push it a bit harder yet but havent tried till I figure out what overclock my new cheapo CPU can handle

Regards Geoff

 

[systemlord]

Hi

Thanks for the response

I wasnt expecting to see a doubling of benchmark even though I doubled the clock to my RAM, but I was expecting better results than I am seeing. The latency and 5-5-15 timing must have way more impact than I guessed.

These memory spec. figures get confusing, you quoted you had gone from 533 Mhz to 1200 Mhz. These cant be the raw input frequencies ? You are quoting after doubling the real frequency due to the 2 transfers per clock I think ?

If I quote my figures in the same way that means I have gone from effective 533 Mhz to 1100Mhz. Just trying to get these figures clear in my head.

Incidentally, my new memory is Crucial Ballistix, seems good so far, I should be able to push it a bit harder yet but havent tried till I figure out what overclock my new cheapo CPU can handle

Regards Geoff

I meant from 533Mhz effective to 1200Mhz (each dimm @600Mhz) effective, these Ballistix's can do 1200Mhz on only 2.25 volts. Higher bin micron memory chips like these can do a lot with little voltage. Get the Ram Fan from CoolIt at NewEgg, your dimms will thank you for it by prolonging the life of these great chips.

 

[Mondoman]

GW, on current Intel systems, all data to/from RAM has to go through the FSB to/from the northbridge (which controls the memory bus). Thus, the FSB normally bottlenecks transfers to/from RAM, assuming you are running in dual-channel mode.