Strange Behaviour P35 NEO2 Memory Throughput

dazman · 22 Apr 2008 at 07:54

Just Found something really odd. I have my MSI NEO2 FR jumpered to the 333 strap rather than the 266 it comes out of the box. There is no setting in the Bios to choose this so you have to change the jumpers. This gave me a big boost in memory performance (from 6500MB/s range to 8600MB/s range)so if you have one of these boards I suggest you do this!

The strange behaviour is that if I boot with FSB of 480 FSB pushing my memory at 1200Mhz and then down clock to 470 FSB using MSI's Dual Core Center with mem at 1175Mhz, I get Sandra XII SP1 memory throughput at 8735. If I boot with FSB of 470 I get Sandra XII SP1 memory throughput at 8550. This is using Memset with the same settings in both modes. This doesn't make any sense to me. I know it is only a 2% difference but a performance gain with the same settings and just a different way of getting there seems quite strange!

Anybody got any ideas why this would be?

Daz

Mr Paul · 22 Apr 2008 at 08:20

NB Strap

dazman · 22 Apr 2008 at 11:44

Mr Paul,

Maybe there's something about this I don't know but here is my understanding:

1. I have my M/b hard jumpered to the 333 strap as I said in my 1st post. As far as I know that is the NB strap and I cannot change it anywhere as jumpering is the only option. No BIOS option.

2. The FSB/RAM divider stays constant at 4/5 (1.25) times FSB under the 470Mhz FSB straight boot or the downclock in Windows from 480Mhz boot to 470Mhz in Windows so this, in theory, should not be changing any of the firmware settings.

3. All adjustable memory timings available in Memset (including performance level) stay the same in both scenarios.

4. The only difference in in the boot time FSB

5. Which strap it is running on has a major impact on performance and this is only a 2% difference.

Could you let me know if I got something wrong here or expand on you explanation?

Thanx in advance,

Daz

Mr Paul · 22 Apr 2008 at 12:25

Hmm if it's hard locked at 333 then I guess it not changing (didn't realise it had a jumper)

It "could" be that the BIOS is changing some other setting not selectable at certain FSB settings (MSI boards are known to do this in days of old).

dazman · 22 Apr 2008 at 18:19

Sandra doesn't really do a benchmark!

Just got home and did some more testing. It is the way that Sandra works not any real increase. After I looked at the result details I noticed that it thought my memory was running at 1212Mhz as it appears to be calulated from something stored in the BIOS rather than what the current FSB/Ram ratio status is, but it was really running at 1175. So Sandra does not appear to do a real benchmark it does a calculation at least partially based on what it thinks the memory is working at!!!!!!! Everest does it correctly and I get the same readings using both scenarios. Really bad of Sandra and totally undermines the purpose of benchmarking as you cannot get more synthetic than that!

Daz