Dual Core Opteron killed my RAM Bandwidth

[Super Nade]

Guys,

Have any of you experienced massive Bandwidth issues with your dual core setups?

I get the same bandwidth as PC2700 RAM running at 2.5-3-3-x-1T. My BH5 is at 2-2-2-2-1T @ 270MHz. Now, its not just windows, but also Memtest which shows a BW of 2741 at those timings. What the heck is going on?

I did install the MS hotfix and the AMD drivers. I followed the exact procedure detailed at xtremesystems. It was useless.

Curiously, the in game FPS remains unaffected. However, SuperPi times are absolutely horrendous. What took 30s with my Venice at the same freq, now takes > 1min 10 sec!

http://img359.imageshack.us/img359/8208/bogus6ap.jpg

Apologies for the rather big image! You see what I'm talking about? What killed my BW? There has been talk that both cores need to be loaded or there is a huge latency bottleneck, but I don't see why that should be the case? No change in results even if I increase process priority or set process affinity to a particular core.

I'm lost here.

 

[DrMrLordX]

That's a first so far as I can tell. Sounds like a bad chip/memory controller. How long have you been running that thing?

 

[Ricemarine]

Specs plz.

 

[coomar]

pc2700 is not getting 2741 in memtest

sometime is clearly wrong like one of the channels not functioning

1T is not going to make a difference at that speed

run everest and read what the timings and speed are

 

[fbrdphreak]

Reformat?

 

[Zebo]

no way ram makes that much diff.. I can run PC2100 vs PC4800 and SPI scores only differ by 10%. Something is using lots of CPU cycles in background.

 

[BroadbandGamer]

Originally posted by: Super Nade

I did install the MS hotfix and the AMD drivers. I followed the exact procedure detailed at xtremesystems. It was useless.

Can you give me a link to that procedure you're talking about at xtremesystems.com? I couldn't find it.

Thanks,

 

[Shimmishim]

270 1:1 or with a divider? and which divider.

 

[Zebo]

Originally posted by: Shimmishim
270 1:1 or with a divider? and which divider.

Doesnt make a difference.. there's no way any AMD 64 running @ 2700Mhz should be scoring over 1 minute in SPI 1M.. should be half that time or less, something is seriously fluxored and it's not the ram.

 

[lyssword]

latest bios too, right?

 

[t3h l337 n3wb]

Yeah, the memory controller isn't very bandwidth hungry at all. There shouldn't be that much of a difference.

 

[RichUK]

this has nothing to do with the memory controller nor the ram, Zebo is right, something wrong is going on with the CPU utilisation.

EDIT: try and boot into safe mode and run Super PI

 

[DrMrLordX]

It's starting to look like a hoax/troll post actually *P No response from the OP.

 

[Furen]

It looks to me like you're running in single-channel mode. Check CPU-Z's memory tab to check if this is true.

 

[RichUK]

I recommend booting into safe mode as this the easiest way to determine whether the operating systems is being foully ran in its normal mode. As you all know ?Safe Mode? just loads minimal basic windows drivers to start the windows explorer shell, you wont get the fastest time your rig can do as only basic windows drivers are initiated, as a result your time will literally be effected by a few seconds, unlike your extremely poor score now. Try it out you should or hopefully find that your Super PI time is greatly reduced to something somewhat normal for your spec hardware. If this is indeed the scenario you can either try and virus scan and adware scan in safe mode and try and clean up your OS, or just simply reinstall it, which I would rather do after moving from single core to dual core.

Also check your motherboard and see if you have actually installed the RAM in the correct DIMM slots i.e. dual channel mode, refer to you motherboard manual for clarification. Hopefully this should clear up your scores within Memtest if you have indeed installed the RAM incorrectly, although this is not really the main issue, the Super PI score is.

 

[arcas]

Out of curiousity, what are other folks getting with for thebandwidth estimate in memtest86?

 

[coomar]

with an opteron 144/146, anywhere from 2400-2700

with the venice, 2800-3400

 

[arcas]

I'm running 2.5-4-4-7 1T @ 280 and memtest claims 2344 or 2402.

What are the RAM settings for your Venice rig?

 

[coomar]

same rig, venice was running 2600 (9x290) or 2700 (9x300), at 200 it was 2850, at 217 it was 3000, at 237 it was 3280, 245 it 3400

i have an a64tweaker shot somewhere

 

[coomar]

http://www.filefarmer.com/coomar/a64tweak.JPG

i think this scored 3280 in memtest

 

[arcas]

Forgive my naivity (and apologies to the OP if I'm hijacking the thread...)

Your Opteron running at 10x300 has RAM bandwidth of 2700.
Your Venice running at 9x300 has a RAM (presumably at the same 7:10 divider?) around 3000.
Why the difference?

This discussion has me wondering about whether my rig is functioning properly...

When I was setting it up a few weeks ago, I jotted down some memtest86 numbers.

2-2-2-5 1T @ 200 (1:1) yielded 1860 (CPU @ 1800)
2-2-2-5 1T @ 233 (5:6) yielded 2188 (CPU @ 2520)

On your Venice rig, you had

2-2-2-6 @ 200 (7:10 presumably) yielded 2850 (CPU @ 2610)
2-2-2-6 @ 217 yielded 3000
2-2-2-6 @ 237 yielded 3280
...

 

[coomar]

with the opteron

i've noticed a significant drop in bandwith in memtest (i'm guessing its the way its calculated, its not really showing up in other programs)

2600 in memtest was running at 3000 with the ram 215