Better performance with better ram, or not?

[acrh2]

I've put together a system like this:
CPU: A64 Venice 3800+ @ 2.8GHz (10.5x266) @ 1.65V
RAM: 4x512 Geil Ulta Platinum PC4000 @ 215MHz @ 5-5-9-2.5 @ 2.5V
Motherboard: Abit AN8 SLI @HTT speed 4x266

I was hoping to get 30s in Super_PI, but I am getting 32.
I think the ram is holding me back. It's sandra benchmark is 5150MB/s.
This piece of :banana::banana::banana::banana: ram won't even do what's in its SPD (5-5-9-2.0 @ 233, 5-5-9-2.5 @ 250).
I can't run anything faster than 225MHz even at CAS 3.
Increasing the voltage above 2.5V results in 1000-2000 errors in memtest. I did get a result once where increasing the voltage to 2.8V seemed to stabilize the ram, but I can't reproduce it, and I have already spent over 6 hours trying to tweak the ram.

Here are the questions for all overclocking gurus:
1) My other options are @ 280x10 the ram would run @ 233MHz, or @ 9.5x295 the ram would run @ 254MHz. Would I see better results in super_pi benchmark at those speeds? With better ram timing settings? How much?

2) If the answer is yes, what ram should I get? Would it be 2x1GB for 1T command rate?

3) If money is not an issue, what are my options?

[g33kski11z]

...A64 Venice 3800+ @ 2.8GHz (10.5x266)...4x512 Geil Ulta Platinum PC4000 @ 215MHz @ 5-5-9-2.5...I'm not much of a Super_Pi Expert but This Corsair Ram (http://www.newegg.com/Product/Product.asp?Item=N82E16820145564), will run CAS 3, which will help your score. But I don't know if it will get you to 30s...

[international2]

This Ram is for DDR2, the AMD haven´t got a DDR2 interface. ;)

[jinu117]

2.8ghz 32 seconds sounds like norm for unoptimized system for Super PI.... (windows XP)... Probably in very low 32s though.

Errr... I retract it... x2 3800+ running 2.65ghz with 3-3-3-8 memory on my cluttered X2 is doing 32s (running all kind of services including SQL server)... hmmm...

[JfRsQ]

I think your best bet would be to get 2 x 1GB kit to be able to use them at 1T. CAS 2.5-5-5-9 timings is somewhat loose + command rate @2T @215Mhz :(

I'm running in the low 28s with ~2950-3000 and memory 480-500Mhz 2-2-2-6 1T. I will try a run when I'll be home with ~2500-2750Mhz to compare.

[DriveEuro]

CPU: A64 Venice 3800+ @ 2.8GHz (10.5x266) @ 1.65V

There is your problem. Only use WHOLE NUMBER multipliers. Half number multipliers don't work properly on A64.

[acrh2]

what does that mean they don't work properly exactly? My system at those speeds seems to work "properly", just has slow ram.

[acrh2]

I think your best bet would be to get 2 x 1GB kit to be able to use them at 1T. CAS 2.5-5-5-9 timings is somewhat loose + command rate @2T @215Mhz :(

I'm running in the low 28s with ~2950-3000 and memory 480-500Mhz 2-2-2-6 1T. I will try a run when I'll be home with ~2500-2750Mhz to compare.

Thanks man. If you could try this
10 x 280 = 2800 5:6, 166, DDR333 = 233.33 (DDR467) 3 x 280 = 840 WITH CAS 2.5

or this

9 x 305 = 2745 5:6, 166, DDR333 = 249.54 (DDR499) 3 x 305 = 915 WITH CAS 3,
I would love to see the results.
Thanks.

[acrh2]

I suspect that my current ram is simply too old for this motherboard (ram was made in 2002, according to spd).
What are these settings in bios, and how will they affect performance/stability: Bank Interleaving (enable/disable), MTRR Mapping Mode (continuous/discrete)?

[brandinb]

dont run your ram 2.5-5-5-9 why are you doing that??? spd results are usless as they are read differently by each motherboard.

try cl 2.5-3-3-8 or cl 3-3-3-8 which are both way way faster then any cl 2-5-5 or 2.5-5-5 as cas latency has little effect compared to the other two which have a much larger role.

bank interleave has a huge impact especially with 4 512 modules as thats like 8 way interleaving so it should help alot.

with your timings fixed start at 2.6 volts and work up to 2.9 and that ram should work well somewhere in there

and like you said if money isnt a hold back get crucial ballistix pc4000 2x1gb sticks that will run 280+mhz with 3-3-3-8 1T timings so you can run 10x280 1:1

[acrh2]

dont run your ram 2.5-5-5-9 why are you doing that??? spd results are usless as they are read differently by each motherboard.

try cl 2.5-3-3-8 or cl 3-3-3-8 which are both way way faster then any cl 2-5-5 or 2.5-5-5 as cas latency has little effect compared to the other two which have a much larger role.

bank interleave has a huge impact especially with 4 512 modules as thats like 8 way interleaving so it should help alot.

with your timings fixed start at 2.6 volts and work up to 2.9 and that ram should work well somewhere in there

and like you said if money isnt a hold back get crucial ballistix pc4000 2x1gb sticks that will run 280+mhz with 3-3-3-8 1T timings so you can run 10x280 1:1

Two different motherboards (Abit AN8 SLI and Epox EP-4PCA3+) read SPD as 2.5-5-5-9. This ram did not post at 200MHz and cas3 with 8-way interleaving, so I have to use 4. Settings of 2.5-4-4-8 passed 4xtest#5 of memtest86 but failed prime95 after 4 minutes, while 2.5-5-5-9 was 8 hours stable. The most stable voltage for these sticks was 2.5V. Anything higher was unstable (this was after many hours of testing). The most common memory error occurs in the same place: 96% into memtest, or at 2002MB. I suspect that in addition to crappy timings, one of the sticks is defective.
While crucial ballistix sounds nice, how about something a little cheaper: Mushkin High Performance Blue 2x1GB?
http://www.mushkin.com/doc/products/memory_detail.asp?ID=194

[craig588]

Half multis don't exists for A64s, the FSB is decresed and the next highest multi is used. This results in a overall slower memory clock.


I have never seen something as high at 5-5-9 with DDR before.

[DriveEuro]

Half multis don't exists for A64s, the FSB is decresed and the next highest multi is used. This results in a overall slower memory clock.
Thanks for restating that. That's what I was trying to say to them earlier.

[brandinb]

Two different motherboards (Abit AN8 SLI and Epox EP-4PCA3+) read SPD as 2.5-5-5-9. This ram did not post at 200MHz and cas3 with 8-way interleaving, so I have to use 4. Settings of 2.5-4-4-8 passed 4xtest#5 of memtest86 but failed prime95 after 4 minutes, while 2.5-5-5-9 was 8 hours stable. The most stable voltage for these sticks was 2.5V. Anything higher was unstable (this was after many hours of testing). The most common memory error occurs in the same place: 96% into memtest, or at 2002MB. I suspect that in addition to crappy timings, one of the sticks is defective.
While crucial ballistix sounds nice, how about something a little cheaper: Mushkin High Performance Blue 2x1GB?
http://www.mushkin.com/doc/products/memory_detail.asp?ID=194


wholy crap man either your boards cannot handle 4 sticks of ram or your ram is some real :banana::banana::banana::banana:e

that mushkin will kill your current ram you will actually be able to tell the difference with how slow your current ram is runnign