RAM Timings vs Mhz for SPI Times

[cyberexia]

Ok, here's a dumb question. I just thought about it while I'm here at work...
If I'm running 295x9 w/ RAM @ 1:1 3-3-3-6 and getting:
SPI 1M = 30.65s
SPI 32M = 26m 50s

Then If I were to use a ram divider to bring it down to 205mhz 2-2-2-5 would I see an increase or decrease in SPI times?

[Vapor]

It would probably be slower. Good question though....I'm really not 100% sure.

[Minnyboy]

Only way to find out is try...

No one has the exact same setup as you (maybe..) so you're probably better off to give it a shot & see for yourself...

[Vapor]

Actually....I do have the same setup as him (and so do others), just I never bothered to try 2655/13 2-2-2-5. I honestly can't imagine how the 205MHz would be faster on further thought....ns latency of the 295 setup is nearly identical AND it has much more bandwidth.

[cyberexia]

yeah, i don't think it would be an increase, and that the bandwidth would be so much of a loss that it would be a significant decrease anyway. But I will try it none-the-less.
Vapor, what are your times if you have the same setup?
Am I to worry about the times I have? I was hoping to be down in the 28-29s 1M and 25m 32M...but I guess I need a single core proc oced past 2.8Ghz to reach those?

[Vapor]

IIRC, 29s are very possible at around 2.7GHz and Ballistix. You might even be able to get it now with some tweaking (like using a different OS). I couldn't tell ya what my times are--my Ballistix are getting ready to go on the RMA-bus (again) and it's been 2 months since I ran SPI on this CPU or on an A64 clocked like this CPU (I've run it on my Opty).

[cyberexia]

well when my opty 175 gets here, hopefully all these scores will increase :) it shipped out today, YEAAA

[carmelo]

Iv notice an increase from having aggresive memory timings. I was getting about 34.4 @ 2600 - 2-3-2-6. At 2580 I have timings of 2-2-2-5 and i get 33.9 seconds. Still has plenty of tweaking .. but im waiting for my new blocks and what not to come in.

[cyberexia]

cool.....i tried it out...for some reason when I run a 166 or 150 divider it acts funny. for instance, I was running 230x11 w/ 166 divider. that puts ram @ 180mhz 2-2-2-5. this booted..
However if i go 235x11, it gets a little iffy... 240x11 is out of the question. my ram is good for 2-2-2-5 @ 200mhz 1:1
I don't know what's up with these dividers making it all stupid.
Really, it doesn't matter to me, b/c I won't run it at this anyway. I would have to run it around 250mhz but the timings would be the same as the ballistix's downfall is mid-range timings :(
Guess I'm going high bandwidth mediocre timings as normal! 300mhz 3-3-3-6 is good for me.

anybody got any ideas how to get this divider thing a little more stable though?

[SpookyMoomin]

I spent last night wondering the exact same thing - I got a p4 650 and some v.loose Samsung DDR2 - but the theory is the same i guess... With loose timings(4-4-4-11) on 3:4 (fsb:RAM) divider I'm about a second quicker on 1M Pi than tight timings (3-3-3-8) on 1:1; at 250MHz fsb for both tests. So MHz gets my vote.

[jrw]

Test it and show us some screenies ;)

[Mr.President]

hm in pi im faster with 295 2,5 3 3 than 326 2,5 4 3 with an opteron wc @ 3260

[DStealth]

If I'm running 295x9 w/ RAM @ 1:1 3-3-3-6 and getting:
SPI 1M = 30.65s

U mean 31.65 right?:)
290x9 1:1 2-2-2-5 ~30.950 (http://img372.imageshack.us/img372/8786/dddd5rr.jpg)
270x10 1:1 1.5-2-2-5 ~30.670 (http://mail.skgt-bg.com/dfi/a64_nf4_2700_14mod.jpg) My best

295x9 1:1 3-3-3-6 30.65 :confused:

[stummerwinter]

Here you can see my results for SuperPi @ A64 from Frequency vs Timings:

CPU: A64 3200+ Venice @ 2 GHz - fix CPU-Frequency by Multi
MoBo: DFI NF4 Ultra-D with 702er BIOS
Graka: X800 XL
RAM 1: OCZ Platinum EL PC3200 mit TCC5-Chips
RAM 2: OCZ VX PC3200 mit CH-5-Chips
OS: WXP SP2

http://oc-x.de/Dia_1M_relativ.jpg

SuperPi 1M in more effected by the Timings...;)

Complete Test, but in German: http://www.computerbase.de/forum/showthread.php?t=161595