How memory timings on A64's affect performance - The truth.

[beau_zo_brehm]

BUMP

let's see some more test cuddles

[madgamer]

Any news on this? frequency vs. latency etc? I look forward to another update.

[kryptobs2000]

I'll do some test if you guys want just let me know what you want tested. I can get the memory in my sig anywhere up to 265 1.5-2-2, can't post at cas 3 tho, since it's bh5.

[FallenAngel]

It's a nice thing you're doing 'kryptobs'

What I would like to know is if there is a performance hit for just playing if you got tight timings and low frequency or in the other hand have loose timings and high frequency, I got a P4 and tight timings doesn't do nothing at all but high frequency yes, I'm switching to a AMD rig so before buying this or that mem I would like to know what's best for playing, do I bought VX for tight timings or GSKill or other for higher bandwidth or frequency, bear with my english please...

[Rabbi_NZ]

Very nice write up cuddles

[kryptobs2000]

It's a nice thing you're doing 'kryptobs'

What I would like to know is if there is a performance hit for just playing if you got tight timings and low frequency or in the other hand have loose timings and high frequency, I got a P4 and tight timings doesn't do nothing at all but high frequency yes, I'm switching to a AMD rig so before buying this or that mem I would like to know what's best for playing, do I bought VX for tight timings or GSKill or other for higher bandwidth or frequency, bear with my english please...

By playing, you mean games? Just lemme know what you'd like me to bench.

[FallenAngel]

By playing, you mean games? Just lemme know what you'd like me to bench.

Thanks, I already know what i want, don't need to bother you, thanks very much for your offer

EDIT: by the time I posted this, I had a pm from a friend who explained me alot, and advised me to pick a mem for high bandwidth

[dexx]

Ras to cas is the most important factor when it comes to timings on A64. lets do another test. how about VX 2-2-2-10 1T 260mhz vs. TCCD 2.5-3-3-10 1T 300mhz

Just a quick test using 3dmark2001. Running the memory of my 3000+ s939 at 240MHz 3-4-4-8 gave a score of 26618. Running 240MHz 2.5-3-3-6 gave 28812. Equates to an 8% improvement.

[wfarid]

wow great job cuddles. It's been mentioned before but I really like how you have organized everything (paying attention to gamers, benchers etc), can't wait to see more tests results!

[saaya]

beautifull!

as soon as the real front page is back this is going to be posted there as an article
(if you dont mind)

[DeMagH]

first of all:
i LOVE ur work/idea, nothing to add really .. u covered it all .. waiting for the other results

second:
just a thought .. those results were taken with a flashed to PE card, means .. a powerful card, would there be a greater gap if u changed to something with lower graphics performance, like 6600GT, 9800 pro ... etc.?! take that in consideration please.

[fareastgq]

that's some great info.. props to u again.

[isp]

Looks like this little gem was hidden for a couple months, nice info though. Thanks.

[WiCKeD]

First, I just wanted to say there are too many stickies in this forum! It's getting crazy. That said, this should be one of them.

Great test cuddles. Very amusing to see the differences in real world gaming - I never would have guessed! Just checking their is no fps cap on D3, right?

For everyone implying the results are bogus, because cuddles ran the results separately... well it's plain to see that if the timings do not do much individually, things aren't going to be much better as a whole.

Here, I did a few tests awhile back on an XP platform, mainly using 3DMark01. It should give everyone an idea of the largest impact timings are going to have on performance, since '01 seems to be the most affected. There is definitely nothing to get too excited about: http://forums.extremeoverclocking.co...dwidth+timings

So figure best case scenario, you get a 3.5% performance increase in 3DMark01 between the worst and the best timings (and this is on an XP where timings are a little more important)... Turn that into MHz and that's less than about a 90MHz difference in performance (2500 *.05). In actual games, it looks like it is way less than half of that!

Now that is not including an increase in fsb, which would contribute a handful of percentage points more, but you get the idea. Bottom line, if you're deciding between spending $300 on any component in your system and you intend to game, spend that extra $200 or so on the video card over some shiny, overpriced RAM.

Me, I have money to burn, so I can afford to be a hypocrite!

[demonR6]

WiCKeD.. are you using the single channel turbo memory from PQI? I am using this currently in my set up and when I drop it to 166 I can play with the timings and OC the FSB on the CPU to 255x10 LDT 3.. when I try to take the mem back to 200 1:1, any higher FSB than 230x10 takes a dump. Same thing with the LDT, if I lower it and try raising the FSB it locks up from the get go.

Last week was a lot of frustrating time and error. It seems like I have an old revision of the 3000+ Clawhammer that is castrated. I am able to crank up the voltage on it high as heck without it even blinking though. I am not well versed in memory yet so I am having a hard time trying to learn how to best tweak it. I did have the box running at 250x10 LDT3 vcore 1.75 vdimm 2.8 with the memory at 166 5:6 2.5-3-3-7 and it runs through SuperPi @ 32m without problems in about a half hour or so. I am able to run games without problems either. Running the memory at 166 though is a crutch since I have to run it slower than 1:1??

It sucks because as soon as I try running the mem at 1:1 my FSB takes the hit and 230 is more or less my peak.

[M.Beier]

I wonder why some of you test in 2-2-2-"10"

Am I totally wrong or....
A64 gains best performance by using 2-2-2-"5"..

I get best results at 2,5-2-3-5 compared combined with max-clock on my ballistix PC4k :\

Best regards
12k

[isp]

I've always thought that a tRAS of 11 yielded best performance on socket 754. I think 6 is the lucky number for 939.

[STEvil]

11 is fastest on nF2. 5 gives fastest latency results on nF2
5, 10, and 0 have proven to be fast on S754 and 939.
2 is an oddball, its doing the best on 939 so far, but nobody has tested on 754 yet that I know of.

1_2_Know: Try 2.5-4-2-x (cas/trcd/trp/tras) on the Ballistix.

[dexx]

11 is fastest on nF2. 5 gives fastest latency results on nF2
5, 10, and 0 have proven to be fast on S754 and 939.
2 is an oddball, its doing the best on 939 so far, but nobody has tested on 754 yet that I know of.

1_2_Know: Try 2.5-4-2-x (cas/trcd/trp/tras) on the Ballistix.

Has anyone with Ballistix/UTT/BH5 tried 2.5-2-2-10 on an NF4 system?

[XanderF]

Did a thread like this with tRAS and Mhz ever get created at any point?

Those posts are still blank...

[G4h4o8s6T]

Yeah Im sure this might have been mentioned somewhere in this thread, and is probably common sense to most of you...........but I still dont know whether its better to run 200mhz 2-3-2-5 with a divider or 1:1 280mhz 3-3-3-8??? All I do really is game, and Ill be running my opty at 2.8ghz 24/7.

[uOpt]

Yeah Im sure this might have been mentioned somewhere in this thread, and is probably common sense to most of you...........but I still dont know whether its better to run 200mhz 2-3-2-5 with a divider or 1:1 280mhz 3-3-3-8??? All I do really is game, and Ill be running my opty at 2.8ghz 24/7.

My results here:
http://cracauer-forum.cons.org/forum/viewtopic.php?t=27

The last entry has a high-clocked Opteron. With that, BH-5 @ 250 MHz 2-2-2-5 beats TCCD @ 322 MHz 3-4-4-8.

[WeStSiDePLaYa]

still updating this cuddles? nice to see a fellow from ontario making real sweet guide.

[slavearm]

Maybe this is a good place to post since people are actually using facts...

I have two options to run my memory, and this is the best I can manage with my 2GBHZ

255Mhz 3-4-4-6 1T
or
280Mhz 3-4-4-8 2T

Thanks in advance.

[uOpt]

Maybe this is a good place to post since people are actually using facts...

I have two options to run my memory, and this is the best I can manage with my 2GBHZ

255Mhz 3-4-4-6 1T
or
280Mhz 3-4-4-8 2T

Thanks in advance.

The 1T option is obviously faster.