3-3-3-2-1T DDR2 is worth it!

[Zeus]

Wow, that was a lot of work!

Beautifully done, thanks for sharing.

[Don_Dan]

my 2 cents for 680i chipset

Oh my God, thank you very much! Seems like you put a LOT of work into this table! Again, thank you!

[Roger_D25]

I realize nobody has posted in this thread for a bit but I was late getting my first P35 board. One of the first things I noticed after getting the board stable at baseline was the huge difference in Read/Write/Latency between my old eVGA 680i and new DFI P35. Granted I didn't do an apples to apples comparison but despite what settings I use with the P35 I cannot come anywhere close to the numbers I got with the 680i (large difference)!

Is this difference between the Read/Write/Latency a direct result of the different chipset or is there something else going on here?

[railmeat]

I realize nobody has posted in this thread for a bit but I was late getting my first P35 board. One of the first things I noticed after getting the board stable at baseline was the huge difference in Read/Write/Latency between my old eVGA 680i and new DFI P35. Granted I didn't do an apples to apples comparison but despite what settings I use with the P35 I cannot come anywhere close to the numbers I got with the 680i (large difference)!

Is this difference between the Read/Write/Latency a direct result of the different chipset or is there something else going on here?

so the 680i is better or worse then p35?please elaborate a little bit

[eva2000]

X38/P35 will have better bandwidth and latency than 680i clock for clock.. just a matter of tuning P35/X38 boards memory timings/sub timings etc

[Roger_D25]

Hey Eva,
I am no expert on this stuff so please don't think I'm questioning your above post but given the results in this thread along with others I have not seen this to be true. The net Read/Write numbers seem to be considerably higher on 680i tests compared to at least the P35 numbers (not familiar with X38 data yet). Of course the whole problem could be that the results in this thread along with my own testing just didn't use properly tuned DRAM settings which I guess could account for the large difference between results. Eva do you happen to have any results you could post or link to which show some of these higher Read/Write/Latency resultts with X38 or P35? Believe me I really want to know because I'd love to figure out why I can't come close to even 10.5k with Read/Write or a Latency of under even 48ns with my P35 but could easily perform past those numbers with 680i? Thanks for the help and clarificaiton, much appreciated!

Edit: I didn't take into account the "clock for clock" aspect of what you said Eva which just got me thinking. The results I attained on my 680i were done running 1:1 (at least with my highest results). Although I ran similar clockspeed, memspeed, timings, and FSB the memclock ratio was not the same? That could account for some of the difference I'm seeing couldn't it?

[eva2000]

Forgot to add while everest read seem to have high numbers on 680i chipset it doesn't translate well to actual performance in memory bandwidth intensive apps like super pi.. So P35 with lower mem bandwith latency than 680i will still out perform 680i chipset.

Both 680i and P35/X38's bandwidth also depends on FSB and cpu speed as well

i.e.

[Roger_D25]

Very true, after noticing the difference in bandwidth between the two chipsets I also noticed how much faster the P35 is in apps like SuperPI (especially anything over 2M or so). My 32M time decreased by 1min21sec on this board compared to my old 680i! Like you have already explained, this board really shines once your able to start getting the memory setup correclty!

[Simps]

LOL
680i / p35 / x38 performance is pretty much the same.
0 - 3% avarage difference clock for clock on games, apps, etc.

If 680i could overclock a yorkfield, I am pretty sure someone like K|ngp|N would top the ORB with ultra's SLI and today's vcard drivers.

And that is with DDR2 too.

[]'s
Simps

[Cronos]

However high single-thread memory bandwidth on 680i may be, in multi-threaded scenario with C2Q it lags very bad behind P35/X38, being almost 2x (two) times! slower.

Given that many number-crunching apps. with todays Intel C2Q CPUs are limited by available memory bandwidth, 680i is a very bad choice for those.

[Simps]

Well, beat this stuff from Pyro, with P35/X38 DDR2/FSB @ 450MHz and CPU @ same clock.



I really doubt P35/X38 can do it.
680i 1T is good stuff.

[]'s
Simps

[Cronos]

What is to beat here? Single-threaded benchmark? Didn't i say 680i is good in single-threaded scenario? It is also may be good enough with Core2 Duo.
But with Core2 Quad, in multi-threaded scenario... its performance is simply abysmal.

Please run RightMark Multi-Threaded Memory test, here is the instruction:
http://www.xtremesystems.org/forums/...&postcount=288
You can also try to run Winrar multi-threaded bench.

Compare with similarly clocked memory on P35/X38 and see how bad 680i is!

[The0men]

What is to beat here? Single-threaded benchmark? Didn't i say 680i is good in single-threaded scenario? It is also may be good enough with Core2 Duo.
But with Core2 Quad, in multi-threaded scenario... its performance is simply abysmal.

Please run RightMark Multi-Threaded Memory test, here is the instruction:
http://www.xtremesystems.org/forums/...&postcount=288
You can also try to run Winrar multi-threaded bench.

Compare with similarly clocked memory on P35/X38 and see how bad 680i is!

But the idea of this thread has nothing to do with overclocking capacity, or even benchmarking for that matter.

It's about the ability to run tight timings like 3-3-3-4-1t and get wicked low latency like the above example.

[Cronos]

But the idea of this thread has nothing to do with overclocking capacity, or even benchmarking for that matter.

It's about the ability to run tight timings like 3-3-3-4-1t and get wicked low latency like the above example.

I think it is worth to know that these high single-threaded bandwidth and low latency does not translate into high real-life performance in multi-threaded apps. , thats all.
Do you agree?
I am sure very few here at XS know this.

[Xvys]

Here is the best latency I've gotten so far...

read: 10026
write: 9027
copy: 9268
latency: 52.2

Perhaps not all that impressive compared to some other chipsets, but pretty good for DDR2-667 ram. I tried 1T command rate, but it would only boot at 740Mhz with poor results. I will try with Cas4 and Cas3 to see what she'll do.
---------------------------------------
Asus P5K-D / E6300 / 2G Crucial 10th Anniv. / ATI X1950pro / 2 x Seagate 320G / Ultra 120 Extreme / FSP 600w GLN / CM690 case

[The0men]

I think it is worth to know that these high single-threaded bandwidth and low latency does not translate into high real-life performance in multi-threaded apps. , thats all.
Do you agree?
I am sure very few here at XS know this.

There are also a thousand other threads with this kind of information. Run your P35 with 1t
I have not come across much success when trying to run 500mhz 4-4-4-10-1t command rate with a P35.

Also in this thread everyone is purely running tighter timmings simply to compare latency. 680i is widely known to have an edge when overclocking and timing your memory, which is why this thread was started.
It really has little to do with multi-threade memory benchmarking.
AFAIK this thread wasn't created to boast about real life situations.

Here is the best latency I've gotten so far...

read: 10026
write: 9027
copy: 9268
latency: 52.2

Perhaps not all that impressive compared to some other chipsets, but pretty good for DDR2-667 ram. I tried 1T command rate, but it would only boot at 740Mhz with poor results. I will try with Cas4 and Cas3 to see what she'll do.
---------------------------------------
Asus P5K-D / E6300 / 2G Crucial 10th Anniv. / ATI X1950pro / 2 x Seagate 320G / Ultra 120 Extreme / FSP 600w GLN / CM690 case

Those by far are not bad results, especially for the RAM.
BUT if you want a really low latency, try dropping FSB to below 450 and use 4-4-4-6 or there about's timing You will automatically use tighter subtiming also I think, but if you want to drop latency even further (should already be below 48 at this stage) try pulling down Tref and TRC and most subtiming have minor effect.

[The0men]

I did this a while back. My read and write sucks, I was just messing around with some latency settings.

[Xvys]

I dropped the FSB to 450Mhz with 4-4-4 and reduced all the subtimings as suggested, but my latency increased from 52ns to 66ns with a 20% drop in my read/write scores, compared to 625Mhz with 5-5-5?

[The0men]

Can you get 500 mhz with 4-4-4? Because there are certain frequencies that work well with certain CAS latency settings. I will look something up and get back to you.

[The0men]

If you are interested in LOW latency and working out some optimal CAS vs Frequency setting here is a start :


A complete write up on how to calculate your own latency according to your frequency can be found here @
http://www.thetechrepository.com/showthread.php?t=160

[Xvys]

I tried tightening the timings to 4-4-4 @ 540Mhz, but the latencies and memory performance was still 15% lower than at 5-5-5 @ 625Mhz. According to the chart the 4-4-4 should be faster, but not with my setup. I also reduced all the sub-timings but the seem to have little effect on latency.

[The0men]

I just realised your useing a P35... I'm not sure where the strap change is, in fact I don't know much about how to get good latencies with it, I have not played with them alot. Put tRAS to 12, will help a little. and also yuou may need to do subtimings, I have done some small changes to those. I also use 1t, I dont think many P35 will do that though.

[firas]

LOL
680i / p35 / x38 performance is pretty much the same.
0 - 3% avarage difference clock for clock on games, apps, etc.

If 680i could overclock a yorkfield, I am pretty sure someone like K|ngp|N would top the ORB with ultra's SLI and today's vcard drivers.

And that is with DDR2 too.

[]'s
Simps

this interest me

[bf2142lol]

I'm pretty sure that 1T isn't possible on any Intel chipset with DDR2.....however I could be wrong.
Intel chipsets benefit from higher frequencies much more than tighter timings (when it comes to bandwidth)
I'd try it.....but I know that I won't POST and I'll have to swap out my value RAM so I can set the voltage, then reboot and install my Ballistix

Ive had my tracers doing 900mhz c4 1t on my x38-ds5, with 2.3volts. I'm thinking of doing a review on them exploring different timing/freq combinations 1t/2t when my board comes back from rma and i get my new WCing parts.

[KTE]

Phenom 800 3-3-3-3 tRC11 tRFC 75ns 1T (lowest options -- not supported in EVEREST yet)



Mine looks pretty low compared to the previous Phenom I tested which was 108xxMB/s at 440 4-4-4-4 1T with 1980MHz IMC. Anyway, just throwing an AMD sample in there.
No its not worth it to run those timings with this CPU unless you're work application covers rendering/encoding/decoding/compressing/decompressing.

EDIT: this is with the TLB errata patch. Without it, we're talking plus 12k.