Cas latency test on Intel chipsets : big or small improvement

Lo lads ( and gals),

a short description why I started this thread... a while back I read lower latency doesn't bring that much on Intel chipsets... we are curently comparing the P5Q dlx and Biostar P45, P5Q dlx hardly gets a boost in ram performance with lower latency ram, the Biostar however does big time... a bios flaw as both mobo's use the same chipset ?

So if you got a few minutes of your spare time plz try the following and post ya results here plz... (a note this is not a show off how good ram bandwith or low latency is on your particular setup)

A test at C4 and C5 latency eg 4-4-4-12 and 5-5-5-15 with all the other sub settings the same... CPU speed and FSB don't matter much... Post an everest bandwith test, superpi,... and include a MemSet in the screenie... just to see how much or little the gain is...

Plz don't mess with any settings and just alter the latency and keep other values like PL and co in line...

I start off with my setup with 400FSB ram at 1066Mhz PL7 : I needed to alter the subtimings as the board set lower sub values for the C5 run...

http://users.pandora.be/OAP/P5QDlx/C4C5.JPG

Thx for any contribution :up: and plz don't just vote, add proof too...

Plz post results too... with just voting the poll, this thread is going nowhere...

We need screenies !! and plz add ya conclusion too if you find the gain negligable or big...

Thx

I have tried it with same settings .... fsb400, cpu 3.6GHz, ram 1066MHz, timing CL4/CL5-55-7-18-13-11-7-7-4/5-4171-2T ...

e8500
P5Q Deluxe Bios 0803
Ballistix 8500 2x1GB

- DRAM Static red control disabled
- DRAM Read Training enabled
- MemOC Charger enabled
- PL7
- timing CL4/CL5-55-7-18-13-11-7-7-4/5-4171-2T ...

CL4
http://img168.imageshack.us/img168/5...4resizetf0.jpg

CL5
http://img168.imageshack.us/img168/8...5resizewb4.jpg

If you need results with other settings, only write me... I will try bios 1306 later if are there some differences :up:

you still got ya commando up and running mate ? would be nice to have different chipsets, mobo types compared...

perhaps in 2 weeks it will be possible, but now I have no time to play with :( and the primary "problem" is that I need to readjust/rebuild all my WC ring to Commando because I have no stock cooling on it, only waterblocks ...

only thing I know for certain Commando boots with my Ballistix max. on 1260MHz CL5, P5Q Deluxe on 1320 MHz... timings auto, voltage 2.2V :)

Here is mine on a DFI P35.
I could not set the performance level to 7 with the cas set to 5, I tried with memset and got a hard lock and with bios I had to clear cmos.
Since I couldn't get cas 5 to run at perf lvl 7 I redid cas 4 at perf lvl 8.
CAS 4
http://webpages.charter.net/chew_toy/DFI8400/EV444.PNG

CAS 5
http://webpages.charter.net/chew_toy/DFI8400/EV555.PNG

You're going to get mixed results with different rams.

http://img147.imageshack.us/img147/1566/55515vj0.jpg

http://img216.imageshack.us/img216/7046/44412fw2.jpg

Thx lads we are getting somewhere now... If ancient could redo his C5 test at PL 7 l think his score will be only a small improvement too... till now not not much gain... can you do a test Zucker on ya magical Rampage ?

Would be nice to test 1:1 divider also, to make shure that CAS doesn't have more effect on that.

on a P5Q with bios 1306 i typically run 1020mhz 1:1 CL555-15 PL10 24/7 stable set to 1.90v bios. (Mushkin PC6400) boosting it to CL455-15 so i can run PL9 shows a slight bandwidth increase/latency decrease and lower SuperPI scores, but no 3D or real world improvement. in fact in some cases 3D was slower. going a step further i boosted to CL444-12 PL9, again pretty much the same results, a slight bandwidth increase/latency decrease and lower SuperPI scores, but no 3D or real world improvement. and again some 3D was even slower. PL8 was a no-go.

:shrug:



for SuperPI CL444-12 PL9 is a must, or i aint in the game.
http://i224.photobucket.com/albums/d...spi32m-516.jpg

Ok bud, here goes:

Everest 5-5-5-15
http://img.photobucket.com/albums/v6...25-5-5crop.jpg

Everest 4-4-4-12
http://img.photobucket.com/albums/v6...24-4-4crop.jpg

SuperPi 1M 5-5-5-15
http://img.photobucket.com/albums/v6...M5-5-5crop.jpg

SuperPi 1M 4-4-4-12
http://img.photobucket.com/albums/v6...M4-4-4crop.jpg

Motherboard is a P5Q-E P45 chipset, can't wait to fire up my Rampage Extreme.
Will do some tests on the RE as soon as i have it up and running, waiting for my mem and backplate to arrive now. :)

My RAM doesn't like much to work at 4-4-4-10 or 4-4-4-12 at this clock, but here's a 5-5-5-15 run. I never noticed in previous tests i made while getting the rig stable any significant gains by using 4-4-4-10/12.

http://www.allsinclair.com/gfx/p45_bench.png

Hey Guys

3600 (400X9) 1:1
TRANSACTION BOOSTER = AUTO
STATIC READ = DISABLED


800 CL5 XP32 SP3
http://img54.imageshack.us/img54/912...xp32lj6.th.jpg

800 CL4 XP32 SP3
http://img233.imageshack.us/img233/4...xp32hj9.th.jpg

800 CL5 VISTA 64 SP1
http://img54.imageshack.us/img54/807...ta64iz2.th.jpg

800 CL4 VISTA 64 SP1
http://img141.imageshack.us/img141/8...ta64iy9.th.jpg

How the chipset responds to CAS 3/4/5 also depends on the implementation (the motherboard).

Anyway, when X38 was launched, I did a chipset test, and one of the issues was how do different chipsets take advantage or not from using low latencies. Using the same settings (clocks) and components (except for the motherboards, of course) I got this:

http://forum.crazypc.ro/attachment.p...4&d=1200927098

http://forum.crazypc.ro/attachment.p...5&d=1200927098

http://forum.crazypc.ro/attachment.p...6&d=1200927098

http://forum.crazypc.ro/attachment.p...7&d=1200927098

The hole story is here, if anyone wants use google translate.Hope it helps :)

I don't have graphs, but I was amazed at how small the difference was at certain timings.
P35, Crucial Ballistix Tracer 4x1gb, times are SuperPi 1m (lowest of 3 runs after a reboot):

1008mhz 4-5-5-25: 13.344s
1050mhz 5-5-5-15: 13.344s
1050mhz 5-4-5-15: 13.328s

I know it looks like random tests, but it was back when I was trying to stabilize my memory overclock. I've temporarily given up.

To get a fair comparisation with SuperPi one has to run 32M as 1M is mainly ran within the CPU's cache.

Tweaking timings hardly helps any but for 32M it does.

Better to show bandwidth to compare the diff.

i found that 1:1 with good timings jacks up the copy and to a lesser extent write more than the faster speed, that said i prefer the 500fsb area wehen i have the ambient to do it and that likes the 1200mhz cas5 over 1000mhz cas4, but at slower speeds like 400mhz fsb the cas3 800mhz works better than 960 cas4 or 11xx cas 5 (i forgot teh exact number)

so it just depends on the clocks that u are doing and i noticed that with 4 sticks timings matter more than 2

I have barely any difference going from CAS 5 to CAS 4 at the same speed, FSB & PL.

DDR 1000 CAS 4, PL 8
http://i266.photobucket.com/albums/i...dr1000cas4.jpg

DDR 1000 CAS 5, PL 8
http://i266.photobucket.com/albums/i...dr1000cas5.jpg

Looking at several results with different ram/mobo combinations on current 775 Intel platform C4 brings little to none increase comign from C5...

I started this thread due to the fab speed increase on the Biostar P45 mobo... so I suspect the biostar mobo has more an issue or is a gift from the Holy man above

Here's the results of the Bio mobo

Quote:

---

Originally Posted by ci2kla

Well finally not only me have this problem so I guess my motherboard isn't faulty but maybe the BIOS...

Another friend of mine tried on BIOSTAR TP I45 and the normal result should look like....(He's using D9GMH also)
http://i317.photobucket.com/albums/m...C4JPGthumb.jpg

http://i317.photobucket.com/albums/m...c5JPGthumb.jpg

---

here is some of mine, all pl7 and 400mhz fsb
http://i34.tinypic.com/2myvi54.jpg
http://i33.tinypic.com/11v5dns.jpg
http://i34.tinypic.com/akzioo.jpg

i never tried 400/1066 with a 400mhz fsb so i was kinda surprised, but windows took about 5 sec extra to load so im staying cas 4, i would have poseted 800mhz ram with cas 3 but my nwe board dosnt like it, the new dfi x38 are much weaker than the launch