?tRAS? 6/7: Best performance on nForce4 boards!!*updated*

MADMAX23

Senior member
Apr 22, 2005
527
0
0
Hi!

I have seen many people around with nForce4 boards using a ?tRAS? of 5, a few with a ?tRAS? of 8 and many others with a tRAS of 6 or 7, including me.
Several well known sites stated the best memory performance on nForce4 motherboards was achieved with a ?tRAS? of 7. Others stated that performance with a tRAS of 6,7 or 8 didn?t make any difference, was exactly the same.
I decided I had to test it by myself in order to see if that also applied to my nForce4 system. My Ram memory was running at 236.5 Mhz with 2-3-3-6 1T timings.

I wanted to test the impact on performance a tRAS of 5, 7 and 8 would have on my Ram memory subsystem.

IMO, the best tool for Ram memory performance comparison is RightMark Memory Analyzer. The actual version is this: 3.5 (bug fixed). You can download it here:
RightMark Memory Analyzer 3.5 (Latest version)

This tool is extremely accurate and offers some very nice features. The best one is the possibility to change your Ram memory timings ?on the Fly?. You can also test the Ram performance and stability.

Note: All these results were obtained under same circumstances (same processes running, 0% Cpu load at the beginning of each test, no reboots?).

Test Method: Every single test has been run 4 times each in order to get the most accurate to real performance average values (Ex: 2-3-3-x timings at 236.5 Mhz > I have run 4 times the ?float copy? test with a tRAS of 5, 4 times with a tRAS of 6 and so on?). So after each single test like 2-3-3-5 @ 236.5 Mhz I had 4 different (very close) average results/scores. To get the final constant average I discarded the highest and lowest result/score (bandwidth Mb/s) so that two average scores remained. With those 2 remaining scores the job is nearly done, just calculate the average value between them.
I think this is the fairest method to obtain real average values.

Ram performance test: Stream (Min 512Kb, Max 10.240Mb) > Selected tests > Float Copy (heaviest)

*All tests were run at 1T Command Rate*
*AVG value: Out of 4 average results, remember!*
*All Ram speeds/timings are stable (no errors) through Memtest+ 1.51 (Test 5 & 6)*
*My Corsair memory cannot run at 200 Mhz with 2-2-2-x nor 2-3-2-x timings on Asus A8N SLI Deluxe that?s why I had to find out the max stable ram speed at those timings. Results at higher speeds will be proportionally equivalent so do not worry about that. My Ram can go very high, past 300 Mhz at 1T (checked it out on the DFI nF4 SLI-DR a friend let me for a week time ago) but couldn?t test that because of the ASUS 1:1 ratio problem, it?s a pity!*


1) TEST 1: Ram at 2-2-2-x @ 168?4 Mhz (Cpu 2020 Mhz > 202x10, DDR333, LDT 5x)

Results with Ram at 2-2-2-5: 2726?5 Mb/s (Second best)
Results with Ram at 2-2-2-6: 2722 Mb/s
Results with Ram at 2-2-2-7: 2730 Mb/s (Best performance)
Results with Ram at 2-2-2-8: 2690 Mb/s

The tRAS 7 wins here. However, what surprises me is the fact that a tRAS of 5 performs way better than a tRAS of 6. Honestly, I didn?t expect this, I thought 6 would perform better, but here the rule 6/7 tRAS breaks out. For 2-2-2-x timings, the rule seems to be 7/5 tRAS.

2) TEST 2: Ram at 2-2-3-x @ 219 Mhz (Cpu 2190 Mhz > 219x10, DDR400, LDT 4x)

Results with Ram at 2-2-3-5: 3342 Mb/s
Results with Ram at 2-2-3-6: 3351 Mb/s (Second best)
Results with Ram at 2-2-3-7: 3354 Mb/s (Best performance)
Results with Ram at 2-2-3-8: 3314?5 Mb/s

tRAS of 7 and tRAS of 6: Clear winners here

3) TEST 3: Ram at 2-3-2-x @ 170 Mhz (Cpu 2040 Mhz > 204x10, DDR 333, LDT 5x)

Results with Ram at 2-3-2-5: 2726 Mb/s
Results with Ram at 2-3-2-6: 2735 Mb/s (Best performance)
Results with Ram at 2-3-2-7: 2728 Mb/s (Second best)
Results with Ram at 2-3-2-8: 2701 Mb/s

Here, we can appreciate that a tRAS of 6 performs better than a tRAS of 7 and the difference is quite important if you compare it to the rest of the tests. I have run this particular test 4 more times just to be sure and results were proportionally the same.

4) TEST 4: Ram at 2-3-3-x @ 236?5 Mhz (Cpu 2601 Mhz > 289x9, DDR333, LDT 4x)

Results with Ram at 2-3-3-5: 3317?5 Mb/s
Results with Ram at 2-3-3-6: 3319 Mb/s (Second best)
Results with Ram at 2-3-3-7: 3321 Mb/s (Best performance)
Results with Ram at 2-3-3-8: 3306 Mb/s

5) TEST 5: Ram at 2.5-3-3-x @ 240?3 Mhz (Cpu 2403 Mhz > 299x8, DDR333, LDT 3x)

Results with Ram at 2.5-3-3-5: 3333 Mb/s
Results with Ram at 2.5-3-3-6: 3333 Mb/s (Second best)
Results with Ram at 2.5-3-3-7: 3333 Mb/s (Best performance)
Results with Ram at 2.5-3-3-8: 3288 Mb/s

In this case, we can see that 5, 6 & 7 Tras values have achieved the same final average result. In order to choose the best performer and second best performer I had to find out the Highest Min and Max bandwidth values those different tRas values got. Results revealed a Tras of 7 get the highest Min bandwidth value and an average Max bandwidth value. IMO, it is more important to keep high a minimum constant bandwidth than to get a Max value only once, that?s why I?ve chosen Tras 7 as the best performer.

Final Conclusions:
tRAS 7: Winner. Best one for top performance over all. (Recommended)
tRAS 6: Very close to tRAS 7 on performance, specially recommended for 2-3-2-x timings. (Recommended)
tRAS 5: Not good performance (Not recommended)
tRAS 8: Worst performance of all. Do not use it (Not recommended)

You should run this tests in order to find which tRAS value performs the best for your system/rig 'cause results may vary depending on your hardware/software.

If anyone wants me to run a test with different Ram timings like 2.5-4-4-x or so, please, tell me, I will post my results here.

Good Luck!
 

Elcs

Diamond Member
Apr 27, 2002
6,278
6
81
Some decent work but Id check numbers up to around 11 or 12. NForce2's liked 11 quite a bit and you may find something like that in your testing.
 

MADMAX23

Senior member
Apr 22, 2005
527
0
0
Originally posted by: Elcs
Some decent work but Id check numbers up to around 11 or 12. NForce2's liked 11 quite a bit and you may find something like that in your testing.

I have tried tRAS values of even 15 (2-3-3-15), results were horrible from 9 and up.
I know nForce2 Mobos liked 11, and nForce3 ones liked 10, but nForce4 likes 6/7.

 

Avalon

Diamond Member
Jul 16, 2001
7,571
178
106
I found that a tRAS of 6 tended to be optimal on my DFI NF4 Ultra board. I honestly wouldn't worry what you set your tRAS to, though. It will make 0% impact in anything you do by itself as a setting.
 

Elcs

Diamond Member
Apr 27, 2002
6,278
6
81
Originally posted by: MADMAX23
Originally posted by: Elcs
Some decent work but Id check numbers up to around 11 or 12. NForce2's liked 11 quite a bit and you may find something like that in your testing.

I have tried tRAS values of even 15 (2-3-3-15), results were horrible from 9 and up.
I know nForce2 Mobos liked 11, and nForce3 ones liked 10, but nForce4 likes 6/7.

I see. Sorry if you had it mentioned above, I missed it.

Another thought, since the A64's have an onchip memory controller, could processor revison make any difference to this? Venice/San Diego came with "improved memory controllers". Any impact on optimal timings?

I think I'll leave this for people with A64 systems to mull over.
 

ericlala

Senior member
Apr 18, 2005
387
0
0
I guess my ram sucks because I cant even boot into windows with 7, using 6 I get the "an error has occured" window after bootup. I am pretty safe at 8 though and thats with 3-3-3-8 @ 193mhz... talk about value ram...;)
 

rise

Diamond Member
Dec 13, 2004
9,116
46
91
Originally posted by: ericlala
... talk about value ram...;)

lol

i'm with avalon, my sli-dr and winnie seem to like 6, although in memtest i don't see any change in bandwidth between 6/7/8 and then she falls off.

i thought the basic formula is tcl+trcd+2 = tras. of course mines at 2.5-3-3-6 so WTH do i know?
 

Painman

Diamond Member
Feb 27, 2000
3,728
29
86
You're right about 7 being the sweet spot, Anandtech's RAM roundup pointed this out too. I haven't been running my RAM optimally and I'm aware of this, I just recently put it into the machine and have been running off of a TCCD template I found, then copied and pasted basically.

This weekend I got off my duff and spent some time tweaking. I'm currently running the RAM @ 2.5-4-3-7 1T DDR560, so far so good. I think I just about have it stabilized.
 

rise

Diamond Member
Dec 13, 2004
9,116
46
91
i'll agree with you too pain. i forgot once i go over 265 i need to drop it to 2.5-3-3-7 along with another tweak or 2.
 

Shimmishim

Elite Member
Feb 19, 2001
7,504
0
76
i've done tests with my chip between 5 to 11 and i found that 6-8 actually gives me the same results.
 

RichUK

Lifer
Feb 14, 2005
10,341
678
126
how reliable are this app .. its just that i have never heard of it , not to say anything with regards to that, as they do look the part, but just out of interest..
 

RichUK

Lifer
Feb 14, 2005
10,341
678
126
So to speak tRAS can be seen as a book,

tRAS is that amount of time that you leave the book open to read a certain part of the page, if the book is closed to early (low tRAS setting) the data/words are not read properly, if the book is open to long then you have lost precious cycles .. ;)
 

kobe8

Member
Feb 12, 2001
188
10
81
Originally posted by: MADMAX23
Ram performance test: Stream (Min 2048,Max 8192) > Selected tests > Float Copy (heaviest)

I get approx 100MB/sec deviation between various test runs at the same settings. It seems like any difference could just be due to error versus a a change in the settings.
 

MADMAX23

Senior member
Apr 22, 2005
527
0
0
Originally posted by: GuitarDaddy
:thumbsup:

Try 2-3-2-7, that seems to be the sweet spot for me:)

Can't run my memory at that speed with those timings, Guitardaddy, I've tried it several times. Try tRAS 6, it may give you better performance, read my topic.

 

Concillian

Diamond Member
May 26, 2004
3,751
8
81
Originally posted by: MADMAX23
Hi!

1) TEST 1: Ram at 2-2-2-x @ 168?4 Mhz (Cpu 2020 Mhz > 202x10, DDR333, LDT 5x)

Results with Ram at 2-2-2-5: 2726?5 Mb/s (Second best)
Results with Ram at 2-2-2-6: 2722 Mb/s
Results with Ram at 2-2-2-7: 2730 Mb/s (Best performance)
Results with Ram at 2-2-2-8: 2690 Mb/s

The tRAS 7 wins here. However, what surprises me is the fact that a tRAS of 5 performs way better than a tRAS of 6.

"way better" to me is a little more than 0.15% in a RAM specific benchmark where overall performance impact will be a fraction of performance differences seen in this kind of benchmark.

Personally I go by the CAS + RAS to CAS + 2 = tRAS and call it a day. It's not worth optimizing these parameters that are going to give you an overall performance increase on the order of 0.05% or so.
 

MADMAX23

Senior member
Apr 22, 2005
527
0
0
Well, that's up to you, my aim is to squeeze my system capabilities and help others do the same.
 

Zebo

Elite Member
Jul 29, 2001
39,398
19
81
Depends on board. Depends on memory.


All tests Ram at 2.5-2-2-x @ 242 Mhz (Cpu 2655 Mhz > 295x9, 166 memsetting, LDT 3x)

Sanda CPU
8 - 12020
7- 12057

Sandra MEM
8 - 7250
7 - 7249

Everest Read
8 - 7112
7 - 7085

Everest Write
8 - 2600
7 - 2601

Everest latency
8 - 36.7
7 - 36.8

And Finally Right Mark Stream (Min 512Kb, Max 10240Mb) > Selected tests > Float Copy
8 - 3460
7 - 3333


As you can see mixed bag... definity 6-8 is where it's at though on any A64 939 setup. I run 8 YMMV...
 

MADMAX23

Senior member
Apr 22, 2005
527
0
0
Well, I haven't tested 2.5-2-2-x timings yet.
I agree with you that every system may behave differently so everybody should try testing in order to know the best tRAS setting but ,on the whole, 7 seems to be the overall winner.
Also, give Rightmark tests 4 runs and then calculate the average as I did.
 

RichUK

Lifer
Feb 14, 2005
10,341
678
126
i dont give a sh1t bout tRAS, i just aim for low CAS, that is the main performance timing from your RAM .. shame my peace of sh1t mem controllers are not able to handle 2.5 like they are supposed to ...

Quick question with soime chips not giving out high OC's due to poor yields, can this sometimes relate to mem controllers aswell .. i ask this because they are mem test stable at what the should be 2.5 no errors, but are not prime stable in XP, therefore i have to use 3, im just hoping that this is not the case when it comes to the X2's mem controllers
 

Zebo

Elite Member
Jul 29, 2001
39,398
19
81
There is a lot easier way to do this. Just boot into memtest86 and see resulting speeds of various tRAS's.

No need to load whole OS and run an app.
 

MADMAX23

Senior member
Apr 22, 2005
527
0
0
Well, very respectable opinion, but I do care of all timings. IMO, Cas latency is not the only or main important timing, tRP is also very decisive.