Relax the tRAS timings for better performance

[BDSM]

KhoiFather..There are nothing unusual about those scores u have. Since you are running the cpu fsb at defaul (133 mhz) you will get the best scores with mem in sync.. Period.

If you want better scores use higher fsb (in sync with mem). I'm at 220 mhz 2-2-2-11 and that yields 6000pts exactly for me.

 

[KhoiFather]

I scored a lot higher with these settings

CPU: 5110
Ram: 4352
HD: 929
Ram timings at 2-2-2-10 at 166MHz

 

[infinite012]

I read about this awhile ago at the amdmb forums (link is above) and yeah just like everyone else, 11 for tras = best/better performance than 5 or 6. I run 200MHz FSB 1:1 with memory and 11 gives out one of the better scores. However, it gives out a lower score than with a tras of 9.

This is with Corsair PC2700 on A7N8X Deluxe - 2-3-2-9 is what it's running at.

 

[maluckey]

Most people should read the disclaimers, and real world comparisons sometimes about the benchmarks. Benchies are to compare to benchies, not real world. I lost FPS in high intensity games like UT2003, period, across the board is what I'm saying. I'm not even close to being alone on this. Go to the FPS sites and explain that slower memory timings make you faster and they tell you to get better memory than Mushkin. Remember when everthing OCZ said, people believed?

 

[Bound_Vortex]

Originally posted by: maluckey
Most people should read the disclaimers, and real world comparisons sometimes about the benchmarks. Benchies are to compare to benchies, not real world. I lost FPS in high intensity games like UT2003, period, across the board is what I'm saying. I'm not even close to being alone on this. Go to the FPS sites and explain that slower memory timings make you faster and they tell you to get better memory than Mushkin. Remember when everthing OCZ said, people believed?

You didn't respond to any of my questions/suggestions. And if you are going to spout off about losing fps here and there and about these fps sites, then provide links, provide examples. You say you lost fps in high intensity games...was this online? Could it have been your computer was accessing a page file on your hard drive, were you still downloading pr0n...

And you not thinking Mushkin is a quality memory provider, you are completely mistaken. Overpriced, I would tend to agree. But apparently you didn't read any of my posts above, so I will reiterate that I referenced the Mushkin page because it was well thought out and accurately explained. And all of my comparisons were produced using Corsair XMS 3200.

 

[Bound_Vortex]

maluckey,
I tried out UT2k3 and I didn't notice any of what you speak of....i.e. degredation of performance. Using Demo version 2206 and HardOCP's UT2k3 benchmarking utility I got the following results at 1280x1024 selecting High Quality using Direct3D.

Tests were done at 210 fsb with a processor clock of 1.894 GHz at 11-3-3-2.5 and 6-3-3-2.5 to isolate the tRAS setting. I performed each test 3 times rebooting after each one.
Below are the lowest frame rates, average frame rates, and highest frame rate.

6-3-3-2.5
Test 1: (dm-antalus/dm-asbestos/ctf-citadel)
Low: (36.8/33.0/20.1)
Ave: (49.5/72.9/61.0)
High: (131.8/184.4/186.7)

Test 2:
Low: (13.2/33.0/20.0)
Ave: (46.5/72.9/60.4)
High: (131.6/186.5/183.3)

Test 3:
Low: (36.6/12.4/19.7)
Ave: (49.0/69.1/59.8)
High: (130.8/179.0/184.1)

11-3-3-2.5
Low: (36.5/32.5/20.1)
Ave: (49.0/71.8/61.0)
High: (130.3/181.4/184.2)

Test 2:
Low: (36.3/33.0/19.8)
Ave: (49.7/72.5/60.4)
High: (132.5/183.7/184.2)

Test 3:
Low: (36.9/32.8/20.0)
Ave: (49.7/72.5/61.0)
High: (132.3/183.7/188.4)

The two dips in scores occurred with tRAS set to 6. These times are probably a direct result of the memory unable to finish with a set of data and it has to be paged to the hard drive. Also notice the ranges in scores for the multiple tests. The lower memory timing generates a greater variance in performance whereas the higher tRAS of 11 produces much more solid repeatable results.

 

[PrinceXizor]

There's an interesting article explaining a whole LOT about tRAS and memory timings and such at Lost Circuits. A VERY good read.

http://www.lostcircuits.com/memory/ocz3700/

The main article is about some PC3700 Gold from OCZ...but he delves deep into latencies and timings.

Edit: Another good article where he tests some DDR400 modules and talks again about varous timings.

http://www.lostcircuits.com/memory/ddr400/

P-X

 

[Noid]

I tried all the settings,,, (my NF7-S goes to 15)

The 2 mb I gained in write speed,,, was lost in write speed (actually 4mb more)

tesing with AIDA32

This was at 215 using my Corsair PC3200c2 at 11,3,3 Cas 2.0 with 2.8v

Using 7 ,,, yielded the same results,,,

 

[maluckey]

To whom it may concern. You search button doen't work? I found plenty of goo interesting stuff, in less time than it took to change my kid. I won't even go over anything OCZ sells, prints or otherwise, as they have proven themselves to be shady, and liars. Also..Some of us have careers, kids, and are not immature enough to believe that every post will be responded to when you want it to be.

Did you notice something about your own "proof". It's what I said. You were faster in botmatches with lower Tras settings than with higher (numerically) Tras settings. Faster where it counts, on the lowest FPS you had, you were faster with lower timings than with the fast, in every case. About 1 FPS like I said....Hmmmmmmm Nuff said.

It was a typo, (I am physically disabled, having fused wrists), and for that I apologize. I meant to type No degradation in-game. because I rarely turn on the swap file, and never swap during game.

 

[Jeff7181]

Gettin nasty in here... maybe ya'll should sit down and have a nice cold :beer:

 

[PrinceXizor]

To whom it may concern. You search button doen't work? I found plenty of goo interesting stuff, in less time than it took to change my kid.

Does your Cntrl and V buttons not work? How about some links to this "interesting stuff" you are finding.

I won't even go over anything OCZ sells, prints or otherwise, as they have proven themselves to be shady, and liars.

I assume this was in response to MY post, since bound_vortex wasn't even talking about OCZ. You won't even go over a review of an OCZ product? How ridiculous is that?? Does the simple fact that a reviewer is reviewing an OCZ product taint him irreparably in some fashion?

Also..Some of us have careers, kids, and are not immature enough to believe that every post will be responded to when you want it to be.

It wasn't a question of when you responded. He was talking about when you DID respond, you did nothing but restate your claims with no facts to back you up. He has done benches and posted his own data. What have YOU done?

Did you notice something about your own "proof". It's what I said. You were faster in botmatches with lower Tras settings than with higher (numerically) Tras settings. Faster where it counts, on the lowest FPS you had, you were faster with lower timings than with the fast, in every case. About 1 FPS like I said....Hmmmmmmm Nuff said.

I don't know what you are talking about.

With tRas = 6
Average of All Tests
Low: (28.8/26.1/19.9)
Average: (48.3/71.6/60.4)
High: (131.4/183.3/184.7)

With tRAS = 11
Average of All Tests
Low: (36.6/32.8/20.0)
Average: (49.5/72.3/60.8)
High: (131.7/182.9/185.6)

Difference b/n tRAS = 11 and tRAS = 6
Low: (+27.0%/+25.7%/+0.5%)
Ave: (+2.5%/+1.0%/+0.7%)
High: (+0.2%/-0.2%/+0.5%)

So there is no performance degradation at the high end of the numbers (they are identical). The only performance degradation is on the low end with tRAS set to 6. Only ONE number in there was tRAS=6 faster than tRAS = 11. Where you got that tRAS=6 was faster from these numbers. I have no idea.

P-X

 

[Noid]

Welp,,, I went and D/L PCmark2002...

This is a very inconsistant benchmark for me. Here are my results at 220 Cas 2.0 at 2.8v

6-3-3 6851 6413 915
7-3-3 6849 6427 922
8-3-3 6847 6430 915
9-3-3 6847 6428 916
1033 6846 6429 917
1133 6831 6422 917

so,,, I picked out 8 for a re-test and I got this ... 6846 6423 919

laff

so,,, I re-re-tested ... lol (no re-boot this time) - 6848 6428 921

So,,, I guess I'll keep 8 and try it on 3dM02 and 03 next.
_______- edit - __________
3dM02 17513
3dM03 5592


Not as high as my highest score (before ATI and FutureMark adjusted drivers and pgms)

My gain was minor ( better than nothing) , and a test I never tried before.

So,,, from what I have read from the posts from members ,,, I'd have to say the results depend on the hardware

and I wouldnt rely on PCmark2002 for accuracy (in my tests)

 

[HoMeZ]

Did anyone catch TheOtherDude's formula on that link Bound_Vortex provided?

 

[Jeff7]

Is there a good tester that'll work in DOS? Peter suggested Cachemem - it runs in DOS, and gives consistent results. The only thing I wonder about is that it doesn't seem to support SSE, 3DNow, or any CPU optimizations newer than MMX. Would it still be a useful benchmark utility?
I will still give this tRAS timings thing a try though, using Cachemem, and see what it turns up.

 

[HoMeZ]

I read that forum and I THINK it said for 166 fsb I should have 7tras.. is that right?