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

[texuspete00]

Yeah, most of us knew cas didnt make a difference. EB was the flavor for awhile based on this fact. Didnt catch on as much as TCCD only to price I believe. I know you said were going to, thanks for that. I just trust that many people on the forum knew this part. On to the real deal!

Thanks for the effort though, but we'd like to see the other timings... I mean you always have the holier ras to cas at 2. Also 1T vrs. 2T in real life benchmarks. There are some arguments there. Im sure 1t vrs. 2T is a big deal but Sandra would have you believe.... well lets just say dual vrs. single channel isnt mega-gigantic in most apps for A64's so how does even this 800MB/s to 1GB/s end up changing the Doom 3 experience.

[reject]

he said hed do them in time.
to finish the whole thing i bet would take a day or two
cuddles

[[ r2 ]]

Nice comparison, as some of the users pointed out it will better to compare RAS to CAS and RAS Precharge as those timings play the role for A64s.

x-2-2-x/x-3-2-x/x-3-3-x/x-4-4-x

You will see a huge difference at x-2-2-x and x-4-4-x.
A 200Mhz o/c on the cpu will not match the above at x-2-2-x for A64.

[Holst]

This is with dual channel memory.

Can anybody be bothered to test single channel 754?

[trans am]

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

[thephenom]

Nice comparison, as some of the users pointed out it will better to compare RAS to CAS and RAS Precharge as those timings play the role for A64s.

x-2-2-x/x-3-2-x/x-3-3-x/x-4-4-x

You will see a huge difference at x-2-2-x and x-4-4-x.
A 200Mhz o/c on the cpu will not match the above at x-2-2-x for A64.

Agreed

And thanks for taking the time to do the review. CL generally matters more on a P4 side if you're running 1:1, the other timings affect A64 the most.

[kryptobs2000]

Testing 2 different types of memory is not a truly fair comparison, even at the same speed with the same timings memorys still will be a little faster or slower in many cases.

[cuddles]

Thanks for all the comments. Expect 2-3-2 vs 2-2-3 vs 2-3-3 / 2-4-4 in the next few days. Depending on the results, I might also try 2.5-3-3 as well and will also be looking into how all the differences scale with different Mhz ratings as well as CPU speed.

[kryptobs2000]

I ran a few tests. Just the HL2 stress test, everything on the highest details except no AA, or AF @ 1280x960, and 3dmark 01. These results are pretty accurate everything was consistant. They were at 2t because I cannot run 1t stable with only 2.8v and my motherboard does not like my bh5 for some reason either so sorry I couldn't test that. But it's with my sig rig. Also I can't boot at cas 3 because bh5 dosn't like that I've heard and I've never beenable to do it anyways.

HL2 Stress test:

2.5-4-4-7............... 121.5
2-2-2-5................. 128.9

3dmark01:

2.5-4-4-7.............. 18987
2-2-2-5................ 19857

Sorry it is not as comprehensive and thorough as cuddles but it should be pretty accurate results. I'm surprised there was almost a 1000 point gain in 3dmark01 I think this shows very much that timings matter a significant amount. I am going to rerun the test for HL2 at 2-2-2-5 because it jerked a few times for some reason, so I'll rerun that and post my results.

edit: I reran HL2 Stress test 3 times on each setting to make sure they were accurate. I took the highest and the lowest run and here they are.

2.5-4-4-7

highest 122.4
lowest 121.5

2-2-2-5

highest 131.9
lowest 129.3

So there's about an 8-10 fps diff with tighter timings. That dosn't really seem like alot since they are high numbers to begin with but I think that's pretty significant. And keep in mind I was at 1280x960 once you lower the settings and the resolution the gap will most likely increase.

[cuddles]

I've updated it with what I'll be investigating hopefully over the next few days (try to squeeze it in without upsetting the girlfriend). RAS to CAS, as well as RAS which will include a look at RAS to CAS and RAS at higher settings. I'll also look into compensating timings with FSB to try to find where the line draws, and whether or not timing changes becoming greater at higher FSB settings.

One problem I will face is that my ram is unable to do 2-2-2 at anything above 215 or so, so most of my tests involving FSB speeds will unforunately have to be done comparing CAS2.5 to CAS3 only, without the CAS2 ability. Expect it to be completely finished within a week (hopefully).

[Holst]

If I have time I can run some benches on my 754 setup using BH5 that will do uber tight timings. @250.

I think the difference between tight and loose may be more significant on single channel 754 systems.

[cuddles]

If you can get some solid and comprehensive results under the S754 platform I'll put it in the main post with specific S754 results. I'll shoot you a PM later today.

[kryptobs2000]

Car Low can vary so much it will skew the results horribly. (can be as much as 40fps)

I tested to see if you were right and it wasn't too off but oddly I got less fps each time I ran it

260
252
252
252
250.9

86
86
85.4
84.7
84.4

[Holst]

The first run would have had the barrel go a different place, run it over a few times more and it will go back up again.

2001 is not consistent between runs, my score changes by 200 odd points between car low runs

[cuddles]

I did consider showing the FPS for each individual test, but thought of it too late. Perhaps I will update it later, or as you suggested only use LH / LL. This isn't really meant for benchmarkers, because we all already know the answers, in general, and all already know that 2-2-2 is king. What I'm mostly looking into is real world applications as to how timings will affect your gaming, if at all. The CAS investigation has been done, and RAS to CAS is next. It will be in the same format, only changing a different clock time. I will update the main thread with all timing listings from A64, even though it's all at defaults and make sure it uses all the same settings for the benchmarks already done.

[Jupiler]

Cleaned up wingless' post.

Good work, cuddles.


@wingless : check PM.

[NoGodForme]

This thread actually deserves to be made into an article and published.
Review sites are probably scared to do it, because they don't want to make anyone mad.

I have Corsair PC3200XLPro which has been advertised at 2225 timings. The problem I have, is that most people CAN'T RUN 2225 TIMINGS TO BEGIN WITH. In my case, I have to run at 2336, or I get errors in Memtest and Prime95. When I built my machine, it defaulted to 2225, and it crashed until I figured out what was going on.

If you notice on the CorsairMicro site, they don't brag about 2225 memory on the main page. It's been replaced with flash memory and upgrades. I wonder why?

And I've also run benchmarks at 2225 vs 2336, and there is no differance.

[kryptobs2000]

lol, rma it. What kind of chips does it use?

[MrQ3W]

Sounds to me like a case of too much optimistic thinking on Corsairs side.

[esdee]

Nice test!!! it deservers a sticky!

Since you are hot from all those test, how about a sPI compare?

[cuddles]

SuperPi 1M using 100th of a second timing will be updated with the previous results, replacing CPU Arithmetic from the list. It will also be used and included in all further benchmarks.

Thank you all for the comments. I did some initial testing and it looks like the flute will play a different tune when it comes to RAS to CAS Many updates coming in the next week.

[Crankster]

cuddles, VERY nice work indeed. I really like that you dealt with the CAS "issue". Everyone seems to be thinking it's of no relevance wich you in my opinion proved it is. When presenting facts in an empirical manner it's always good to question the gospel. In fact i might chime in and do some testing, will be spending a lot of time off-line for a while. I'll put up (provided someone cares of course) some SPI 32M results with variying timings and FSB.

Anyways, i'm looking forward to your next installment.


@mods: Sticky this you must.

[iddqd]

I'm not even going to bother arguing with the op ...

[Jupiler]

Already to many stickies in this section.
This section might get a sub-section, like the AMD one, where all the "stickies" are gathered.
Until that happens, you'll need to have some patience.

[fareastgq]

I like to see this test done for timings:

2 2 2 10
2 3 2 10
2 2 3 10
2 3 3 10
2 3 4 10
2 4 4 10.....

2.5 2 2 10
2.5 3 2 10
2.5 2 3 10
2.5 3 3 10.....

and so on, you get the picture.. up to 3 4 4 10 that combined with those 3d runs should give us a very clear bench and real world pic. Each timing would have to be run multiple times of course and averaged, may be ran at 200 mhz, 220mhz, 240mhz, 260mhz, 280mhz, 300? and on in 20 mhz increments? same multi, just diff cpu spds? somone with the capability to go all the way up to 300mhz on their ram, hehe. also clarify whether it's 2x256 or 2x512 or whatever... there was a thread a long time ago where someone had done this, but I don't believe I've seen it done for a64 754/393 yet. props to you btw cuddles.