CAS1.5 vs. CAS2 (My Findings)

Well this came up today as to whether CAS1.5 offered any real gains over CAS2. And although it's been discussed before, and openly stated by AMD that their CPU's do not support CAS1.5 why does everyone still run it? I spent the last hour testing this theory so i can have some hard evidence of my own.

What i did was test 6 common benchmarks using an identical system with CAS being the only changes to the system. My goal was to see if having that sexy looking CAS1.5 actually offered anything over CAS2. Well lets see what i found out.

Test Setup:

DFI NF4 SLI-D 6/23-2 Beta
AMD X2 4400+ Toledo 1.55v 2900MHz
Sapphire x850 XT Cat 5.7 600/600
2x512 OCZ Gold VX PC-4000 3.4v
2x250GB Hitachi T7K250 RAID 0
PCP&C 510 Deluxe SLI
Windows XP Professional x64 Eddition SP1

Results:

Super PI 1m:

CAS1.5 - 28.219



CAS2 - 28.250



Sandra Bandwidth

CAS1.5 - 7,596



CAS2 - 7,593



PCMark04:

CAS1.5 - 8,765



CAS2 - 8,776

Everest Latest Beta:

CAS1.5 - Read - 7,329

CAS1.5 - Latency - 35.3





CAS2 - Read - 7,335

CAS2 - Latency - 35.3





PCMark05:

CAS1.5 - 6,868



CAS2 - 6,817




Conclusion: Well there you have it. It's pretty obvious to me CAS1.5 offers nothing over CAS2. All results are within such a small margin that the differences are negligable. With that said, there is obviously no reason not to run CAS1.5 at the same time. It is certainly the placebo effect in my personal opinion. The only reason i can see to run a CAS of 1.5 is for the plain looks of it.

Nice thread, and very interesting results. I must say I'm surprised!

Dread-Star said:

Sucka,

Again great comparison! This answered many questions about settings and overall performance.

As I can see by your testing and the testing I have done there is not a very noticable difference in overall performance.

Notice my scores with what appears to be much loser settings yet the results are so close that there is really no benefit to run any tighter @ say higher voltage.


Mushkin Redline @ 11x262 1:1 Sandra>Super Pi :

Sandra 7556 ; Super Pi 1M 28.484s

Click to expand...

Thanks for the kind words.

Looks to be right about on target. You also need to remember Toledo's are just slightly slower than San Diego's clock for clock, so i think it's the memory timings more than anything there. I'm also using Windows x64.

I had kinnda known this/read about this for awhile now, but i'm glad that i have some evidence to back these statments up. If nothing else, i proved this to myself

Nice one Sucka.

From my findings, going from CAS 2.5 to 2.0 doesn't increace performance a significant amount. I ran PCmark04 on my sig computer at 2.5CAS, then at 2.0.

2.5 CAS: 4219

2.0 CAS: 4248

While there is an improvement, it is very neglegable.

I will try SuperPi now, just for kicks. I know the best time I got at 2.5 CAS was about 40.1. I will close everything, and try it now to see if there is an improvement.

Addon: Okay, when I tried with 2.5 CAS, my results ranged from 40.08-40.250 after multiple runs. I could never get below 40. But with 2.0 CAS, I range from 39.875-40.06. While it is not extremely significant, it is notable. Since the system is stable, I guess I'll leave it at 2 CAS. It won't boot at 2-2-2-6 at stock voltage, though.

Good info for Athlon 64's. I'd like to see it on an Athlon XP or Pentium 4 where the memory controller does support the possibility of CAS1.5.

Great post like all your comparisons. Thanks.

Sucka

Conclusion: Well there you have it. It's pretty obvious to me CAS1.5 offers nothing over CAS2. All results are within such a small margin that the differences are negligable. With that said, there is obviously no reason not to run CAS1.5 at the same time. It is certainly the placebo effect in my personal opinion. The only reason i can see to run a CAS of 1.5 is for the plain looks of it.

Click to expand...

Ah, but there is another legitimately good reason to run a CAS 1.5 latency, besides the sex appeal.

This timing is different than CAS 2 - when set in the BIOS, it alters some paramater (the details of which are unknown to me), which does affect our memory behaviour. Oskar Wu has done something with this timing... it tweaks something. We could call it CAS ?.?, or CAS 2.x if we wanted to - it is NOT legitimately a CAS 1.5 value (reserved).

The affect on performance is quite marginal, as you have indisputeably proven (It's awesome to see more of your great work, by the way!! ). But performance isn't the only thing that our memory tweaking is about, right?

The affect of DFI's "CAS 1.5" CAS latency value on scaling, and memory overclockability, is somewhat signifigant for many people. My memory will not run clean through memtest at 278 MHz with a CAS 2 latency... but it does with a reserved CAS 1.5 value.


Oskar has done something similar with tRAS - we have a few reserved tRAS values available to us in the DFI nFarce4 BIOS, that are fully functional values (and I assume/guess that they also adjust some basic, unknown, parameters past the adjustment of the "normal" values), if not actually adjusting the timing as implied by the value's name. A tRAS of 3 is also reserved, and also gives us marginal performance gains with low-latency memory... but my BH-5 sees overclocking gains from it.

Try a tRAS of "00" with some TCCD/TCC5...


What we have here, are some neat little tweaked timings to play with. The labelling of these tweaked settings, using reserved and non-existant values, is deceptive to their true nature.

CAS 1.5 isn't REALLY CAS 1.5. tRAS 2/3/etc. isn't REALLY such a tight latency.

The question I am really, really keen on having answered, is "what specifically and exactly do these settings change?"

What are your thoughts?

I have heard this same thing before, and plan to test if it actually changes the overall stability of the memory at given speeds. On a performance standpoint it's pretty obvious it doesn't actually do anything one way or the other. I did the changes in BIOS, not using A64Tweaker as well, it's there just for timings. I'll see if i come up with anything, but i somehow doubt it will be significant. Thanks for the input.

Sucka, you might find this thread on XS to be of interest: http://www.xtremesystems.org/forums/showthread.php?t=69242&highlight=CAS+1.5

Like me, several DFI nFarce4 users who have posted in that thread have noticed that these reserved values are beneficial to their memory's overclocking.

Thanks. I saw that thread and actually posted in it

Fact still remains that clock for clock, there is no advantage between the two. I'll test further, but at the same clocks CAS1.5 in my opinion offers nothing over CAS2. Again, that is at ***HTT, not based on max clocks.

I'll let you know what i find. Do you have any evidence to counter this yourself? I would be interested in seeing some results myself. Thanks.

Sucka

Fact still remains that clock for clock, there is no advantage between the two. I'll test further, but at the same clocks CAS1.5 in my opinion offers nothing over CAS2. Again, that is at ***HTT, not based on max clocks.

Do you have any evidence to counter this yourself?

Click to expand...

Nope, I don't have any evidence that CAS 1.5, or reserved tRAS values offer any sort of meaningful performance gain . I'm not saying that they do. I run CAS 1.5, and tRAS 3 purely for the sex appeal, and for the stability gains that my specific memory sees with these reserved values, as set via the BIOS. At a higher clock frequency, the results will be the same.

felinusz

The affect on performance is quite marginal, as you have indisputeably proven (It's awesome to see more of your great work, by the way!! ). But performance isn't the only thing that our memory tweaking is about, right?

Click to expand...

There is one thing that puzzles me a great deal. Your results show that some very, very, very slight difference is present between these two BIOS CAS values, if only from a performance standpoint. My BH-5 memory, and several other user's BH/UTT memory, behaves differently with the two value - there is some difference from a stability standpoint as well.

We know that the CAS 1.5 value in the nFarce4 BIOS isn't actually a CAS 1.5 latency (reserved). We also know from the data you have collected, very slightly different memory performance, and from reported overclocking differences from multipler users including me (ALL anecdotal evidence, so far), that it isn't the same as CAS 2, and isn't actually CAS 2.

So, what is it?

I wish i were an expert on this and could better explain how things actually work, but alas i can't really comment on the technical aspects of this.

With that said to the end user it's all the same at the end of the day. CAS1.5 will perform within a reasonable margin of CAS2 to not make a difference at all (imo). The differences in my results i posted are all negligable however. If i ran each test 10 times and took an average i can assure you the results would be within a few % of each other.

Thanks for the info Sucka (your work makes us all a little smarter )...... I would love to know what AMD would think of this kind of data? I myself have seen little to no difference between cas2 and cas1.5.

CAS 1.5 improved my OC ability. I was only ably to get 258 stable, now I am at 262. It is just a little, but it helps.

thanks for the info, informative as always.

Glad some of you like the read

I did a little testing last night and having CAS1.5 didn't change my overclock at all. All i did was change the CAS and ran Memtest #5 and came up with the same results. While i'm glad some are having mixed results on stability i guess it will very from set to set? I can't do enough testing to conclude this one way or another unless you folks want to flood me with UTT memory for testing

Dread-Star said:

Sure bro let's hookup I live in Temecula, maybe you can get these sticks of mine to pass 262 .

I had a bad bios flash last night and my PC is down for a few days until the new bios chip arrives. So this ram is just sitting here.


Eric

Click to expand...

Yeah, you're close to me. Shoot me a PM sometime if you would like to do that

Great job Sucka. you should round of your tests by doing a CAS 2.5 and CAS 3 run also.