CAS 1.5 vs CAS 2.0 vs CAS 2.5

[GreenBeret]

I have always been wondering about the real effect of CAS setting on performance, especially since I got my Pentium 4 rig. The following is my little review on that

Configuration:

CPU: Intel Pentium 4 1.6A @ 2.75GHz (172.5x16)

Motherboard: Abit BD7-II RAID (AJ BIOS)
Settings: PCI bus = 33 MHz
System BIOS cacheable = Disabled
Video BIOS cacheable = Disabled
Video RAM cacheable = Disabled
AGP speed = 4X
AGP Aperture Size = 64 MB

RAM: Samsung "CTL" PC2700
Settings:
RAS to CAS Delay = 2
RAS Precharge = 2
Cycle TIme (Tras) = 5
Enhance DRAM Performance = Enabled (DRAM Idle Timer = 0)
FSB:RAM = 1:1

Graphic card: ATI Radeon 8500
Settings:
Clock speed: 301.5 Mhz core / 301.5 MHz memory
BIOS: 1009
Driver: Catalyst 2.3 ver 6166

HDD: Seagate Barracuda IV on HPT 372 RAID controller

OS: Windows XP Professional

3D settings:
Resolution: 1024x768 x 32 bit, 120 HZ
FSAA & AF = OFF
Texture and Mipmap quality = Highest
VSYNC = OFF
TruForm = OFF

If the difference is under 2 fps, it is insignificant and within the error margin.
I don't run the game tests at 640x480x16 bit (so that the video card is no longer the bottleneck), as nobody plays games at that resolution, and I want to see the performance benefit that the CAS setting brings at the most frequently use resolution: 1024x768x32bit.

OK let's benchmark !

Sandra 2002 Pro mem:

cas 1.5: buffered: 2654/2653 unbuffered: 1376/1379
cas 2.0: buffered: 2654/2654 unbuffered: 1403/1399
cas 2.5: buffered: 2653/2655 unbuffered: 1407/1398

Comment: I'm very surprised with the cas 1.5 unbuffered result. I have reran the test many times to confirm it. Strangely, the result is always worse than cas 2.0 and cas 2.5

Science Mark 2.0 beta:

cas 1.5: 2444.72 MB/s
cas 2.0: 2474.04 MB/s
cas 2.5: 2458.65 MB/s

PCMark 2002:

cas 1.5: CPU = 6779 / Memory = 6471
cas 2.0: CPU = 6776 / Memory = 6484
cas 2.5: CPU = 6795 / Memory = 6508

Comment: Again, I'm very surprised that cas 2.5 gives the highest score !

3DMark 2001 SE:

cas 1.5: 10600 Car Chase: 166.8/68.0 Lobby: 166.8/79.8
cas 2.0: 10572 Car Chase: 167.3/66.5 Lobby: 165.9/79.5
cas 2.5: 10621 Car Chase: 169.5/68.4 Lobby: 166.3/79.4

Comment: Car Chase and Lobby are the two most CPU and memory intensive tests in 3DMark. The highest score again is at cas 2.5 !

UT2003:

cas 1.5: Flyby = 131.0 / Botmatch = 65.0
cas 2.0: Flyby = 131.6 / Botmatch = 65.0
cas 2.5: Flyby = 131.6 / Botmatch = 65.0

Comment: The scores are identical, even in the Botmatch test, which is very CPU and memory intensive

AquaMark v 2.1: Texture = 24MB, Pixel Shader = ON.

cas 1.5: min 36.8 max 79.4 avg 56.2
cas 2.0: min 33.7 max 81.2 avg 55.8
cas 2.5: min 31.9 max 80.2 avg 55.5

Comment: It seems that CAS does affect the minimum fps, CAS 1.5 is the fastest, and CAS 2.5 is the slowest in this case.


Commanche 4: DXTC = On, AUDIO = OFF, SHADERS = ON

cas 1.5: 42.8
cas 2.0: 42.3
cas 2.5: 42.1

Comment: Commance 4 is known as a very CPU intensive benchmark. It seems that changing the CAS setting doesn't help much in performance, although CAS 1.5 is the fastest and CAS 2.5 is the slowest in this test.


Note: I will test the RAM at 3:4 soon.

[reupbert]

That is... strange to say the least. I was shocked in disbelief, so I went back into my BIOS and lowered my aggressive memory timings back to SPD defaults. I ran 3DMark2k1SE again, and came up with about 100 more points than with my prior CAS latency lowering. This could definitely be attributed to the statistical variation of 3DMark (I get + or - 100's of points on a regular basis).

There's a little blurb here about timings and front side bus. You'd think that with shorter timings (i.e. clock cycles to respond) there would be a boost in preformance. Anyone else have ideas about this randomness?

[sysfailur]

What the f...?

Well that is REALLY weird. Maybe I'll fiddle around with my settings as well and see what happens. That's freaky though. :\

[Colin]

You were running CAS 1.5 with your Samsung DDR RAM at 172 MHz FSB?!? What voltage were you giving that memory?

I can't get any of my RAM to run at CAS 1.5!

[GreenBeret]

Originally posted by Colin
You were running CAS 1.5 with your Samsung DDR RAM at 172 MHz FSB?!? What voltage were you giving that memory?

I can't get any of my RAM to run at CAS 1.5!

Only 2.67V That's all I have on this BD7-II to give her. I can run it up to 182 cas 1.5 Haven't tried higher though.

Other ram don't like cas 1.5. I've tried 2x Corsair XMS3200 C2's and none of those could do cas 1.5, even at 133Mhz.

Btw, do you use the same nick on MadOnion ? If so, long time no see my friend My nick there is GreenBeret

[Colin]

Originally posted by Niko
Only 2.67V That's all I have on this BD7-II to give her. I can run it up to 182 cas 1.5 Haven't tried higher though.

Other ram don't like cas 1.5. I've tried 2x Corsair XMS3200 C2's and none of those could do cas 1.5, even at 133Mhz.

Btw, do you use the same nick on MadOnion ? If so, long time no see my friend My nick there is GreenBeret

Hey man! Ya, this is Colin at MO... I was banned earlier this year (don't know why). Was pissed off at MO over the summer but I eventually went back there as Coca_Cola... It's a shame the place has been overrun by kids.

More of a lurker there now. Nice to see ya too dude!

[GreenBeret]

Originally posted by Colin
Hey man! Ya, this is Colin at MO... I was banned earlier this year (don't know why). Was pissed off at MO over the summer but I eventually went back there as Coca_Cola... It's a shame the place has been overrun by kids.

More of a lurker there now. Nice to see ya too dude!

Cool ! Nice to see you again yeah that place has been overrun by d!ckhead kids

Anyway, back on topic, it'd be good to see more results (cas 2 vs 2.5 vs 1.5 (if possible) ) on both Intel and AMD rigs

Cheers

[Colin]

Well when I get home tonight after work and class I'll be trying to get my BG7 to run at CAS 1.5!

If I'm successful, I'll try and get out some numbers for ya too.

[docah]

I doubt i can get my gigabyte ga 7vax to run cas 1.5 ... though it can run 2, 2.5 and 3. we'll see eh?

[Qkjhfhaiguihfma]

What we need is someone with an Abit board, KX7-333 to run these tests, since Abit puts a lot into the timings of memory. Oh wait, I have one of these boards, I'll do it.

MrIcee could probably verify all this, he is the god of RAM and memory.

PS I have almost no experience in P4 boards, but I'll post an Athlon reference.

[GVCryan]

Hey I'm a kid, just not a bad one.
I quit hanging around mo since it became a flame haven. The modding and cooling section is the only one with good people still.

Back on question.
I have always run slightly faster in 3dmark at cas 2.5 on my soyo board, however my new abit loves the cas 2.

I honestly think its mostly just board complaints. My soyo had a crap memory controller (wouldn't break 138 fsb), so I'll attribute my speed loss to that. Perhaps p4 boards suffer the same problems. Just a guess.

Hope it helped, Ryan

[Qkjhfhaiguihfma]

Okay here's the setup.

AthlonXP 1600+ @ 1754Mhz (166 x 10.5)
167Mhz on the RAM, timings as noted.
Samsung DTL 256MB
WindowsXP

[GreenBeret]

Thanks Qkjhfhaiguihfma. See ! You score higher at cas 2.5 :p Btw please run more benchies if you can, like the unbuffered test, PCMark and 3DMark

Also, before you paste the screenshots in Paint, go to Image Attributes and change the size to 1-1. That way, you can get rid of the white spaces.

[Bushboy]

This is a bit odd, I thought buffered sandra scores were almost identical with CAS2 or 2.5. That's it I have to do some testing maybe I can improve my scores using slacker timings:p

[Qkjhfhaiguihfma]

Yea. sorry about the screenshots, I forgot and it was too late to redo them. Should be better now.

[Qkjhfhaiguihfma]

Here are the results, I tried to do as some of you asked.

Test System this time differed slightly, to add another variable into the mix. The system was:

Same Abit KX7-333 (9k BIOS)
AthlonXP 1600+@ 1.4Ghz (133 x 10.5)
Samsung Original PC2700 DTL @ 133Mhz, timings as noted
Windows98SE

[QuadDamage]

Hi Mike.

i must admit i noticed no diffrence between cas 2 / 2.5 timings on P4 rig, cas 1.5 even lowered framerate in 3d mark. very strange as i always thought that cas 1.5 could improve bandwidth noticably

good job Beret. i wish i had some free time on my hands.

[GreenBeret]

Update:

I ran the tests again at the following timings:

RAS to CAS Delay = 3
RAS Precharge = 3
Cycle TIme (Tras) = 6
Enhance DRAM Performance = Disabled (DRAM Idle Timer = Infinite)
FSB:RAM = 1:1 , 172MHz.

Sandra 2002 Pro mem:

cas 1.5b: buffered: 2625/2627 unbuffered: 1259/1257
cas 2.0b: buffered: 2627/2628 unbuffered: 1280/1272
cas 2.5b: buffered: 2624/2627 unbuffered: 1263/1266


Science Mark 2.0 beta:

cas 1.5b: 2433.51 MB/s
cas 2.0b: 2393.93 MB/s
cas 2.5b: 2401.94 MB/s

PCMark 2002:

cas 1.5b: CPU = 6790 / Memory = 6391
cas 2.0b: CPU = 6777 / Memory = 6371
cas 2.5b: CPU = 6757 / Memory = 6357


3DMark 2001 SE:

cas 1.5b: 10431 Car Chase: 165.8/65.0 Lobby: 162.7/76.7
cas 2.0b: 10462 Car Chase: 167.2/66.0 Lobby: 161.8/76.4
cas 2.5b: 10426 Car Chase: 165.4/65.1 Lobby: 162.2/76.5


UT2003:

cas 1.5b: Flyby = 131.5 / Botmatch = 63.6
cas 2.0b: Flyby = 131.3 / Botmatch = 63.2
cas 2.5b: Flyby = 131.4 / Botmatch = 63.0


AquaMark v 2.1: Texture = 24MB, Pixel Shader = ON.

cas 1.5b: min 29.6 max 73.9 avg 54.7
cas 2.0b: min 30.5 max 74.2 avg 52.9
cas 2.5b: min 31.8 max 73.0 avg 54.6



Commanche 4: DXTC = On, AUDIO = OFF, SHADERS = ON

cas 1.5b: 42.3
cas 2.0b: 41.3
cas 2.5b: 41.5


I conclude that changing CAS doesn't affect performance. The performance differences between the above tests and the tests I've done before lie in these settings:

RAS to CAS Delay
RAS Precharge
Cycle TIme (Tras)
Enhance DRAM Performance (DRAM Idle Timer)

[hipro5]

Niko , you are doing hell of a job there man!!!!....hell of a job!!!!.....I'm just IMPRESSED.....Keep on good work.....and thanks for the HOURS you have spent on this for you and for the rest of us here...!!!....

[TheHunter]

Originally posted by hipro5
Niko , you are doing hell of a job there man!!!!....hell of a job!!!!.....I'm just IMPRESSED.....Keep on good work.....and thanks for the HOURS you have spent on this for you and for the rest of us here...!!!....

Yeah, awesome job m8, keep up the good work

[reupbert]

Well, this thing just seemed to get more curious as we went on. So, I too have done some testing, and the following is what I've found. The tests were by no means conclusive or accurate, as I simply rebooted into Windows, ran the SANDRA buffered memory benchmark, and repeated. What I should have done - and if someone has the time and inclination to do so - is run these tests several times and average them to find the standard deviation and so forth, very scientific like. The system I tested on is my Pentium III 1ghz with an Asus CUV266 motherboard, using a stick of Crucial.com PC2100 DDR rated for CAS 2.5 so feel free to laugh at my sub-par results.

CAS 2.5 w/ 125fsb: 866/843
CAS 2.5: 923/913

CAS 2.0: 951/922
CAS 2.0 w/ 143fsb: 1024/997

You might be wondering why I chose the nonstandard FSB for the two runs. Well, aside from 133mhz, apparently CAS 2.5 is meant to run at 8ns or 125mhz, and CAS2.0 is 7ns or 143mhz. So, the only thing I can attribute your odd scores to is motherboard memory timing issues or sticks of RAM which don't like to be overclocked.

[GreenBeret]

Thanks guys

Hi hipro5, I'm GreenBeret from OCAU

to reupbert:

CAS 2.5 w/ 125fsb: 866/843
CAS 2.5: 923/913 *

CAS 2.0: 951/922 *
CAS 2.0 w/ 143fsb: 1024/997

What FSB were those tests (marked with *) done at ?

1: I think P4 and AXP have enough CPU bandwidth for most things, that's why CAS doesn't affect performance much. Also, it seems that Intel chipsets don't need so many timing adjustments and aggressive timigns to run at full speed, while VIA chipsets do.

2: P3 has very low CPU bandwidth, therefore timings do affect a bit.

3: Sandra isn't the most accurate thing. Can you run other benchmarks ? Like 3DMark 2001 for instance ? Thanks

[OPPAINTER]

Niko,

That is a great data base your collecting there.
Good job, Great info

OPP

[Kurupt]

Niko,

As Opp said very nice work on that DB

[hipro5]

Originally posted by Niko
Thanks guys

Hi hipro5, I'm GreenBeret from OCAU

to reupbert:



What FSB were those tests (marked with *) done at ?

1: I think P4 and AXP have enough CPU bandwidth for most things, that's why CAS doesn't affect performance much. Also, it seems that Intel chipsets don't need so many timing adjustments and aggressive timigns to run at full speed, while VIA chipsets do.

2: P3 has very low CPU bandwidth, therefore timings do affect a bit.

3: Sandra isn't the most accurate thing. Can you run other benchmarks ? Like 3DMark 2001 for instance ? Thanks

Hi there Nico(Green Beret-OCAU)......

I've noticed my shelf that I do not have any difference in my memory bandwidth at Cas 2,5 or at Cas 2 when I run buffered or even unbuffered tests upon Windows based programs.....Only in Dos based programs I can see some difference in bandwidth and not at all of them.....