CAS 2.0 vs 2.5 on Athlon64's

[[Mmmm]Homer]

I wanted to see the CAS timings of 2.0 vs. 2.5 on my system by pulling one stick out, then the other. My system also overclocks 10% easily. My wireless network and Norton AV were disabled during testing.

 

I currently have 2 x 512mb of RAM, but 2 different types:

 

Corsair XMS Extreme Memory Speed Series 184 Pin 512MB DDR PC-3200

Cas Latency: 2-3-3-6 1T

$118 currently

 

Taken from Corsair's website:

These parts support 2-3-3-6 latency on Intel platforms, and 2.5-3-3-6 latency on AMD platforms.

 

The other:

 

Corsair Value Select 184 Pin 512MB DDR PC-3200

Cas Latency: 2.5

$83 currently

 

 

System specs:

 

Athlon64 3000+

Asus K8V Deluxe

9800 Pro

2x 36gb Raptors in RAID 0

1x 250gb Hitachi

1x 80gb WD800JB

Windows XP Pro

 

 

CPU-Z says on timings:

 

XMS (auto): 2.5-3-3-8-11-16 Auto settings in BIOS

XMS (manual): 2-3-3-6-11-16 Manual settings in BIOS

XMS (manual, 2.2ghz): 2-3-3-6-11-16 Manual settings in BIOS, 10% overclock

Value Select (auto): 2.5-3-3-8-11-16 Auto settings in BIOS

Value Select (manual, 2.2ghz): 2-3-3-6-11-16 <does not boot>

Value Select (auto, 2.2ghz): 2.5-3-3-8-11-16 Auto settings in BIOS

Both sticks (auto): 2.5-3-3-8-11-16 Auto settings in BIOS

Both sticks (auto, 2.2ghz): 2.5-3-3-8-11-16 Auto settings in BIOS, 10% overclock

 

 

Super Pi:

1m digits

 

44 seconds XMS (auto)

44 seconds XMS (manual)

40 seconds XMS (manual, 2.2ghz)

44 seconds Value Select (auto)

40 seconds Value Select (auto, 2.2ghz)

45 seconds Both sticks (auto)

40 seconds Both sticks (auto, 2.2ghz)

 

 

PCMark2004

 

3969 XMS (auto)

3980 XMS (manual)

4376 XMS (manual, 2.2ghz)

3970 Value Select (auto)

4343 Value Select (auto, 2.2ghz)

4003 Both sticks (auto)

4365 Both sticks (auto, 2.2ghz)

 

 

Prime95 version 23.8, RdtscTiming=1

 

XMS (auto):

Best time for 384K FFT length: 23.205 ms.

Best time for 448K FFT length: 27.830 ms.

Best time for 512K FFT length: 31.251 ms.

Best time for 640K FFT length: 39.571 ms.

Best time for 768K FFT length: 48.183 ms.

Best time for 896K FFT length: 58.325 ms.

Best time for 1024K FFT length: 65.528 ms.

Best time for 1280K FFT length: 87.700 ms.

Best time for 1536K FFT length: 104.800 ms.

Best time for 1792K FFT length: 127.801 ms.

Best time for 2048K FFT length: 143.805 ms.

 

XMS (manual):

Best time for 384K FFT length: 23.207 ms.

Best time for 448K FFT length: 27.864 ms.

Best time for 512K FFT length: 31.139 ms.

Best time for 640K FFT length: 39.546 ms.

Best time for 768K FFT length: 48.252 ms.

Best time for 896K FFT length: 58.220 ms.

Best time for 1024K FFT length: 65.705 ms.

Best time for 1280K FFT length: 87.544 ms.

Best time for 1536K FFT length: 104.530 ms.

Best time for 1792K FFT length: 127.201 ms.

Best time for 2048K FFT length: 143.515 ms.

 

XMS (manual, 2.2ghz):

Best time for 384K FFT length: 21.046 ms.

Best time for 448K FFT length: 25.207 ms.

Best time for 512K FFT length: 28.415 ms.

Best time for 640K FFT length: 35.923 ms.

Best time for 768K FFT length: 43.832 ms.

Best time for 896K FFT length: 53.056 ms.

Best time for 1024K FFT length: 59.610 ms.

Best time for 1280K FFT length: 79.703 ms.

Best time for 1536K FFT length: 94.932 ms.

Best time for 1792K FFT length: 116.111 ms.

Best time for 2048K FFT length: 130.992 ms.

 

Value Select (auto):

Best time for 384K FFT length: 23.137 ms.

Best time for 448K FFT length: 27.820 ms.

Best time for 512K FFT length: 31.245 ms.

Best time for 640K FFT length: 39.509 ms.

Best time for 768K FFT length: 48.135 ms.

Best time for 896K FFT length: 58.157 ms.

Best time for 1024K FFT length: 65.540 ms.

Best time for 1280K FFT length: 87.862 ms.

Best time for 1536K FFT length: 104.755 ms.

Best time for 1792K FFT length: 127.426 ms.

Best time for 2048K FFT length: 143.797 ms.

 

Value Select (auto, 2.2ghz):

Best time for 384K FFT length: 21.114 ms.

Best time for 448K FFT length: 25.386 ms.

Best time for 512K FFT length: 28.418 ms.

Best time for 640K FFT length: 36.036 ms.

Best time for 768K FFT length: 43.885 ms.

Best time for 896K FFT length: 53.074 ms.

Best time for 1024K FFT length: 59.671 ms.

Best time for 1280K FFT length: 79.667 ms.

Best time for 1536K FFT length: 95.146 ms.

Best time for 1792K FFT length: 116.107 ms.

Best time for 2048K FFT length: 131.284 ms.

 

Both sticks (auto)

Best time for 384K FFT length: 23.146 ms.

Best time for 448K FFT length: 27.803 ms.

Best time for 512K FFT length: 31.158 ms.

Best time for 640K FFT length: 39.435 ms.

Best time for 768K FFT length: 48.036 ms.

Best time for 896K FFT length: 58.165 ms.

Best time for 1024K FFT length: 65.504 ms.

Best time for 1280K FFT length: 87.564 ms.

Best time for 1536K FFT length: 104.516 ms.

Best time for 1792K FFT length: 127.068 ms.

Best time for 2048K FFT length: 143.468 ms.

 

Both sticks (auto, 2.2ghz):

Best time for 384K FFT length: 21.087 ms.

Best time for 448K FFT length: 25.309 ms.

Best time for 512K FFT length: 28.394 ms.

Best time for 640K FFT length: 35.896 ms.

Best time for 768K FFT length: 43.713 ms.

Best time for 896K FFT length: 52.946 ms.

Best time for 1024K FFT length: 59.577 ms.

Best time for 1280K FFT length: 79.624 ms.

Best time for 1536K FFT length: 95.084 ms.

Best time for 1792K FFT length: 115.647 ms.

Best time for 2048K FFT length: 130.551 ms.

[All Kill3r]

heh, its not the cas latency that is my beef with RAM, its the middle timings...

 

that 3-3 has already lost you some performance from 2-2, but its better then 4-4. I'll see if i can't do a bit of testing between 2-2/3-3/4-4. Thats where the big lose is.

 

Nice test !

[[Mmmm]Homer]

The "some performance" that is lost is what? 1%? And this is Athlon64 too, do a test on your P4 to see with these timings:

 

2-2-2

 

2-3-3

 

2.5 -3-3

[All Kill3r]

Test Bed:

 

P4 3.0C at 230mhz

Abit Ai7

x800pro stock

2x256 Corsair v1.1 BH5

 

2-2-2-5

 

SiSandra Memory Bandwidth Test

INT- 5419

FLOAT- 5294

 

3DMark01 - 15873

 

2-3-3-5

 

SiSandra Memory Bandwidth Test

INT- 5244

FLOAT- 5196

 

3DMark01 - 15392

 

2-4-4-5

 

SiSandra Memory Bandwidth Test

INT- 4934

FLOAT- 5001

 

3DMark01 - 15081

 

going to 2.5-4-4-5 was identical to 2-4-4-5. Going to 3-4-4-5 or 3-4-4-8 results in no boot. My RAM cannot boot with CL3.

 

The difference from 15873 to 15081 in 3DMark01 is an average of 10fps less, this was done on gameplay settings to see realworld performance. Full quality setting and full AA/AF settings.

[Crowbar]

too many numbers...hurting my eyes...ow...

 

what's all that mean, Homer?

[All Kill3r]

now, same testbed at 225mhz, 5mhz/75mhz overall less the above tests.

 

2-2-2-5 CPC enable

 

SiSandra Memory Bandwidth Test

INT- 5975

FLOAT- 5826

 

3DMark01 - 16421

 

225mhz is the fastest i can run with CPC on with only 3.2v. Above 225 kicks me to desktop in games or benches.

 

This brings me to my conclusion. Either buy the best RAM you can, or the cheapest. There is no middle ground. The best will have a noticable difference over the cheapest. The middle will not have a noticable difference over the cheapest. With the same ram i can run 16421 with tighest timings and bios options turned on, or i can 'mimick' cheap ram and score a 15081. Noticable ? I can notice it.

 

And cheap RAM will not allow you to play with CPC options, or the extra bells and whistles in you bios that give you performance jumps, only the nicer chips can take advantage of this, and even then they need extra volts.

[Crowbar]

I've never understood all this talk about CAS latency CPC timing voltage crap.

 

I did understand that part about buying the cheapest or best RAM though. I'm going with the cheapest, and I'm not going to notice a difference.

[All Kill3r]

The "some performance" that is lost is what? 1%?

 

8% i think ? double check my math from 225mhz 2-2-2 CPC on to 230mhz 2-4-4-5 , i think 8-9% is about right... thats a noticable difference in my book, and at 5mhz slower speed, image when the ocz ddr booster gets here, i might just be able to do cpc on at 240mhz. Then that extra money i spend on bh5 will pay off huge.

 

Good choice Crowbar, there is big gains to be had with performance RAM, but it takes work and skill, with a little luck to get the gains to pay off. For gaming the value ram is just fine, i just wanted to put the numbers out there for people to see, along with Homer's numbers.

 

Hope it helps someone.