CAS Latency?

[MisterRaven]

What exactly is CAS Latency, and what does it affect?

Also, would I suffer much of a performance hit if I chose to use Kingston PC3200 Value Select memory instead of Kingston PC3200 HyperX Memory?

Here are my specs:

Athlon XP 1900+ (soon to be replace with 2500+ Barton)
Shuttle AK32A Motherboard, with VIA KT266 (Soon to be rplaced by Albatron KX18D Pro II, with nForce2)
512MB Premium PC2100 (soon to be replaced by one of the above types of RAM)
Radeon 9800np
Antec 400W PSU
Lite-On 52x32x52
120GB Western Digital Hard Drive with 8MB of cache

Thanks in advance.

 

[dmcowen674]

Originally posted by: MisterRaven
What exactly is CAS Latency, and what does it affect?

Also, would I suffer much of a performance hit if I chose to use Kingston PC3200 Value Select memory instead of Kingston PC3200 HyperX Memory?

Here are my specs:

Athlon XP 1900+ (soon to be replace with 2500+ Barton)
Shuttle AK32A Motherboard, with VIA KT266 (Soon to be rplaced by Albatron KX18D Pro II, with nForce2)
512MB Premium PC2100 (soon to be replaced by one of the above types of RAM)
Radeon 9800np
Antec 400W PSU
Lite-On 52x32x52
120GB Western Digital Hard Drive with 8MB of cache

Thanks in advance.

Here is an analogy for you:

Think of CAS as the delay it takes to get through an intersection. If you have a traffic Cop you may wait 2.5 minutes to get through the intersection while if it had a traffic light you could get through in 2 minutes. So that would be an example of CAS 2.0 Vs CAS 2.5 or 3.0

You really won't notice the difference however unless you are running something extremely dependent on fastest Memory traffic time possible though.

If you are just running Business Apps like Word or Excel for example you would not notice any difference but if you are running SETI@home on the machine it would be the difference in the amount of time it takes the machine to crunch through data. At CAS 2.5 a 1.4 Gig Athlon may take 6 hrs to do a Work Unit while at CAS 2.0 it would take 4.5 hrs.

Enjoy

 

[Peter]

The difference isn't that big. For one, it's only in the main RAM, caches are not affected. Particularly on AMD machines where intelligent caching does not let RAM latencies do too much damages, there will be little difference. P4 relies more on fast RAM, but it'll still never be a 33 percent speed loss from CL2 to CL3.

This is simply because the CAS Latency occurs only once per burst transfer, meaning that one extra CL cycle slows a typical DDR burst down from 12+4 to 13+4 cycles (total addressing latency plus eight half-cycles for data transfer)
So that's actually 17 over 16, not 3 over 2. That's a mere six percent. Best case, only occurring if the CPU doesn't cache anything ever - which hardly is the case in real life.

Actual real life performance gain is hardly visible in benchmarking, let alone in actual applications - except for the rare few that do eat through HUGE datasets without looking back. Then you'll get some two to five percent.

 

[MisterRaven]

Thanks a bunch!

So you guys are saying that the lower the CAS Latency, the better, and that there won't be such a huge difference with Athlon XP systems?

This is primarily a gaming machine I'm using, by the way.

For about $100, I could get 512MB of HyperX memory, whereas I could get 1GB (2 512MB sticks) of Dual-channel value select for $150. . . So if the lower CAS latency of the HyperX won't make such a huge difference, I'll go for the value select.

Am I right or wrong? Please correct me if I'm still not understanding this completely.

Thanks!

 

[Peter]

Yes and Yes. Particularly if that's going to be a dual channel RAM machine, you'll be MUCH better off with two CL3 sticks than with one CL2.

 

[MisterRaven]

Originally posted by: Peter
Yes and Yes. Particularly if that's going to be a dual channel RAM machine, you'll be MUCH better off with two CL3 sticks than with one CL2.

Well, I'll go with the 1GB of dual-channel value select RAM then.

So I would actually be better off to use two dual-channel CL3 sticks rather than one single-channel CL2 stick? Why is that? I'm certainly not complaining, I'm just curious.

Another Question: I've also seen CAS Latency ratings displayed in other ways (rather than just "2.5" and so on). Ways like "2-3-2-6 1T" and "2-2-2-6 1T." What do these mean?

Thanks for being patient with me, by the way.

 

[Peter]

(1) Dual channel gives you extra bandwidth. Continuing my example from above, dual channel CL3 gets you sixteen datasets (128 Bytes) in 17 cycles, while single channel CL2 gets you eight (64 Bytes) in 16.

Would you rather like your 128 bytes in 17 or in 32 cycles?

Those "long" specifications also tell you the less relevant other timing values of SDRAM. There are lots more, that specification telling you them in descending order of relevance.

 

[MisterRaven]

Okay, I think I understand it now.

Thank you, oh-so-wise-Peter (and dmcowen674 too. . . but mostly Peter), you are among the hardest of core.

 

[dmcowen674]

Originally posted by: MisterRaven
Okay, I think I understand it now.

Thank you, oh-so-wise-Peter (and dmcowen674 too. . . but mostly Peter), you are among the hardest of core.

Sometimes a Motherboard doesn't get along with two Memory chips. System builders used to test pairs of chips together because of this. Getting all this bus data to cooperate at these blazing speeds is part science and part luck.

I used to design Computer Hardware back in the slower bus days and was able to see the binary data riding on the bus, it has got to be hard to measure and see anything at these speeds.

Good luck.

 

[MisterRaven]

Originally posted by: dmcowen674

Originally posted by: MisterRaven
Okay, I think I understand it now.

Thank you, oh-so-wise-Peter (and dmcowen674 too. . . but mostly Peter), you are among the hardest of core.

Sometimes a Motherboard doesn't get along with two Memory chips. System builders used to test pairs of chips together because of this. Getting all this bus data to cooperate at these blazing speeds is part science and part luck.

I used to design Computer Hardware back in the slower bus days and was able to see the binary data riding on the bus, it has got to be hard to measure and see anything at these speeds.

Good luck.

Thanks! From what I hear, the motherboard I'll be getting (Albatron KX18D Pro II) is quite stable, so I should okay.

Thanks for being patient with my questions!

 

[cockeyed]

I'll be setting up a new 865PE 2 channel mobe next week and was thinking the same question. I read a test report some time ago about the real world difference between CLK2 to CLK3 memory. It was minimal and IMO only worth it if you had to have the best and wanted bragging rights. I'll be using CLK3 Kingston value ram (dual pack 512mb) for my mobo. My thanks also to the other posters for their excellent replies; knowledge is a wonderful thing!