Several Beginning Memory Questions

[Firewind]

I am trying to learn more about computers and memory as a whole. Most of my performance questsion are reference to gaming or multi-tasking (virus-scanning/gaming/chatting/office).

1) What exactly does higher memory clockspeed do for performance? For that matter, what exactly does better latency do for performance? For instance, how would a CAS3 DDR2 533 memory compare to a CAS5 DDR2 800 memory?

2) I was told that AMD processors are far more sensitive to better memory, and to consider getting CAS4 DDR2 800 to run with an AMD socket AM2 motherboard. Why is this the case?

3) Without any overclocking, do AM2 processors (such as the X2 3600+ brisbane) have any default memory needed for a 1:1 ratio like Intel processors have?

4) Is there any performance increase by running memory at a non-1:1 ratio? Like DDR2 800 on a 1066 FSB?

5) I'm told that DDR 400 939 socket motherboards outperform DDR2 800 AM2 socket motherboards. Why is this the case?

6) I often see people overclocking their memory from something like DDR2 800 CAS 4 to DDR2 950 CAS 5. Why would they be willing to make their latency worse for faster speed.

7) I have a E6700 currently overclocked on an Asus P5W DH Deluxe at 3.2 ghz. I have 2gb of Corsair 6400c4 memory. Currently the memory is running at DDR2 800 CAS4 at a 5:4 ratio with the motherboard. Should I lower it to DDR2 640 and try to get better latency?


Thank you very much for your time,

Firewind

 

[Firewind]

8) Is there a really big increase in dual-channel memory as opposed to single-channel? Do socket 754 motherboards support dual-channel?

 

[agentsmith957]

1) Higher memory clockspeeds will allow information to be read and written to your memory faster. Latencies have to do with how long the memory has to wait between each cycle. Therefore memory performance is directly related to both of these aspects. Usually, an increase in clock speed will do more for your system than a lower latency, so the CAS5 800 will perform faster than a CAS3 533 module, assuming your chipset can take advantage of DDR2 800 speeds.

2) Memory performance on the more recent AMD platforms is much better than its corresponding intel counterparts becuase their memory controller method is much better. The memory controller is integrated right on the CPU, while on the intel platform, the memory controller is integrated on the northbridge chipset, therefore adding an extra step between memory and CPU.

3) There are always specific memory speeds any platform supports at stock speeds. Usually, this is not a 1:1 ratio.

4) The ratio is not what defines the performance, the speeds define the performance. If you had memory that could run at 1066 speeds, and you could run it at 1:1, it would be faster than 800:1066. However, if you have a motheboard with high overclockability, you might be able to reach higher speeds such as DDR2 1066 with a 1600 FSB, which is a 2:3 ratio, which will be much faster that the 1:1 1066:1066.

5) This is only the case under certain configurations with certain benchmarks, however there is some truth to that statement. The reason the memory increase does not make all that much of a performance difference, is due to the may the platform is built. The current AM2/939 processors are not particularly starved for memory bandwidth. The reason for the move to AM2 is for future upgradability and higher FSB's that the arcitechture will allow.

6) You said it yourself, its faster. The faster speeds make up for the higher latencies.

7) No.

 

[nobly]

I'll try to answer these questions together since they're related:

1) What exactly does higher memory clockspeed do for performance? For that matter, what exactly does better latency do for performance? For instance, how would a CAS3 DDR2 533 memory compare to a CAS5 DDR2 800 memory?

6) I often see people overclocking their memory from something like DDR2 800 CAS 4 to DDR2 950 CAS 5. Why would they be willing to make their latency worse for faster speed.

Higher clockspeed translates into a higher CPU overclock. Each CPU has a multiplier, which mulitplies the FSB speed to get the CPU clockspeed. So if your FSB is 400Mhz w/ a CPU multiplier of 5, then you have a CPU clocked at 2Ghz.
Latency is another matter in itself. As with all circuits, its all about timing. CAS4 means 4 clock cycles. Now this varys because it depends on the speed of the FSB. A FSB of 200Mhz, 1 clock cycle is 5ns. At 400Mhz, 1 clock cycle is 2.5ns. So you can see that a CAS4 at 200Mhz is the same time as a CAS8 at 400Mhz. The key is to lower your overall CAS timing. So, taking your example, CAS4 at 800Mhz is 5ns, and CAS5 at 950Mhz is 5ns. So its the SAME timing. BUT with 950Mhz, your CPU overclock is higher than at 800Mhz.

2) I was told that AMD processors are far more sensitive to better memory, and to consider getting CAS4 DDR2 800 to run with an AMD socket AM2 motherboard. Why is this the case?

5) I'm told that DDR 400 939 socket motherboards outperform DDR2 800 AM2 socket motherboards. Why is this the case?

The integrated memory controller vs the northbridge memory controller is a matter of much debate. AMD processors are more sensitive to memory timings because its integrated onto the CPU core. You can see in tomshardware's benches that you need DDR2 800 to outperform the old 939 DDR 400Mhz RAM. Perhaps its just because they need to refine it etc.
Intel CPUs are less sensitive to the memory because the memory controller is on the northbridge, a separate chip. True, this causes more latency that Intel counters w/ a larger CPU cache, but its also alot easier with memory upgrade paths, because the team only has to worry about the northbridge chip, and not the CPU as a whole.
This is why AMD has to release new CPUs when new memory standards come out, and intel just has to release new chipsets.

Sorry can't really comment on 3,4, or 7 cuz I'm not entirely sure myself.

Oh yeah, usually its a boost of less than 10% for dual channel memory. And socket 754 is SINGLE channel.

 

[Mondoman]

1) What exactly does higher memory clockspeed do for performance? For that matter, what exactly does better latency do for performance?

This post should help: http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=1233230#1233230

2) I was told that AMD processors are far more sensitive to better memory, and to consider getting CAS4 DDR2 800 to run with an AMD socket AM2 motherboard.

Recent AMD 64 CPUs have the memory controllers built into the CPU, reducing controller <-> CPU latency and increasing controller <-> CPU speed (current Intel CPUs have the memory controllers as part of the northbridge chip; the northbridge <-> CPU FSB thus limits overall transfer speed and increases total latency). Thus, on such AMD systems with current and near-future memory, the memory/memory bus is the speed and latency bottleneck; running memory at faster speeds and lower latency settings should directly improve performance.

3) Without any overclocking, do AM2 processors (such as the X2 3600+ brisbane) have any default memory needed for a 1:1 ratio like Intel processors have?

I don't quite understand this question.

4) Is there any performance increase by running memory at a non-1:1 ratio? Like DDR2 800 on a 1066 FSB?

Assuming you mean that the memory is running *faster* than the FSB, normally yes, with a few rare exceptions.

5) I'm told that DDR 400 939 socket motherboards outperform DDR2 800 AM2 socket motherboards. Why is this the case?

Not true in general. It depends on the memory used and board design.

6) I often see people overclocking their memory from something like DDR2 800 CAS 4 to DDR2 950 CAS 5. Why would they be willing to make their latency worse for faster speed.

Depending on the apps you are running, sometimes throughput is more important than latency. In any case, 4 clock ticks at 400MHz (DDR2-800) is almost the same amount of time as 5 clock ticks at 475MHz (DDR2-950), so no big latency difference.

7) I have a E6700 currently overclocked on an Asus P5W DH Deluxe at 3.2 ghz. I have 2gb of Corsair 6400c4 memory. Currently the memory is running at DDR2 800 CAS4 at a 5:4 ratio with the motherboard. Should I lower it to DDR2 640 and try to get better latency?

Try it and see how it affects your apps (there's no general answer).

 

[Firewind]

Thank you very much for the answers, this helps me out a lot.