Memory Speed

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Or maybe it's not so simple...

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Or maybe it's not so simple...

Click to expand...

Memory speed isn't usually of great importance for overall performance, but be sure that if you're adding modules rather than replacing them all that the new memory has the same characteristics as the old. Buying the exact same modules if possible is a good idea.

I would take clock speed over latency, so you can overclock your processor a little better .

If your board allow to make adjustments, buy a 16GB stick and overclock to squeeze out all the performance you can get. Then sell off your old 8GB.

If just a adding, agree that it's only going to run at the slowest speeds of all sticks, so best to buy same to save money.

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory...

Click to expand...

Out of curiosity...

What is the motherboard model?

What current memory modules, make, model, speed and latency spec'?

What CPU, and is it over-clocked?

BG454 said:

...and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Click to expand...

As a broad generalisation/simplification, clock speed is usually the more significant - but not always.

If a decrease in latency was large enough, then that could outweigh a single step up in clock speed.

These test articles are quite informative...

https://www.anandtech.com/show/4503/sandy-bridge-memory-scaling-choosing-the-best-ddr3

https://www.anandtech.com/show/7364/memory-scaling-on-haswell

BG454 said:

Or maybe it's not so simple...

Click to expand...

Indeed - in the conclusion of the second article, it shows that (at least on 'Haswell' platforms) the highest clock speeds with their accompanying longer latencies, could sometimes actually lose overall performance - and the overall performance is a function of both clock speed and latency in combination (as it is obviously bound to be).

All that said and done - the difference to overall system performance is not all that much (sometimes non-existent) except perhaps where the step up in memory specification is really significant, e.g. perhaps going from 1333 to 2133+. while keeping low latency.

Small gains at best - but, it can 'still add up' in combination with other things, like CPU over-clocking, swapping from HDD to SSD, and such like.

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating.

Click to expand...

In all the time that I was building computers, memory speed was something that I never worried about. At each stage of hardware evolution, my wholesaler only ever sold one type of memory and that was that.

At one stage everyone became excited by RamBus memory, but that proved to be a passing fad.

The amount of memory (and reliability) was always far more important than the gritty details of clock speed etc. etc.

Austinian said:

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Or maybe it's not so simple...

Click to expand...

Memory speed isn't usually of great importance for overall performance, but be sure that if you're adding modules rather than replacing them all that the new memory has the same characteristics as the old. Buying the exact same modules if possible is a good idea.

Click to expand...

Thank you.

It will be a complete new set of DDR3 RAM.
I've experienced the vagaries of mixing RAM in the past.

Mark H said:

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory...

Click to expand...

Click to expand...

Mark H said:

Out of curiosity...

What is the motherboard model?

Click to expand...

Thanks for your interest,

It's a Gigabyte Z97x-UD5H-BK "Black Edition."

Mark H said:

What current memory modules, make, model, speed and latency spec'?

Click to expand...

Corsair "Vengance" 1600MHz, I'm not certain but I think they are CL9

Mark H said:

What CPU, and is it over-clocked?

Click to expand...

Intel i7 - 4790K

not overclocked (running at 4GHz)

Mark H said:

Mark H said:

...and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Click to expand...

As a broad generalisation/simplification, clock speed is usually the more significant - but not always.

If a decrease in latency was large enough, then that could outweigh a single step up in clock speed.

These test articles are quite informative...

https://www.anandtech.com/show/4503/sandy-bridge-memory-scaling-choosing-the-best-ddr3

https://www.anandtech.com/show/7364/memory-scaling-on-haswell

Mark H said:

Or maybe it's not so simple...

Click to expand...

Indeed - in the conclusion of the second article, it shows that (at least on 'Haswell' platforms) the highest clock speeds with their accompanying longer latencies, could sometimes actually lose overall performance - and the overall performance is a function of both clock speed and latency in combination (as it is obviously bound to be).

All that said and done - the difference to overall system performance is not all that much (sometimes non-existent) except perhaps where the step up in memory specification is really significant, e.g. perhaps going from 1333 to 2133+. while keeping low latency.

Small gains at best - but, it can 'still add up' in combination with other things, like CPU over-clocking, swapping from HDD to SSD, and such like.

Click to expand...

Thanks for the information, I'll study those links.

WryCuda said:

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating.

Click to expand...

In all the time that I was building computers, memory speed was something that I never worried about. At each stage of hardware evolution, my wholesaler only ever sold one type of memory and that was that.

At one stage everyone became excited by RamBus memory, but that proved to be a passing fad.

The amount of memory (and reliability) was always far more important than the gritty details of clock speed etc. etc.

Click to expand...

Thank you.

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Or maybe it's not so simple...

Click to expand...

Only the amount matters more than the speed. The only other thing is the speed compatibility.

MisterBG said:

I'm considering expanding my computer memory from 8GB to 16GB.

Click to expand...

Good idea,

BG454 said:

My M/B takes DDR3 memory and I was wondering, what is the most significant speed factor influencing DDR3 memory, the clock speed or the CL rating?

Click to expand...

Populate all your memory slots with identical memory sticks. Get the rating that is suggested for your motherboard.

BG454 said:

Or maybe it's not so simple...

Click to expand...

It's simple.

Both.

If your want to overclock, you can increase performance a fair bit but you might have to make a couple changes. Get a good cooling system - a decent tower cooler or an AIO water cooler. It's anywhere from $50 to $150. Then, consider the Vengeance Pro RAM. It's rated at 2400MHz C10.

This is a pretty good write-up on overclocking your motherboard and CPU:

Austinian said:

Memory speed isn't usually of great importance for overall performance...

Click to expand...

Exactly - modern microprocessors are very heavily cached so they don't have to access main memory all that often, relatively speaking.

Typical cache hit rates are well north of 90%, which means that even if you were able to completely eliminate all DRAM memory access time (totally impossible, but useful as a thought experiment) you'd only be able to increase overall system performance by less than 10%.

Sean Nelson said:

Austinian said:

Memory speed isn't usually of great importance for overall performance...

Click to expand...

Exactly - modern microprocessors are very heavily cached so they don't have to access main memory all that often, relatively speaking.

Typical cache hit rates are well north of 90%, which means that even if you were able to completely eliminate all DRAM memory access time (totally impossible, but useful as a thought experiment) you'd only be able to increase overall system performance by less than 10%.

Click to expand...

Assume cache access time is .2 nsec and DRAM access time is 100 nsec.

90 cache hits + 10 memory hits = 18 nsec + 1000 nsec = 1018 nsec.

100 cache hits = 20 nsec.

50X improvement!

Chris Noble said:

Sean Nelson said:

Austinian said:

Memory speed isn't usually of great importance for overall performance...

Click to expand...

Exactly - modern microprocessors are very heavily cached so they don't have to access main memory all that often, relatively speaking.

Typical cache hit rates are well north of 90%, which means that even if you were able to completely eliminate all DRAM memory access time (totally impossible, but useful as a thought experiment) you'd only be able to increase overall system performance by less than 10%.

Click to expand...

Assume cache access time is .2 nsec and DRAM access time is 100 nsec.

90 cache hits + 10 memory hits = 18 nsec + 1000 nsec = 1018 nsec.

100 cache hits = 20 nsec.

50X improvement!

Click to expand...

Yep, my logic was obviously very Fawlty.

Chris Noble said:

Sean Nelson said:

Austinian said:

Memory speed isn't usually of great importance for overall performance...

Click to expand...

Exactly - modern microprocessors are very heavily cached so they don't have to access main memory all that often, relatively speaking.

Typical cache hit rates are well north of 90%, which means that even if you were able to completely eliminate all DRAM memory access time (totally impossible, but useful as a thought experiment) you'd only be able to increase overall system performance by less than 10%.

Click to expand...

Assume cache access time is .2 nsec and DRAM access time is 100 nsec.

90 cache hits + 10 memory hits = 18 nsec + 1000 nsec = 1018 nsec.

100 cache hits = 20 nsec.

50X improvement!

Click to expand...

50X improvement for the small percentage of workloads that are able to rely solely on processor cache.

It's kind of like saying my tires have 50x more grip so my car will go 50X faster now.

I don't remember which site Tom's Hardware or Anand or somebody else who actually builds computers tested the slowest memory (DDR3) to the fastest memory and even overclocked it. The final delta between the slowest and the fastest was........2%

So just follow what manufacturer recommends and don't worry about it.

tkbslc said:

Chris Noble said:

Sean Nelson said:

Austinian said:

Memory speed isn't usually of great importance for overall performance...

Click to expand...

Exactly - modern microprocessors are very heavily cached so they don't have to access main memory all that often, relatively speaking.

Typical cache hit rates are well north of 90%, which means that even if you were able to completely eliminate all DRAM memory access time (totally impossible, but useful as a thought experiment) you'd only be able to increase overall system performance by less than 10%.

Click to expand...

Assume cache access time is .2 nsec and DRAM access time is 100 nsec.

90 cache hits + 10 memory hits = 18 nsec + 1000 nsec = 1018 nsec.

100 cache hits = 20 nsec.

50X improvement!

Click to expand...

50X improvement for the small percentage of workloads that are able to rely solely on processor cache.

It's kind of like saying my tires have 50x more grip so my car will go 50X faster now.

Click to expand...

I agree with you. Just pointing out that RAM access time dominates processor throughput (much more than processor clock rate), even with big processor caches -- there's just too much concurrent process switching, which is the main reason for hard memory faults.

SushiEater said:

I don't remember which site Tom's Hardware or Anand or somebody else who actually builds computers tested the slowest memory (DDR3) to the fastest memory and even overclocked it. The final delta between the slowest and the fastest was........2%

So just follow what manufacturer recommends and don't worry about it.

Click to expand...

Thanks, that's quite reassuring.