Memory clock (frequency) Vs. Latency (timings)

[prankstare]

From what I gather reading this great post/thread, having higher memory clocks as opposed to tighter memory timings does not always translate into better performance. In the scenario described here, CL6@1600MHz might be better than CL10@2133MHz because the former represents 3.75ns of latency in contrast to 4.68ns, which is significantly less latency.

Hence, I am aware that clock speed also affects latency, but what about bandwidth or throughput? Is memory bandwidth only tied to clock speeds, or latency as well? I thought this was more of a matter of how wide the memory bus is in relation to the frequency at which the memory is operating.

Thanks!

 

[P4-630]

@sneekypeet should be able to answer your questions.

 

[little cat]

As if the frequency is more important then on second place the Cl . Look at the mark of the RAM stick/s . It says the bandwidth e.g. PC3 14900. . In gaming with dedicated video card , system RAM with DDR3 1866 Cl9 gives results equal to DDR4 2133 Cl15 .

 

[silentbogo]

Here's a little article from crucial, which might answer some of your questions:

http://www.crucial.com/usa/en/memory-performance-speed-latency

EDIT:
Here's one more on timings:
http://www.hardwaresecrets.com/understanding-ram-timings/

 

[little cat]

Here's a little article from crucial, which might answer some of your questions:

http://www.crucial.com/usa/en/memory-performance-speed-latency

Glad You posted it . And now it is updated . I read it a long time ago

 

[prankstare]

Alright guys, let's try and not oversimplify this matter okay. Memory is something of a hot-topic especially now that CPUs seem to have hit a brick wall in technological development.

In gaming with dedicated video card , system RAM with DDR3 1866 Cl9 gives results equal to DDR4 2133 Cl15

Sorry, I'm having a hard time trying to understand your statement. Do you mean CAS9@1866MHz is as fast as CAS15@2133MHz in terms of latency, bandwidth or both?

Nevertheless, I think one of the most relevant questions anyone could make is are we really getting any faster RAM now with DDR4? And if CPUs are not really getting any faster anymore, then what are the next steps for the industry to reinvent itself and make x86 hardware attractive again if not with newer, faster and more power-efficient CPUs? The answer may rely solely on one thing, memory!

 

[cadaveca]

Alright guys, let's try and not oversimplify this matter okay. Memory is something of a hot-topic especially now that CPUs seem to have hit a brick wall in technological development.


Sorry, I'm having a hard time trying to understand your statement. Do you mean CAS9@1866MHz is as fast as CAS15@2133MHz in terms of latency, bandwidth or both?

Nevertheless, I think one of the most relevant questions anyone could make is are we really getting any faster RAM now with DDR4? And if CPUs are not really getting any faster anymore, then what are the next steps for the industry to reinvent itself and make x86 hardware attractive again if not with newer, faster and more power-efficient CPUs? The answer may rely solely on one thing, memory!

CAS Latency is but one solitary timing out of many, and in the end, is simply the first timing (Column Active Strobe) before any "action". So you can not judge memory performance by this single timing.

That's kind of what the article hosted by Crucial was trying to say, by mentioning CAS latency vs Real latency.

Ultimately, data passing through the processing chain can be much faster... M.2 SSDs now top out over 2000MB/s, memory on Skylake CPUs can hit 50,000+ MB/s, and so on. So memory isn't the sole thing that can improve performance. Designs in the complete architecture account for those delays, since data does not arrive instantly at the CPU or GPU, and as the supporting technologies truly catch up to "real-time" processing speeds, significant performance benefits can be had. But then, we need everything to be contained wqithin the same slice of silicon for that to be truly possible, and that's simply not affordable at this time, even though the physical limits of silicon tech are close at hand.

 

[prankstare]

You're overthinking things IMO. It's easier to just trust in those who's job it is to solve these problems. There's no real question of whether things will get faster or more efficient. They will. Or it will all come to a grinding halt. Which is a highly unlikely event. Any time real soon that is.

Really, I don't see the point in going off-topic all the time (as you explicitly do), but quite frankly, everytime I hear I'm overthinking I always take it as a compliment. On a more serious note, no one should leave the questioning to others, especially when you are dealing with greedy companies out there who's sole purpose is to live and survive out of other people's pockets (or in other words, to convince you to buy something you might not always necessarily want or need). Enough off-topic? Alright!

But then, we need everything to be contained wqithin the same slice of silicon for that to be truly possible, and that's simply not affordable at this time, even though the physical limits of silicon tech are close at hand.

Well, HBM/HBM2 for graphics is a good example of how things could be much improved for system RAM also, though I'm not sure why CPU vendors still insist on the same old slow RAM architecture, relying for decades on expensive on-die caches to make up for such structural bottlenecks.

Maybe, I don't know maybe in 5-10 years we may see a totally new approach for system RAM that is as prolific as AMD's HBM/HBM2 technologies are for graphics, and realize that perhaps our now almost 4 year-old 8350's, that most consider "obsolete", had in fact a lot more to offer if not for poor software implementation (true multi-threaded coding) and system RAM performance.

Just as DirectX 12 did wonders for gaming performance in Windows, maybe system RAM should be the next big thing for the x86 platform. Correct me if I'm wrong but I believe we have never been "CPU-bound" in the first place.

 

[little cat]

Alright guys, let's try and not oversimplify this matter okay. Memory is something of a hot-topic especially now that CPUs seem to have hit a brick wall in technological development.


Sorry, I'm having a hard time trying to understand your statement. Do you mean CAS9@1866MHz is as fast as CAS15@2133MHz in terms of latency, bandwidth or both?

Nevertheless, I think one of the most relevant questions anyone could make is are we really getting any faster RAM now with DDR4? And if CPUs are not really getting any faster anymore, then what are the next steps for the industry to reinvent itself and make x86 hardware attractive again if not with newer, faster and more power-efficient CPUs? The answer may rely solely on one thing, memory!

What I mean is :
Rig 1
i7 4790K
2x8 GB DDR3 1866 Cl9
GTX 980 ti

Rig 2
i7 6700K
2x8GB DDR4 2133 Cl15
GTX 980 ti

The gaming score should be equal

Tips : DDR4 3000MHz is faster and cheap

 

[cadaveca]

Well, HBM/HBM2 for graphics is a good example of how things could be much improved for system RAM also, though I'm not sure why CPU vendors still insist on the same old slow RAM architecture, relying for decades on expensive on-die caches to make up for such structural bottlenecks.

Maybe, I don't know maybe in 5-10 years we may see a totally new approach for system RAM that is as prolific as AMD's HBM/HBM2 technologies are for graphics, and realize that perhaps our now almost 4 year-old 8350's, that most consider "obsolete", had in fact a lot more to offer if not for poor software implementation (true multi-threaded coding) and system RAM performance.

Just as DirectX 12 did wonders for gaming performance in Windows, maybe system RAM should be the next big thing for the x86 platform. Correct me if I'm wrong but I believe we have never been "CPU-bound" in the first place.

Sure, HBM1/2 is a good example of how things can have higher bandwidth... but...

The cheapest HBM "system" right now costs well over $500 locally. Which means the cheapest CPU would cost just as much...

So like I said, its not affordable. In order for technology to advance in such a way, it needs to be affordable, not just faster. So we are taking "baby steps" in order to keep things affordable, and slightly faster with each iteration.

Also, when it comes to memory specifically, a big part of the reason that increasing memory speeds right now show only small performance benefits is because memory is just one link in the chain, and currently, is "fast enough" to match with all the other bits and pieces. There's no point in super-fast memory when there are other things that are "holding it back".

 

[MrGenius]

Not a problem. We've got a cool feature that allows me to delete my off topic blatherings.

Off topic? Are you...yep...I suppose you are. Nice talking to ya!

 

[prankstare]

Not a problem. We've got a cool feature that allows me to delete my off topic blatherings.

Off topic? Are you...yep...I suppose you are. Nice talking to ya!

Good to see folks like you disseminating bad ideology and then retracting.

Nice talking to you too, "Genius". Have a good one!

 

[MrGenius]

For someone as smart as you seem to think you are, you sure ask a lot of "dumb" questions and say a lot of "stupid" things...IMO. And big words don't make you smart. Especially when you misuse them.

You make up some false accusation about going off topic when I did no such thing. Then you have the audacity to take one of those "bad ideologies" I provided and run with it. As if it wasn't such a bad idea after all. Make up your mind. I retracted my statements because you are obviously not worthy of them. I was trying to be helpful. Apparently it was a lost cause. And I never said anything to you to give you any reason to start criticizing me. Yet you proceeded to do so. I find that offensive. Thus I deem you as a waste of my breath. And you're no longer welcome to my input. Or to steal my ideas or thoughts and claim them as your own.

IMO means In My Opinion. Which means you have no right to deny me of it. And shows a lack of maturity that it would be taken by you as an insult. When it was not intended as such. It was intended as a friendly suggestion. Most certainly not as any kind of insolence. But if you'd like to take it as such that's your prerogative. It just makes you look bad, not me. Say go right ahead. Doesn't hurt my feelings in the least.

BTW, those whom are offended by my screen name are those who have obvious insecurities about their own IQ. And they tend to try and mock me about it to make themselves feel better. Like it hurts my feelings or something. Please. As if. You're just making yourself look even stupider. I've even went so far as to give you an idea of what the word "genius" means to me(see my sig). Lest you insecure types get the wrong idea. Since I'm starting to get pretty tired of the same old comments about it. Get over yourselves already.

And congratulations! You're the first member here to make my ignore list. Good for you! Sorry I don't find your pranks amusing.

 

[little cat]

Guys , take it easy ! You both are good ones .

 

[prankstare]

Well, if you find my questions "dumb" and such, then why do you even bother answer them?

Seriously though, why are you even talking about IQ in the first place?

Plus, I don't recall offending anybody at any given moment, so don't be.

 

[rtwjunkie]

That's how I feel, because there are obviously whole posts missing in this thread. I can't follow it at all. There is no smooth flow of conversation.