Frequency or Latency?

[EpIckFa1LJoN]

Short Question: What is more important for gaming, frequency or true latency?

So I know I have looked through these threads enough to know by now but I am still kind of confused about how this works. I know dividing the CAS latency by the frequency gives you the relative speed of memory (true latency) but I have been doing the math since I plan on upgrading to z370 soon and I want the fastest RAM I can get, and it doesn't make too much sense.


I have looked at the Corsair Dominator Platinum 16GB (2x8) kits and doing the aforementioned math. Here are some of the kits I am looking at:

True Latency Formula (more or less)- (CAS Latency/Frequency(GHz))x2 = True Latency (ns)

Current: 3200MHz C16 = 10ns

Seemingly Fastest: 2400MHz C10 = 8.33ns

Others:
3200MHz C14 = 8.75ns
3466MHz C16 = 9.23ns
3733MHz C17 = 9.11ns
3866MHz C18 = 9.31ns



Really my question is what is more important for gaming? I know we're talking a fraction of a second more, but if I'm buying new RAM anyways, I might as well get the one that is going to perform the best.

So what would be faster in a gaming situation? Something like the lower latency with a decent frequency? (Like the 2400 C10 or 3200 C14) or something with a higher frequency and a medium latency? (Like the 3733 C17) Or is it really all about True Latency?

 

[Pinhedd]

Short Question: What is more important for gaming, frequency or true latency?

So I know I have looked through these threads enough to know by now but I am still kind of confused about how this works. I know dividing the CAS latency by the frequency gives you the relative speed of memory (true latency) but I have been doing the math since I plan on upgrading to z370 soon and I want the fastest RAM I can get, and it doesn't make too much sense.


I have looked at the Corsair Dominator Platinum 16GB (2x8) kits and doing the aforementioned math. Here are some of the kits I am looking at:

True Latency Formula (more or less)- (CAS Latency/Frequency(GHz))x2 = True Latency (ns)

Current: 3200MHz C16 = 10ns

Seemingly Fastest: 2400MHz C10 = 8.33ns

Others:
3200MHz C14 = 8.75ns
3466MHz C16 = 9.23ns
3733MHz C17 = 9.11ns
3866MHz C18 = 9.31ns



Really my question is what is more important for gaming? I know we're talking a fraction of a second more, but if I'm buying new RAM anyways, I might as well get the one that is going to perform the best.

So what would be faster in a gaming situation? Something like the lower latency with a decent frequency? (Like the 2400 C10 or 3200 C14) or something with a higher frequency and a medium latency? (Like the 3733 C17) Or is it really all about True Latency?

You're confused about how it works because... that's not how it works. Dividing the transfer rate (the people perpetuating this nonsense always call it frequency, which is an immediate tell) by the read latency to get some fictional "true latency" is something that some amateur twat made up absent any technical, empirical, or scientific foundation.

Anyway... Frequency trumps latency pretty much across the board... and it should as long as the memory controller can keep the DRAM busses busy. Past memory controllers have struggled to keep the busses busy because there really wasn't that big of a demand. Now that microprocessors are pushing 6/8 cores on mid level desktops along with bandwidth hungry IGPs, serious effort has been put into memory controllers that was largely absent in the later half of the 2000s and early half of the 2010s.

So... frequency, frequency, frequency. With that said, overclocking a memory controller places additional thermal and electrical stressors on the package just the same as overclocking the ISA cores. Tightening timings does not.

DDR4-2400 should be plug and play with no drama, even in large capacities. DDR4-4000 on the other hand is likely going to require some TLC before you find that sweet spot where the OS loads, nothing crashes, and stress tests don't spit out errors.

 

[EpIckFa1LJoN]

Short Question: What is more important for gaming, frequency or true latency?

So I know I have looked through these threads enough to know by now but I am still kind of confused about how this works. I know dividing the CAS latency by the frequency gives you the relative speed of memory (true latency) but I have been doing the math since I plan on upgrading to z370 soon and I want the fastest RAM I can get, and it doesn't make too much sense.


I have looked at the Corsair Dominator Platinum 16GB (2x8) kits and doing the aforementioned math. Here are some of the kits I am looking at:

True Latency Formula (more or less)- (CAS Latency/Frequency(GHz))x2 = True Latency (ns)

Current: 3200MHz C16 = 10ns

Seemingly Fastest: 2400MHz C10 = 8.33ns

Others:
3200MHz C14 = 8.75ns
3466MHz C16 = 9.23ns
3733MHz C17 = 9.11ns
3866MHz C18 = 9.31ns



Really my question is what is more important for gaming? I know we're talking a fraction of a second more, but if I'm buying new RAM anyways, I might as well get the one that is going to perform the best.

So what would be faster in a gaming situation? Something like the lower latency with a decent frequency? (Like the 2400 C10 or 3200 C14) or something with a higher frequency and a medium latency? (Like the 3733 C17) Or is it really all about True Latency?

You're confused about how it works because... that's not how it works. Dividing the transfer rate (the people perpetuating this nonsense always call it frequency, which is an immediate tell) by the read latency to get some fictional "true latency" is something that some amateur twat made up absent any technical, empirical, or scientific foundation.

Anyway... Frequency trumps latency pretty much across the board... and it should as long as the memory controller can keep the DRAM busses busy. Past memory controllers have struggled to keep the busses busy because there really wasn't that big of a demand. Now that microprocessors are pushing 6/8 cores on mid level desktops along with bandwidth hungry IGPs, serious effort has been put into memory controllers that was largely absent in the later half of the 2000s and early half of the 2010s.

So... frequency, frequency, frequency. With that said, overclocking a memory controller places additional thermal and electrical stressors on the package just the same as overclocking the ISA cores. Tightening timings does not.

DDR4-2400 should be plug and play with no drama, even in large capacities. DDR4-4000 on the other hand is likely going to require some TLC before you find that sweet spot where the OS loads, nothing crashes, and stress tests don't spit out errors.

If that's true, i've got some butts to kick . However, that still doesn't explain how 2400 C10 memory is over twice as expensive as the basic DDR4 Kit (2400 C14) (since marketing gimmicks are actually geared towards pushing higher FREQUENCY RAM despite it being theoretically slower due to the high latency; 4000 C19 with a timing of 9.5ns making it barely faster than 3200 C16 and being twice as expensive). Or how the second supposedly fastest (3200 C14) costs 20% more than my current RAM (3200 C16). There has to be some sort of tradeoff. The way I understand it, again in fractions of a second, is that the lower the latency the more processes can be made in the same amount of time. If latency weren't a factor why advertise it at all? Also the information I came across comes from Crucial's own website.

Don't mind my scrutiny, you're literally the first person to say that in a sea of people saying basically what I said above.

It just makes sense that higher frequencies are offset by the higher latencies. Unfortunately I can't just buy two or three kits and do the legwork and see what actually works. We're talking about $1,000 in RAM kits to test out a theory. I might just go with the 3200 C14. That way if it really is frequency it won't be any slower than what I have now, but if latency does come into play it should be a little faster than what I have now. The problem with that is that it is significantly more expensive than any other kit except 4000 C19, at almost $300 for a 16GB kit.

 

[EpIckFa1LJoN]

You know... now that I'm looking at it again... there will be a fraction of a second (1-2 nanoseconds) difference in any RAM....

Put that way I might as well get the RGB ram and just get a 32GB kit of 2400 C14 and call it a day. That sound about right?

 

[Pinhedd]

Short Question: What is more important for gaming, frequency or true latency?

So I know I have looked through these threads enough to know by now but I am still kind of confused about how this works. I know dividing the CAS latency by the frequency gives you the relative speed of memory (true latency) but I have been doing the math since I plan on upgrading to z370 soon and I want the fastest RAM I can get, and it doesn't make too much sense.


I have looked at the Corsair Dominator Platinum 16GB (2x8) kits and doing the aforementioned math. Here are some of the kits I am looking at:

True Latency Formula (more or less)- (CAS Latency/Frequency(GHz))x2 = True Latency (ns)

Current: 3200MHz C16 = 10ns

Seemingly Fastest: 2400MHz C10 = 8.33ns

Others:
3200MHz C14 = 8.75ns
3466MHz C16 = 9.23ns
3733MHz C17 = 9.11ns
3866MHz C18 = 9.31ns



Really my question is what is more important for gaming? I know we're talking a fraction of a second more, but if I'm buying new RAM anyways, I might as well get the one that is going to perform the best.

So what would be faster in a gaming situation? Something like the lower latency with a decent frequency? (Like the 2400 C10 or 3200 C14) or something with a higher frequency and a medium latency? (Like the 3733 C17) Or is it really all about True Latency?

You're confused about how it works because... that's not how it works. Dividing the transfer rate (the people perpetuating this nonsense always call it frequency, which is an immediate tell) by the read latency to get some fictional "true latency" is something that some amateur twat made up absent any technical, empirical, or scientific foundation.

Anyway... Frequency trumps latency pretty much across the board... and it should as long as the memory controller can keep the DRAM busses busy. Past memory controllers have struggled to keep the busses busy because there really wasn't that big of a demand. Now that microprocessors are pushing 6/8 cores on mid level desktops along with bandwidth hungry IGPs, serious effort has been put into memory controllers that was largely absent in the later half of the 2000s and early half of the 2010s.

So... frequency, frequency, frequency. With that said, overclocking a memory controller places additional thermal and electrical stressors on the package just the same as overclocking the ISA cores. Tightening timings does not.

DDR4-2400 should be plug and play with no drama, even in large capacities. DDR4-4000 on the other hand is likely going to require some TLC before you find that sweet spot where the OS loads, nothing crashes, and stress tests don't spit out errors.

If that's true, i've got some butts to kick . However, that still doesn't explain how 2400 C10 memory is over twice as expensive as the basic DDR4 Kit (2400 C14) (since marketing gimmicks are actually geared towards pushing higher FREQUENCY RAM despite it being theoretically slower due to the high latency; 4000 C19 with a timing of 9.5ns making it barely faster than 3200 C16 and being twice as expensive). Or how the second supposedly fastest (3200 C14) costs 20% more than my current RAM (3200 C16). There has to be some sort of tradeoff. The way I understand it, again in fractions of a second, is that the lower the latency the more processes can be made in the same amount of time. If latency weren't a factor why advertise it at all? Also the information I came across comes from Crucial's own website.

Don't mind my scrutiny, you're literally the first person to say that in a sea of people saying basically what I said above.

It just makes sense that higher frequencies are offset by the higher latencies. Unfortunately I can't just buy two or three kits and do the legwork and see what actually works. We're talking about $1,000 in RAM kits to test out a theory. I might just go with the 3200 C14. That way if it really is frequency it won't be any slower than what I have now, but if latency does come into play it should be a little faster than what I have now. The problem with that is that it is significantly more expensive than any other kit except 4000 C19, at almost $300 for a 16GB kit.

Those price differences are primarily marketing. If someone can be convinced to spend more than twice the price of DDR4-2400/CL14 on DDR4-2400/CL10 only a fool would choose to not provide such a product.

Tight timings are a nice thing to have. On many older chipsets the memory controllers were not sophisticated enough to schedule commands far into the future, so added bandwidth was hard to utilise even when there was a demand for it. Many DDR2 and early DDR3 controllers were also incredibly hostile to fast memory and could easily burn out if pushed too hard. In these cases, a slight tightening of timings was much easier to stomach than playing around with a million settings trying to get a finicky batch of wicked fast memory to place nice.

As I have said, this is not the case anymore; modern memory controllers are substantially more capable than those used even a few years ago and the introduction of high core count packages, PCI-E SSDs, and powerful IGPs have greatly increased the demand for memory bandwidth.

Even if a 1-2 nanosecond penalty is incurred, this is insignificant and meaningless. There's already a substantial round-trip penalty that is incurred every time an L3 cache miss occurs anyway. This is a problem that us computer engineers (my profession) worked on a long time ago. Prefetching, speculative execution, out-of-order execution, simultaneous multithreading, DRAM bursting, DRAM pipelining, and many other things serve to not only mitigate this but effectively use latency as a tool.

As for what's best for your needs? I run 32GiB of DDR4-2400/CL14

 

[EpIckFa1LJoN]

Short Question: What is more important for gaming, frequency or true latency?

So I know I have looked through these threads enough to know by now but I am still kind of confused about how this works. I know dividing the CAS latency by the frequency gives you the relative speed of memory (true latency) but I have been doing the math since I plan on upgrading to z370 soon and I want the fastest RAM I can get, and it doesn't make too much sense.


I have looked at the Corsair Dominator Platinum 16GB (2x8) kits and doing the aforementioned math. Here are some of the kits I am looking at:

True Latency Formula (more or less)- (CAS Latency/Frequency(GHz))x2 = True Latency (ns)

Current: 3200MHz C16 = 10ns

Seemingly Fastest: 2400MHz C10 = 8.33ns

Others:
3200MHz C14 = 8.75ns
3466MHz C16 = 9.23ns
3733MHz C17 = 9.11ns
3866MHz C18 = 9.31ns



Really my question is what is more important for gaming? I know we're talking a fraction of a second more, but if I'm buying new RAM anyways, I might as well get the one that is going to perform the best.

So what would be faster in a gaming situation? Something like the lower latency with a decent frequency? (Like the 2400 C10 or 3200 C14) or something with a higher frequency and a medium latency? (Like the 3733 C17) Or is it really all about True Latency?

You're confused about how it works because... that's not how it works. Dividing the transfer rate (the people perpetuating this nonsense always call it frequency, which is an immediate tell) by the read latency to get some fictional "true latency" is something that some amateur twat made up absent any technical, empirical, or scientific foundation.

Anyway... Frequency trumps latency pretty much across the board... and it should as long as the memory controller can keep the DRAM busses busy. Past memory controllers have struggled to keep the busses busy because there really wasn't that big of a demand. Now that microprocessors are pushing 6/8 cores on mid level desktops along with bandwidth hungry IGPs, serious effort has been put into memory controllers that was largely absent in the later half of the 2000s and early half of the 2010s.

So... frequency, frequency, frequency. With that said, overclocking a memory controller places additional thermal and electrical stressors on the package just the same as overclocking the ISA cores. Tightening timings does not.

DDR4-2400 should be plug and play with no drama, even in large capacities. DDR4-4000 on the other hand is likely going to require some TLC before you find that sweet spot where the OS loads, nothing crashes, and stress tests don't spit out errors.

If that's true, i've got some butts to kick . However, that still doesn't explain how 2400 C10 memory is over twice as expensive as the basic DDR4 Kit (2400 C14) (since marketing gimmicks are actually geared towards pushing higher FREQUENCY RAM despite it being theoretically slower due to the high latency; 4000 C19 with a timing of 9.5ns making it barely faster than 3200 C16 and being twice as expensive). Or how the second supposedly fastest (3200 C14) costs 20% more than my current RAM (3200 C16). There has to be some sort of tradeoff. The way I understand it, again in fractions of a second, is that the lower the latency the more processes can be made in the same amount of time. If latency weren't a factor why advertise it at all? Also the information I came across comes from Crucial's own website.

Don't mind my scrutiny, you're literally the first person to say that in a sea of people saying basically what I said above.

It just makes sense that higher frequencies are offset by the higher latencies. Unfortunately I can't just buy two or three kits and do the legwork and see what actually works. We're talking about $1,000 in RAM kits to test out a theory. I might just go with the 3200 C14. That way if it really is frequency it won't be any slower than what I have now, but if latency does come into play it should be a little faster than what I have now. The problem with that is that it is significantly more expensive than any other kit except 4000 C19, at almost $300 for a 16GB kit.

Those price differences are primarily marketing. If someone can be convinced to spend more than twice the price of DDR4-2400/CL14 on DDR4-2400/CL10 only a fool would choose to not provide such a product.

Tight timings are a nice thing to have. On many older chipsets the memory controllers were not sophisticated enough to schedule commands far into the future, so added bandwidth was hard to utilise even when there was a demand for it. Many DDR2 and early DDR3 controllers were also incredibly hostile to fast memory and could easily burn out if pushed too hard. In these cases, a slight tightening of timings was much easier to stomach than playing around with a million settings trying to get a finicky batch of wicked fast memory to place nice.

As I have said, this is not the case anymore; modern memory controllers are substantially more capable than those used even a few years ago and the introduction of high core count packages, PCI-E SSDs, and powerful IGPs have greatly increased the demand for memory bandwidth.

Even if a 1-2 nanosecond penalty is incurred, this is insignificant and meaningless. There's already a substantial round-trip penalty that is incurred every time an L3 cache miss occurs anyway. This is a problem that us computer engineers (my profession) worked on a long time ago. Prefetching, speculative execution, out-of-order execution, simultaneous multithreading, DRAM bursting, DRAM pipelining, and many other things serve to not only mitigate this but effectively use latency as a tool.

As for what's best for your needs? I run 32GiB of DDR4-2400/CL14

Yeah that's what I am thinking. I don't really need 32GB since all I do is game, and 16GB is more than enough for that. And the difference is so small it may be about a 1-2 fps difference from 2133 C15, and 2400 C10.