DDR4 2666MHz cas 15 or 2800MHz cas 16?

S

sblantipodi

2[H]4U
Joined
Aug 29, 2010
Messages
3,765
As title.
I should choose between the vengeance lpx 2666MHz cas 15 and the vengeance lpx 2800MHz cas 16.

IMHO the 2666mhz are faster and cost 20% less. Am I wrong?
 
sblantipodi said:
IMHO the 2666mhz are faster and cost 20% less. Am I wrong?
Click to expand...

The speed difference won't be noticeable, however, what makes you think the 2666 would be faster? You're just thinking it has lower latency, so it should be faster! Going by that logic a DDR 400 RAM of CL2 should be faster than anything higher rated than it.

RAM becomes faster as speed increases, but as speed increases, it's expected for the latency to increase as well. That's the rule of the thumb.
 
X72 said:
The speed difference won't be noticeable, however, what makes you think the 2666 would be faster? You're just thinking it has lower latency, so it should be faster! Going by that logic a DDR 400 RAM of CL2 should be faster than anything higher rated than it.

RAM becomes faster as speed increases, but as speed increases, it's expected for the latency to increase as well. That's the rule of the thumb.
Click to expand...

Frequency divided cas is a good meter when comparing the memory of the same technology.

So, talking about ddr4
2666/15 is greater than 2800/16.
 
My reading implies 2666 CAS15 is definitely the better choice, at least now.
From comparison of benchmarks to date speed increases over the 21xx memory are minor, between 2666 and 2800 imperceptible. With the CAS difference I see no reason to spend more. From a cheapskates perspective 21xx or 24xx, depending on price, would do equally well.

The only "question" (or doubtful validity) is Sandy Bridge was introduced to the tune there was no reason to purchase faster memory than 1333, which became 1600. Current DDR3 recommendations are between 1866 and 2400 with mixed CAS opinions.

A common statement is Intel favor speed over low CAS, AMD favors low CAS over pure speed, except for APUs.

Again, this is a personal, otherwise unsupported impression from reading articles. And as usual, if you are benching for record, faster is better. Probably.
 
Last edited:
CAS 16 is 6% higher latency than CAS15.
But 2800MHz is 5% faster than 2666MHz.

The CAS rating is the number of cycles it takes to perform the Column Address Strobe.
5% more cycles pass in the same time frame with the 2800MHz ram so the 6% deficiency in latency is reduced to 1%.
And the 2800MHz ram is 5% higher bandwidth.
So its likely that the 2800MHz ram will perform better in memory intense operations.

But you probably wont be able to tell the difference in games.
 
I think the guys at Hexus and TweakTown did glancing reviews of Corsair DDR4 RAM that can give you a bit more perspective. It seems to me the sweet spot is 2800 with CAS 15 with a slight voltage bump. I personally elected for 2666 and plan on attempting to get 2800 at CAS 15 out of it with 1.3v. I will be building Friday and testing Saturday. If you can wait that long, my results should at least be good enough to base a purchase on.
 
LAiN_OfTheWired said:
I think the guys at Hexus and TweakTown did glancing reviews of Corsair DDR4 RAM that can give you a bit more perspective. It seems to me the sweet spot is 2800 with CAS 15 with a slight voltage bump. I personally elected for 2666 and plan on attempting to get 2800 at CAS 15 out of it with 1.3v. I will be building Friday and testing Saturday. If you can wait that long, my results should at least be good enough to base a purchase on.
Click to expand...

keep us posted :)
 
sblantipodi said:
Frequency divided cas is a good meter when comparing the memory of the same technology.

So, talking about ddr4
2666/15 is greater than 2800/16.
Click to expand...

Because it's easy to make up things out of ass rather than understanding how they work. Why do you think there is faster memory available? Because it performs worse?

wtourist said:
My reading implies 2666 CAS15 is definitely the better choice, at least now.
Click to expand...

It may be the "better choice" but it's not the "faster choice".
 
X72 said:
Because it's easy to make up things out of ass rather than understanding how they work. Why do you think there is faster memory available? Because it performs worse?
It may be the "better choice" but it's not the "faster choice".
Click to expand...

I'll wait your technical answer. All the answer on this thread was not convincing.
140MHz are not enough to justify a 1 cas latency difference.

there are dozens of benchmarks on ddr3 that shows that memory with a smaller frequency with smaller latency are faster than memory with faster frequency and greater latency.
I think that this is the case but only real/syntethic benchmark could give us the truth.
 
another things that I don't understand is why corsair sets SPD latency, lower than tested latency...

they set more aggressive latency in SPD but they test ram and more conservative timings.
what is the sense of this?

The 2800MHz module for example has:
SPD Latency 15-15-15-36
Tested Latency 16-18-18-36

what is the sense?
 
drescherjm said:
Nenu explained that 2666/ cas 15 is slightly faster than 2800/ cas 16. Was he wrong?
Click to expand...

Sorry if I didnt make it clear, the edge goes to the 2800MHz.
The 2666MHz ram has 1% lower latency but has 5% less bandwidth.
But in this case, the negligible difference in performance is not worth a 20% price hike for the faster memory.
 
sblantipodi said:
another things that I don't understand is why corsair sets SPD latency, lower than tested latency...

they set more aggressive latency in SPD but they test ram and more conservative timings.
what is the sense of this?

The 2800MHz module for example has:
SPD Latency 15-15-15-36
Tested Latency 16-18-18-36

what is the sense?
Click to expand...

Possibly because there is no performance difference?
By quoting the slower speeds achieving the same result or close, they leave it to your imagination that it will be faster with tighter timings.
Perhaps :)

Or maybe its a typo, or a testing error.
 
drescherjm said:
Hmm, sorry. It looks like you made it clear. I have to pay more attention while reading.
Click to expand...

I don't agree with that assertion.
Bandwidth without latency means nothing.

Bandwidth is the amount of work you can get in a fixed period of time, latency is the amount of time that is required to process this job.

When you think at bandwidth you should think at the diameter of a flow. A tunnel.
Suppose that you need to carry 100 big things from A to B.
Once a thing is carried to B, a person work on it to store the thing in a garage.

Between A and B there is a tunnel.
A bigger tunnel permits to carry more things from A to B at the same time.

But what happen if at the checkpoint there is no one to store the thing in the garage?
What happen if you can carry so many things but you have too few people that stores things in the garage?

Frequency and timings are strongly related and there is no absolute way to measure performance.
Sometimes frequency can help, sometimes latency can help. It depends on the workloads, on the os, on the architecture.

So all this counts has no matemathical meaning.
All suppositions that can only be proved by synthetic and real world benchmark.
 
Latency in this respect is a measure of individual times involved in "setting up" a memory transfer.
It is not a measure of overall latency.

This is why these latency timings dont have a marked impact on performance.
 
Nenu said:
Latency in this respect is a measure of individual times involved in "setting up" a memory transfer.
It is not a measure of overall latency.

This is why these latency timings dont have a marked impact on performance.
Click to expand...

Probably you don't know what you are saying.
At same frequency, Ddr3 with cas 9 performs quite equals to ddr4 with cas 15.

Ddr4 has way more bandwidth that ddr3 due to its architecture but due to latency difference they performs quite the same at same frequency
 
sblantipodi said:
Probably you don't know what you are saying.
At same frequency, Ddr3 with cas 9 performs quite equals to ddr4 with cas 15.

Ddr4 has way more bandwidth that ddr3 due to its architecture but due to latency difference they performs quite the same at same frequency
Click to expand...
Um ok.

This is about how changes in latency and speed can impact the performance of different DDR4 modules.
You are now cross comparing with DDR3.
How does this help you?
 
JaiWebb said:
My vote goes to DDR4 3000MHz at CAS 15 even with the price premium.
Click to expand...

LOL, good ol newegg price gouging. That kit was selling for $329.99 not long ago.

I went with two separate 2666 kits (one g skill cl15, the other corsair lpx cl16). I did not feel like breaking the bank on ram, knowing much better stuff will be released later on.
 
Nenu said:
Um ok.

This is about how changes in latency and speed can impact the performance of different DDR4 modules.
You are now cross comparing with DDR3.
How does this help you?
Click to expand...

I'm not here to convince anyone.
Buy the ultra gigahertz ram and be happy :)
 
drescherjm said:
Nenu explained that 2666/ cas 15 is slightly faster than 2800/ cas 16. Was he wrong?
Click to expand...

No he didn't. He said the other way around.

sblantipodi said:
I'll wait your technical answer. All the answer on this thread was not convincing.
140MHz are not enough to justify a 1 cas latency difference.

there are dozens of benchmarks on ddr3 that shows that memory with a smaller frequency with smaller latency are faster than memory with faster frequency and greater latency.
I think that this is the case but only real/syntethic benchmark could give us the truth.
Click to expand...

Find some benchmark where memory scales on a 133mhz and a 1 CAS row addition. For example, DDR 400 to 533. You WILL find that it will perform better at 533 than at 400. If it didn't there wouldn't have been a point in the jump from DDR to DDR2 at all.
 
X72 said:
Find some benchmark where memory scales on a 133mhz and a 1 CAS row addition. For example, DDR 400 to 533. You WILL find that it will perform better at 533 than at 400. If it didn't there wouldn't have been a point in the jump from DDR to DDR2 at all.
Click to expand...

I think that it is very easy to understand that a 133MHz difference between 400 and 533 RAM is way bigger than the difference between 2666MHz and 2800MHz right? :)
 
sblantipodi said:
I'm not here to convince anyone.
Buy the ultra gigahertz ram and be happy :)
Click to expand...

You have things backward, you are trying to convince us of untruths and something you made up.
We are trying to get you to realise what you are saying is wrong.

This is what was said.
sblantipodi said:
I don't agree with that assertion.
Bandwidth without latency means nothing.

Bandwidth is the amount of work you can get in a fixed period of time, latency is the amount of time that is required to process this job.
Click to expand...
Nenu said:
Latency in this respect is a measure of individual times involved in "setting up" a memory transfer.
It is not a measure of overall latency.

This is why these latency timings dont have a marked impact on performance.
Click to expand...
sblantipodi said:
Probably you don't know what you are saying.
At same frequency, Ddr3 with cas 9 performs quite equals to ddr4 with cas 15.

Ddr4 has way more bandwidth that ddr3 due to its architecture but due to latency difference they performs quite the same at same frequency
Click to expand...
Nenu said:
Um ok.

This is about how changes in latency and speed can impact the performance of different DDR4 modules.
You are now cross comparing with DDR3.
How does this help you?
Click to expand...
Click to expand...

It is you that is buying ram based on false pretences, not me.
If you believe what you have typed is reasonable, you can buy some CAS 9 DDR4 :p

As I stated, the latency times discussed "ARE NOT" the time it takes to complete the Job (a memory transfer), they are individual times it takes to perform part of setting up the transfer process.
Your assertion that
sblantipodi said:
Frequency divided cas is a good meter when comparing the memory of the same technology
Click to expand...
Is incorrect.
If you disagree, please provide evidence that correlates with your statement.
 
sblantipodi said:
I think that it is very easy to understand that a 133MHz difference between 400 and 533 RAM is way bigger than the difference between 2666MHz and 2800MHz right? :)
Click to expand...

And I think it's very easy to make things up, right?

Here, something seeing material for you.

http://www.anandtech.com/show/7583/adata-xpg-v2-review-2x8-gb-at-ddr32800-121414-165-v/6

If you honestly think that memory becomes slower at higher mhz/higher CAS, then you're horribly misinformed and spouting out misinformation.

Even so, you're the one asking for which one is faster then calling people "don't know what you are talking about".
 
As promised, I will be building and testing my X99 machine tomorrow night. I did the benchmarking on my P67 tonight with my 2600K at 4.4GhZ (44x100) and Samsung RAM set to DDR3-1866 w/ 9-9-9-24 (this is the Samsung super awesome RAM that was built as DDR3-1600 but most people were putting up to 2133 or higher with zero issues before 2133 was a regular thing). These are quick tests, but they should satisfy the craving. I'm just using Passmark and Sandra. Passmark for, well, the passmark numbers to compare because it's easy. All I'm taking from Sandra is the aggregated bandwidth and latency numbers. I will test the DDR4 at stock (2666, 16-18-18-35) as well as whatever I manage to tweak it to. Then all of this stupid bickering can stop, especially by the people who probably aren't even planning on building an X99 system anytime soon, and the OP can make a purchasing decision.
 
LAiN_OfTheWired said:
As promised, I will be building and testing my X99 machine tomorrow night. I did the benchmarking on my P67 tonight with my 2600K at 4.4GhZ (44x100) and Samsung RAM set to DDR3-1866 w/ 9-9-9-24 (this is the Samsung super awesome RAM that was built as DDR3-1600 but most people were putting up to 2133 or higher with zero issues before 2133 was a regular thing). These are quick tests, but they should satisfy the craving. I'm just using Passmark and Sandra. Passmark for, well, the passmark numbers to compare because it's easy. All I'm taking from Sandra is the aggregated bandwidth and latency numbers. I will test the DDR4 at stock (2666, 16-18-18-35) as well as whatever I manage to tweak it to. Then all of this stupid bickering can stop, especially by the people who probably aren't even planning on building an X99 system anytime soon, and the OP can make a purchasing decision.
Click to expand...

I ordered a lot of cool stuff, i7 5930K, Asus X99 Deluxe, Corsair 2666MHz, Corsair AX860i, Corsair H80i.

All parts is waiting me at the shop except for the RAM that is not available in Italy yet.
If 2800MHz is worth the switch, I'm in time to do the switch.

Thank you for your good intention and for your kind answer
 
Here's the quick and dirty results of my testing.

Quick notes: One of my sticks is defective and had to be removed for the system to even boot, so these results are with only triple-channel. Also, despite the RAM being sold as DDR4-2666, it defaulted to DDR4-2133 when installed and only had XMP profiles for 2800 and 3000. I checked the part numbers and they should all be 2666 instead. These sticks don't have a profile for 2666 anywhere that I can see from SPD/XMP. The 2800 profile was selected for my second round of DDR4 testing, and it still sets voltage at 1.2v.

Core i7 2600K @ 4.4GhZ, 8GB Samsung DDR3-1866 9-9-9-24 1.5v
passmark memory
memory mark: 2023
database operations: 79.5
read cached: 31541
read uncached: 11478
write: 5922
latency: 19.9
threaded: 11991

sandra:
11.86GB/s
19.8ns

----------

!! DEFECTIVE STICK OF RAM - ONLY TRIPLE CHANNEL !!

Core i7 5820K @ stock, 16GB Corsair DDR4-2133 15-15-15-36 1.2v
passmark memory
memory mark: 2421
database operations: 87.8
read cached: 27003
read uncached: 11497
write: 9500
latency: 30.1
threaded: 35281

sandra:
33.24GB/s
30.0ns

----------

Core i7 5820K @ 3.69GhZ, 16GB Corsair DDR4-2800 16-18-18-36 1.2v

passmark memory
memory mark: 2490
database operations: 89
read cached: 27745
read uncached: 12247
write: 9784
latency: 30.0
threaded: 37070

sandra:
37.63GB/s
29.9ns
Click to expand...

As you can see, the latency is slightly higher, but the bandwidth is much higher and overall test scores from passmark memory test are better (with the except of read cached). Overall, I'd say that if it's a significant amount more, go with the slightly cheaper 2666 RAM because it will almost assuredly perform at the 2800 and 3000 specifications. If it isn't a ton more, don't worry about changing your order.
 
LAiN_OfTheWired said:
Here's the quick and dirty results of my testing.

Quick notes: One of my sticks is defective and had to be removed for the system to even boot, so these results are with only triple-channel. Also, despite the RAM being sold as DDR4-2666, it defaulted to DDR4-2133 when installed and only had XMP profiles for 2800 and 3000. I checked the part numbers and they should all be 2666 instead. These sticks don't have a profile for 2666 anywhere that I can see from SPD/XMP. The 2800 profile was selected for my second round of DDR4 testing, and it still sets voltage at 1.2v.



As you can see, the latency is slightly higher, but the bandwidth is much higher and overall test scores from passmark memory test are better (with the except of read cached). Overall, I'd say that if it's a significant amount more, go with the slightly cheaper 2666 RAM because it will almost assuredly perform at the 2800 and 3000 specifications. If it isn't a ton more, don't worry about changing your order.
Click to expand...

Good.
can you tell me if the BCLK Strap must be changed to use the 2800MHz XMP profile?
Should you change it from 1 to 1.25???
 
That appears to be a limitation of the BIOS from what I can see. The stock one on my motherboard does force a bclk overclock when using the XMP profiles, but it appears most motherboards are starting to get BIOS revisions that don't require it.
 
LAiN_OfTheWired said:
That appears to be a limitation of the BIOS from what I can see. The stock one on my motherboard does force a bclk overclock when using the XMP profiles, but it appears most motherboards are starting to get BIOS revisions that don't require it.
Click to expand...

I would like to understand more this think.
Is there someone with an Asus X99 Deluxe board with a 2800MHz RAM that can tell what BCLK is using and if overclocking the CPU may negatively influence the use of the XMP profile?
 
Thanks LAiN there is very little information on DDR and how its frequency and latency affect the performance of X99 platforms out there. I'm honestly not sure what to get at the moment myself, still waiting for the dust to settle a little.
 
Tych-0 said:
Thanks LAiN there is very little information on DDR and how its frequency and latency affect the performance of X99 platforms out there. I'm honestly not sure what to get at the moment myself, still waiting for the dust to settle a little.
Click to expand...

too much people talk, too few people knows (and buy) :D
 
Tych-0 said:
Thanks LAiN there is very little information on DDR and how its frequency and latency affect the performance of X99 platforms out there. I'm honestly not sure what to get at the moment myself, still waiting for the dust to settle a little.
Click to expand...

No problem. Once my new RAM kit arrives (probably still have a week or so, Newegg does not have an advance replacement policy and unfortunately they were the only ones with this particular kit in stock with a discount on them) I will do an actual full review. I don't expect gaming performance to be any different, but I will test it in other aspects as well. If there are any specific benchmarks you want run before making a purchasing decision, just let me know.
 
Is the purpose of this discussion to see which kit would benchmark higher? If so, carry on.

But otherwise, why isn't anyone asking the more obvious question - does the speed difference practically matter- which I think was already answered with DDR3 where 2133 was the cutoff of it not mattering if it was faster.
 
You must log in or register to reply here.
Back
Top