DDR4 3200 14-14-14-34 VS DDR4 3600 17-19-19-39 Which is better

[Crunchakah]

Hey Everyone,
I have picked up some G. Skill DDR4 3200 2x16gb 14-14-14-34 memory for my new Ryzen build. I just saw in Ryzen news that it will support 3600mhz memory. However the G. Skill 2x16gb DDR4 3600 17-19-19-39 is slow timings. Curious if the 3200 will be a better choice still over all.

3600mhz is also $130 more.


here are the links too:
3200mhz http://

3600mhz http://

 

[greens]

Your title and your post details have conflicting info. Is the 3200 RAM 14 or 17?

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

(17/3600)*1000)= 4.72ns
(14/3200)*1000)=4.375ns
(17/3200)*1000)=5.3125ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

 

[Isokolon]

for 130$ more?

the 3200 will be good enough, should be around the same speed
very few applications show enough difference to justify buying new RAM anyway

 

[Crunchakah]

Your title and your post details have conflicting info. Is the 3200 RAM 14 or 17?

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

(17/3600)*1000)= 4.72ns
(14/3200)*1000)=4.375ns
(17/3200)*1000)=5.3125ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

Sorry I have corrected the clock speeds on the original post.

3200 14-14-14-34
3600 17-19-19-39

 

[KingReZo]

It'll literally make no difference in performance in Gaming or Rendering, None that you'll notice. Doesnt justify the extra 130 dollars for it in my opinion

 

[Pinhedd]

Your title and your post details have conflicting info. Is the 3200 RAM 14 or 17?

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

(17/3600)*1000)= 4.72ns
(14/3200)*1000)=4.375ns
(17/3200)*1000)=5.3125ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

Yeah... that's not how it works at all.

 

[greens]

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

Yeah... that's not how it works at all.

Yes, yes it is.

https://www.google.com/search?sclient=psy-ab&site=&source=hp&q=ram+timing+vs+latency&btnK=Google+Search

Feel free to try and find a source stating otherwise! But please do find one, so we don't misinform users on Toms.

Not that anyone will probably be looking at this necro'd post most likely.

 

[Pinhedd]

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

Yeah... that's not how it works at all.

Yes, yes it is.

https://www.google.com/search?sclient=psy-ab&site=&source=hp&q=ram+timing+vs+latency&btnK=Google+Search

Feel free to try and find a source stating otherwise! But please do find one, so we don't misinform users on Toms.

Not that anyone will probably be looking at this necro'd post most likely.

No... it isn't.

The fact that you think that DDR4-3200 has a 3200Mhz clock betrays your ignorance of the subject.

Google isn't a source, search results are individualized.

This is a source, https://www.dropbox.com/s/awpw6cle38straf/Memory_FAQ.pdf?dl=0

 

[The Paladin]

psst check a wiki one day...

DDR4 operates at a voltage between 1.2 V and 1.4 V with a frequency between 800 and 4266 MHz, compared to frequencies between 400 and 1067 MHz[9][a] and voltage requirements of 1.5 or 1.65 V of DDR3. Due to the nature of DDR, speeds are typically advertised as doubles of these numbers (DDR3-1600 and DDR4-2400 are common)

DDR-3200 actually has a frequency of 1600. That way a DDR4-1600 chip always shows as DDR4 800mhz in speccy for example.
DDR4-3600 req is 1800.
so
(CL/(freq-in-MHZ) * 1000) = timing in ns
(14/1600)*1000 = 8.75ns
(17/1800)*1000)= 9.44ns
so the DDR4-3600 is 0.61ns slower for 130$ less... anything under 1 ns is meaningless.
tidbit for you nerds out there "Light travels exactly 29.9792458  centimeters in 1 nanosecond. This is equivalent to 11.8 inches"


gets off soap box... schools out .. you are dismissed

 

[greens]

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

Yeah... that's not how it works at all.

Yes, yes it is.

https://www.google.com/search?sclient=psy-ab&site=&source=hp&q=ram+timing+vs+latency&btnK=Google+Search

Feel free to try and find a source stating otherwise! But please do find one, so we don't misinform users on Toms.

Not that anyone will probably be looking at this necro'd post most likely.

No... it isn't.

The fact that you think that DDR4-3200 has a 3200Mhz clock betrays your ignorance of the subject.

Google isn't a source, search results are individualized.

This is a source, https://www.dropbox.com/s/awpw6cle38straf/Memory_FAQ.pdf?dl=0

The D in DDR stands for double. When solving for something it's perfectly acceptable. 2x=2y+4 Is the same as X=y+2.

No one in there right mind is gonna click a drop box link from a troll man...

Additionally we don't really like to throw personal attacks around here on toms.

Google is a fairly poor source indeed, but when every single result supports my basic calculation... It doesn't serve so much as a source, but rather proof that such things are common knowledge.

 

[greens]

It's the cheaper ram that is .61 faster, not the more expensive kit...

 

[The Paladin]

Greens thanks It was late when I typed that, I corrected it...

 

[Pinhedd]

Here:
(CL/(freq-in-MHZ) * 1000) = timing in ns

14/3200 is by far fastest.
RAM frequency is a big part of it, but so is CAS/CL - and being the numerator - a change to CAS/CL has a greater impact in overall performance than frequency.

Yeah... that's not how it works at all.

Yes, yes it is.

https://www.google.com/search?sclient=psy-ab&site=&source=hp&q=ram+timing+vs+latency&btnK=Google+Search

Feel free to try and find a source stating otherwise! But please do find one, so we don't misinform users on Toms.

Not that anyone will probably be looking at this necro'd post most likely.

No... it isn't.

The fact that you think that DDR4-3200 has a 3200Mhz clock betrays your ignorance of the subject.

Google isn't a source, search results are individualized.

This is a source, https://www.dropbox.com/s/awpw6cle38straf/Memory_FAQ.pdf?dl=0

The D in DDR stands for double. When solving for something it's perfectly acceptable. 2x=2y+4 Is the same as X=y+2.

No one in there right mind is gonna click a drop box link from a troll man...

Additionally we don't really like to throw personal attacks around here on toms.

Google is a fairly poor source indeed, but when every single result supports my basic calculation... It doesn't serve so much as a source, but rather proof that such things are common knowledge.

The D in DDR does indeed stand for double, but to finish the acryonym it's Double Data Rate, not Double Clock Frequency.

SDRAM timings are expressed in clock cycles not data transfers.

Every single result on google (which again, is personalized) supporting your calculation simply shows how massively ignorant the masses are of how SDRAM and memory controllers actually work.

Every authoritative source, including those from JEDEC, Hynix, Samsung, and Micron shows otherwise.

Just for the record, I've built my own SDRAM memory controller and used it in FPGA designs. I know how SDRAM timings work.

 

[Tanyac]

Just for the record, I've built my own SDRAM memory controller and used it in FPGA designs. I know how SDRAM timings work.

I know this thread is old...
I am looking into a Threadripper / X399 Taichi Mobo build in early 2018 and trying to figure out what memory to get

So after all these various answers what is the correct answer.
It seems to me that the 3200 is a better choice, but there is so much disagreement on calculations who really knows what the real/correct answer is?

thanks
Sorry to open old wounds

 

[ddcservices]

Ryzen does not follow the typical performance pattern you see from Intel chips, so there are a number of things going on. From what I have gathered, DDR4-3433 memory with a CAS 14 latency will beat out DDR4-3600 CL 16 memory when it comes to performance. While Ryzen runs Infinity Fabric at half the RAM speed(hinting a higher RAM speed is the key), that lower latency will make a bigger difference when it comes to performance, but obviously, 2133 to 3200 memory will be a big enough jump where 3200 CL16 would be better than 2133 CL14 memory. You will want the CL14 memory in the case of 3200 vs 3600, but the difference overall won't be huge. 3466 CL 14 beats 3600 CL 16.

Just for the record, I've built my own SDRAM memory controller and used it in FPGA designs. I know how SDRAM timings work.

I know this thread is old...
I am looking into a Threadripper / X399 Taichi Mobo build in early 2018 and trying to figure out what memory to get

So after all these various answers what is the correct answer.
It seems to me that the 3200 is a better choice, but there is so much disagreement on calculations who really knows what the real/correct answer is?

thanks
Sorry to open old wounds

 

[Isokolon]

Doesn't really matter. RAM speed makes very little difference apart from some very specific tasks.

With AMDs Infinity fabric get the highest clock rate your board supports, timings are less relevant.

 

[jharrison1185]

I bought Gskill trident RGB 16gb (8x2) 3200mhz CL14. Using the Ryzen dram calculator app for PC and the recommended timings, I was able to overclock them to 3466 @CL14 on my Asus ROG Strix gaming X370-F. It was a pain, and I had to bump the ram voltage to 1.41V and the SOC to 1.2V. All the timings and subtimings (about 20 of them!) are different. Now running at 14-14-14-28 3466 CL14. These DRAM's were not on my QVL however since I knew these were high quality Samsung B-die I was confident they would work at least at rated speeds, and they did once I went into bios and selected DOCP (XMP) standard. The benchmarks are only slightly faster, but I paid $229 on sale (which is robbery) so I wanted to get the best speed I possibly could. 3466 cl14 with tightened sub timings (thanks to Ryzen Dram calculator @ fast settings) is faster than 3600 cl16. I got the idea from a channel called Mindblank on YouTube. I hope this helps because I was able to get great speed and low latency.

 

[WINTERLORD]

am just curiuos to get c14 to work with 3466 data rate with samsung b dies how much memory voltage to you need to give them?