What do you consider fast RAM?

[Nena Trinity]

what is fast, average & slow?

[Moonzy]

depends on my financial department_{aka my wallet}

[Mateyyy]

Fast - 3600 CL14, 4000 CL15, somewhere along those lines

Fast enough for most people - average 3600 CL16 (or 3200 CL14 at XMP, but why wouldn't you overclock B-die)

Average - the almighty, bog standard, cheaper than 2400MHz, 3200 CL16

Slow - 3000 CL16

 

Slower than that just doesn't make sense since 3000/3200MHz kits nowadays cost as much as 2133/2400MHz ones.

[Chris Pratt]

Really don't understand this poll, seems like you're comparing on timings only. The "average" CL timing is going to be basically 1/2000 the clock speed. So 3200Mhz CL16 or 3600Mhz CL18, for example. Lower than that is "faster", higher is "slower". However all of the clock speeds are "average". Slow would be something in the 2000 range and fast would be something in the 4000 range.

[Nena Trinity]

5 minutes ago, Chris Pratt said:

Really don't understand this poll, seems like you're comparing on timings only. The "average" CL timing is going to be basically 1/2000 the clock speed. So 3200Mhz CL14 or 3600Mhz CL16, for example. Lower than that is "faster", higher is "slower". However all of the clock speeds are "average". Slow would be something in the 2000 range and fast would be something in the 4000 range.

https://notkyon.moe/ram-latency.htm

 

This can be used to measure NS how ever there is truth that some programs prefer bandwith over something with lower MHz that is theoretically faster...

[Tan3l6]

Samsung b-die and Micron e-die are fast. Although some trash binned B-dies are very average.

Many are average, but can't bring out any exceptionally slow ones.

[DildorTheDecent]

What I think is fast for Coffee Lake/Comet Lake (CAS latency not included cos idk):

Fast: 4400

Average: 4000

Slow: 3600

 

What I think is fast for Zen 2 (assume FCLK 1:1):

Fast: 3800 C16

Average: 3600 C18

Slow: 3400 C17

[dfsgsfa]

ram stays in ram slot and will not move

[Nena Trinity]

Update: (10-08-2020)

*Faster is considered fast. 

*3000mhz cl15, 3200mhz CL16 & so on is considered most average. 

*Slower is considered slow huh? 2933mhz and slower then? Wonder if 2666MHz CL10 is considered slow?

[Haro]

for zen 2, assuming we keep the fclk at 1:1.

3800cl17/16: fast.

3600cl16/3200cl14: average.

3200cl16: slow but still good enough.

for coffee lake and comet lake.

4000/4400: fast.

3600/3800: average.

3200: slow but good enough.

 

Edited August 10, 2020 by TofuHaroto

[Nena Trinity]

1 minute ago, TofuHaroto said:

for zen 2, assuming we keep the fclk at 1:1.

3800: fast.

3600: average.

3200: slow but still good enough.

for coffee lake and comet lake.

4000/4400: fast.

3600/3800: average.

3200: slow but good enough.

 

I think it's more likely and easier to get 3600MHz CL16 to 3333MH CL15 or better working, so I consider that fast and 3200MHZ CL16, 3000MHz CL15 or 2666MHz CL13 as average. Anything below is slow or below average IMO then again 3000mhz at cl14 would at the very least be above average.

[Haro]

4 minutes ago, Nena Trinity said:

3000mhz at cl14 would at the very least be above average.

I mean considering that would perform a out the same as cl16 3200 I would say it's good enough. 

But yea I edited my post. Forgot to add timings. 

[Nena Trinity]

31 minutes ago, TofuHaroto said:

I mean considering that would perform a out the same as cl16 3200 I would say it's good enough. 

But yea I edited my post. Forgot to add timings. 

Well the 3000MHz CL15 would be the same, a CL14 should be a tier above it but it real world use we would be able to tell then apart huh?

[Haro]

10 hours ago, Nena Trinity said:

a CL14 should be a tier above it but it real world use we would be able to tell then apart huh?

well yea assuming it's the same frequency, yea it wont make a difference. 

[Nena Trinity]

13 minutes ago, TofuHaroto said:

well yea assuming it's the same frequency, yea it wont make a difference. 

So we need 2 tiers above or more then? 3000MHz CL13 would be able to perform better in any notable way? In which case we now are at under 9ns territory!

[Haro]

10 hours ago, Nena Trinity said:

3000MHz CL13 would be able to perform better in any notable way?

for  a casual user? no. 

in fact 3200 cl16 amd faster wouldn't make a huge difference for a casual user.

[Nena Trinity]

4 minutes ago, TofuHaroto said:

for  a casual user? no. 

in fact 3200 cl16 amd faster wouldn't make a huge difference for a casual user.

Well I got more CB score both when replacing the 2933mhz CL16 ram with 3066mhz CL16 and even more when lowering timings to CL14. How many fps I get more in fortnite nao? (didn't really check before and after but the score in R20 wasn't so much so maybe 2-4fps max?) 

[Mateyyy]

23 minutes ago, Nena Trinity said:

Well I got more CB score both when replacing the 2933mhz CL16 ram with 3066mhz CL16 and even more when lowering timings to CL14. How many fps I get more in fortnite nao? (didn't really check before and after but the score in R20 wasn't so much so maybe 2-4fps max?) 

Cinebench doesn't care about memory, like at all. I'm on single-channel now and I literally get the same scores as before.

When it comes to gaming though, every little bit counts  

[Nena Trinity]

12 minutes ago, Mateyyy said:

Cinebench doesn't care about memory, like at all. I'm on single-channel now and I literally get the same scores as before.

When it comes to gaming though, every little bit counts  

It matters more on AMD due to how its Infinity fabric works, but lowering timings gave me a slightly higher single core score on my 10600 going from 499 to 503 that could be withing the margin of error but still.

[Guest]

I bench at 4000mhz Cl 16-16-16-36.

Faster would be CL14 or tighter.

Motherboard makes a difference.

 

People scared to unlinked IF to run fast memory with AMD.

So AMD on average runs memory slow.

[DildorTheDecent]

 

9 hours ago, ShrimpBrime said:

People scared to unlinked IF to run fast memory with AMD.

So AMD on average runs memory slow.

Latency hits the shitter when it's unlinked. Doesn't really matter for some benches but matters for video games.

 

1900 FCLK is way too low of a "limit"

[Guest]

26 minutes ago, DildorTheDecent said:

 

Latency hits the shitter when it's unlinked. Doesn't really matter for some benches but matters for video games.

 

1900 FCLK is way too low of a "limit"

It used to be that HT overclocking (now called IF) didnt produce much gains because it was more used for multiple sockets vs now, the multiple CCXs. 

So that's why we get good gains.

 

Latency loss isnt terrible if you can make it up in bandwidth. 

 

I mean there's been plenty of examples with 5ghz DDR4 memory frequency, but this is on stupid expensive memory and definitely a waste of moneys for average users.

[porina]

9 hours ago, ShrimpBrime said:

I bench at 4000mhz Cl 16-16-16-36.

Faster would be CL14 or tighter.

Motherboard makes a difference.

 

People scared to unlinked IF to run fast memory with AMD.

So AMD on average runs memory slow.

For the ram I have I've been unable to find a use case where faster unsync ram beats 3600C17/18 depending on Intel or AMD system. The ram I have I never got running at lower latency, and suspect the secondary/tertiary is more limiting at this point. And 3200C16 dual rank ram will beat 3600+ single rank. People never seem to consider rank or DPC. Having said that, I've never managed to show rank/DPC to make a difference in gaming which seems to be pretty much all anyone cares about around here.

 

On the original poll, I'd consider reasonably fast to be 3600C16. Decent mid range would be 3200C16. Slow is anything below 2933.

[Guest]

33 minutes ago, porina said:

For the ram I have I've been unable to find a use case where faster unsync ram beats 3600C17/18 depending on Intel or AMD system. The ram I have I never got running at lower latency, and suspect the secondary/tertiary is more limiting at this point. And 3200C16 dual rank ram will beat 3600+ single rank. People never seem to consider rank or DPC. Having said that, I've never managed to show rank/DPC to make a difference in gaming which seems to be pretty much all anyone cares about around here.

 

On the original poll, I'd consider reasonably fast to be 3600C16. Decent mid range would be 3200C16. Slow is anything below 2933.

The 2 kits I have are A2 PCB B rated B-Die. 

They are fast. Single rank.

 

Looking at voltage scaling here, the memory is readily capable.

 

3600mhz CL16 is slow when that kit would be running CL12 at this frequency.

 

Competitive Benching, sweet spots vary on better kits, but 4000mhz CL12-12-12 is desirable, however my kits won't do it, partly because of the motherboard, I'd want to invest into the Apex for that.

 

So a lot of this, as you mention there, we are not taking into many variables that makes memory fast.

 

But of course I love reading at OCN because of all the testing Woo has done. But I also keep in mind his hardware is his and mine is mine.... our results all vary.

[Nena Trinity]

13 minutes ago, ShrimpBrime said:

The 2 kits I have are A2 PCB B rated B-Die. 

They are fast.

 

Looking at voltage scaling here, the memory is readily capable.

 

3600mhz CL16 is slow when that kit would be running CL12 at this frequency.

 

Competitive Benching, sweet spots vary on better kits, but 4000mhz CL12-12-12 is desirable, however my kits won't do it, partly because of the motherboard, I'd want to invest into the Apex for that.

 

So a lot of this, as you mention there, we are not taking into many variables that makes memory fast.

 

But of course I love reading at OCN because of all the testing Woo has done. But I also keep in mind his hardware is his and mine is mine.... our results all vary.

That is so true often the board hold you back, I tested 2 kits on a ASRock B450 board none of em wanted to go very fast one, my 3200MHz CL16 kit nope only 2933MHz CL16, the 2666MHz from my other PC which many reviewers say can easilly hit 3333MHz? Nope it only went to 3066MHz... Well I did end up keeping the sticks in the opposite machines and now I got this mess!

 

SK Hynix (single rank) @3066MHz CL14-17-17-34 (in Ryzen system)

&

Samsung (single rank) @2666MHz CL13-15-15-30 (in Intel system)

 

acording to the DDRAM calculator both kits should be faster than 10ns so yay?