Overclock.net banner
Status
Not open for further replies.
1 - 20 of 34 Posts

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #1 · (Edited)
Edit

If you came here from clicking on "recommended for you," please take note that this was posted in the AMD CPU section. This discussion is about AMD CPUs (specifically Zen 3) and how memory overclocking benefits them and where the point of diminishing returns seems to occur. This is not a general discussion about memory overclocking, so this excludes Intel CPUs. If you came here because it showed up on your "recommended" feed and you don't own a Zen 3 AMD CPU, this post is not for you.

End Edit

I'm all for free performances. I have successfully overclocked 4X SR sticks to 3800 CL16 with a latency of 54.3ns. But awhile back, I remember spending several hours trying to get 4000 MHz when it dawned on me. Why was I doing this? I wasn't even aware of any benefits of clocking any higher memory speeds and lowering my latencies.

I then remembered an article from Hardware Unboxed: https://www.techspot.com/article/2140-ryzen-5000-memory-performance/

From testing 8 games on a 5900X, from their data, memory speeds around 3600 Mhz with latencies at 60ns and lower really had no benefit. It wasn't even a 1% increase in performance at times.

2518938


I just wanted to share this data to have a discussion on memory OC because I feel that many of us here are obsessed with OC memory but we lose sight and the benefits of it, if any.
 

·
H₂O Aficionado
Joined
·
4,762 Posts
My favourite article on the matter.


Note that they identified the sweet spot as 3200MHz CL16. OCN might not be interested in sweet spot so…

Alternatively, 3800MHz CL16 is a great target if you can get the infinity fabric to 1900MHz (Zen 2 testing).
 

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #3 ·
My favourite article on the matter.


Note that they identified the sweet spot as 3200MHz CL16. OCN might not be interested in sweet spot so…

Alternatively, 3800MHz CL16 is a great target if you can get the infinity fabric to 1900MHz (Zen 2 testing).
I think memory OC is like sheep following one another. I think the majority of people that are doing it don't know why they're doing it or what it's going to do for them.

I am and was one, not going to deny that. At some point, I just wondered "why am I doing this? What is it actually doing for me?" And I found out it's literally doing nothing.

I bet a lot of people testing for several hours and buying $300+ kits probably don't know why they're doing what they're doing either.
 

·
H₂O Aficionado
Joined
·
4,762 Posts
I think memory OC is like sheep following one another. I think the majority of people that are doing it don't know why they're doing it or what it's going to do for them.

I am and was one, not going to deny that. At some point, I just wondered "why am I doing this? What is it actually doing for me?" And I found out it's literally doing nothing.

I bet a lot of people testing for several hours and buying $300+ kits probably don't know why they're doing what they're doing either.
I’m not sure if I agree with the sheeple comment. Sure, the net gains may be minimal if you have a solid foundation or kit, but it’s still worth some effort.

It most scenarios, going above and beyond the recommendations you or I acknowledge may be fruitless. However, there are some pretty popular games that are exceptions to the matter.

I haven’t played Call of Duty Warzone but I’ve read a few comments from @Veii and he has some interesting results and info with regards to ram, timings, and that title.

Note that 100 million people play that game.
 

·
Registered
Joined
·
81 Posts
Glad to see 3600 cl18, especially dual rank, competing well with the old gold standard that was 3200 cl14.

When diving deep into the hole that was finding the best IC for zen 1/+ it easily got expensive real quick. It's nice to know now seeing 3600 cl18 k(its around those specs in general) are becoming cheaper, and more widely available that people don't have to do the same amount of research for getting good performance.
 

·
Registered
Joined
·
3,357 Posts
Hardware reviewers will show you that CPU choice does not make a huge impact in games. One could argue that the performance improvement of a $800 5950X is negligible over a $270 10700K. However, they are comparing CPUs on a test setup that uses the same memory kit. So if DRR4000 is 10% faster than DDR3000, what does that tell you? The higher the core count and IPC of your processor, the more memory bandwidth you need. Memory unlocks the performance of your CPU.
 

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #7 ·
I’m not sure if I agree with the sheeple comment. Sure, the net gains may be minimal if you have a solid foundation or kit, but it’s still worth some effort.

It most scenarios, going above and beyond the recommendations you or I acknowledge may be fruitless. However, there are some pretty popular games that are exceptions to the matter.

I haven’t played Call of Duty Warzone but I’ve read a few comments from @Veii and he has some interesting results and info with regards to ram, timings, and that title.

Note that 100 million people play that game.
The sheep comment wasn't meant to be an insult. The meaning of it means people doing something because others are doing it but they don't know why they're doing it.

Hardware reviewers will show you that CPU choice does not make a huge impact in games. One could argue that the performance improvement of a $800 5950X is negligible over a $270 10700K. However, they are comparing CPUs on a test setup that uses the same memory kit. So if DRR4000 is 10% faster than DDR3000, what does that tell you? The higher the core count and IPC of your processor, the more memory bandwidth you need. Memory unlocks the performance of your CPU.
No one said otherwise. We're discussing the point of diminishing returns. We're not saying that all attempts at any memory OC is pointless.

Did you see the test? 8 games tested and there was little to no improvement going from 3600 Mhz 60ns vs. 3800 Mhz and 54.6ns.
 

·
H₂O Aficionado
Joined
·
4,762 Posts
The sheep comment wasn't meant to be an insult. The meaning of it means people doing something because others are doing it but they don't know why they're doing it.



No one said otherwise. We're discussing the point of diminishing returns. We're not saying that all attempts at any memory OC is pointless.

Did you see the test? 8 games tested and there was little to no improvement going from 3600 Mhz 60ns vs. 3800 Mhz and 54.6ns.

I didn’t see as insulting, I just disagreed with it.

Curious to know the stats and identify how many people really OC their ram. I bet you most people, like the Eurogamer article mentioned, don’t even enable XMP or DOCP.

The knowledge for the masses is out there. Just as you or I have found. Just remember this is OCN and not all discussions are based on value or optimizations.

We’re usually trying to push hardware to its limit. Even if it reaches diminishing returns.

Most people here try to push their components to the limit or experiment with alternative approaches to computer hardware. Like peltier pads
 

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #9 ·
I didn’t see as insulting, I just disagreed with it.

Curious to know the stats and identify how many people really OC their ram. I bet you most people, like the Eurogamer article mentioned, don’t even enable XMP or DOCP.

The knowledge for the masses is out there. Just as you or I have found. Just remember this is OCN and not all discussions are based on value or optimizations.

We’re usually trying to push hardware to its limit. Even if it reaches diminishing returns.

Most people here try to push their components to the limit or experiment with alternative approaches to computer hardware. Like peltier pads
But the audience for this post is a OC forum. I get your point otherwise.
 
  • Rep+
Reactions: Slaughtahouse

·
Iconoclast
Joined
·
31,435 Posts
Some people try to set records or compete against others, I mostly just like to tinker and get as much as I can from whatever it is that I've got.

Seeing what I can do with cheap memory is more rewarding to me than trying to get that last ounce of performance with a high-end kit where the bulk of the sorting has already been done. Even if I have the budget for high-end stuff, I'm much more likely to buy a stack of OEM memory and bin it myself then I am to pay extra to deprive myself of that entertainment. I do have some reasonably well (pre)binned B-die stuff (a couple of kits of Patriot 4400), but honestly I have a lot more fun with my OEM sticks and budget off-brand kits.

Of course, gaming is not all, or even the majority of what I do on most of my systems and memory performance can matter much more for some tasks. What I do have, I push as far as is practical--meaning as far as I can maintain unconditional stability and continue to see measuable, even if small, gains.
 
  • Rep+
Reactions: i n f a m o u s

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #11 ·
Some people try to set records or compete against others, I mostly just like to tinker and get as much as I can from whatever it is that I've got.

Seeing what I can do with cheap memory is more rewarding to me than trying to get that last ounce of performance with a high-end kit where the bulk of the sorting has already been done. Even if I have the budget for high-end stuff, I'm much more likely to buy a stack of OEM memory and bin it myself then I am to pay extra to deprive myself of that entertainment. I do have some reasonably well (pre)binned B-die stuff (a couple of kits of Patriot 4400), but honestly I have a lot more fun with my OEM sticks and budget off-brand kits.

Of course, gaming is not all, or even the majority of what I do on most of my systems and memory performance can matter much more for some tasks. What I do have, I push as far as is practical--meaning as far as I can maintain unconditional stability and continue to see measuable, even if small, gains.
Yeah, not saying there's no reason to OC if you enjoy benchmarking.

I'm presenting data for those that are trying to shoot for gaming performance, giving them a target to aim for because, there's no actual target for someone to aim for right now.
 

·
Registered
Joined
·
778 Posts
Hardware reviewers will show you that CPU choice does not make a huge impact in games. One could argue that the performance improvement of a $800 5950X is negligible over a $270 10700K. However, they are comparing CPUs on a test setup that uses the same memory kit. So if DRR4000 is 10% faster than DDR3000, what does that tell you? The higher the core count and IPC of your processor, the more memory bandwidth you need. Memory unlocks the performance of your CPU.
hardware reviewers do not test online games, due to standardisation. Taking their data as 100% truth is a mistake i will never repeat. Ryzen b-die 3400 cl14 tunned on 2600 at 4.2ghz and 8700k at 5ghz. Reviewer test in bf 1 campaign spits out 172fps for the ryzen and 175 for 8700k, results completely replicable and repeatable by any user, i know cus i did it with 2 pcs. You go into the online mode and after 10min of gameplay the 8700k, gives you 25% better framerate on avg.
CPUs make a huge impact, if you have the GPU power for it.
 

·
Not New to Overclock.net
Joined
·
5,373 Posts
I'm all for free performances. I have successfully overclocked 4X SR sticks to 3800 CL16 with a latency of 54.3ns. But awhile back, I remember spending several hours trying to get 4000 MHz when it dawned on me. Why was I doing this? I wasn't even aware of any benefits of clocking any higher memory speeds and lowering my latencies.

I then remembered an article from Hardware Unboxed: https://www.techspot.com/article/2140-ryzen-5000-memory-performance/

From testing 8 games, from their data, memory speeds around 3600 Mhz with latencies at 60ns and lower really had no benefit. It wasn't even a 1% increase in performance at times.

View attachment 2518938

I just wanted to share this data to have a discussion on memory OC because I feel that many of us here are obsessed with OC memory but we lose sight and the benefits of it, if any.
funny youtubers

y dont they choose say a 10700k?? it has no gearing. scales. so u can see actual impact of ram oc.

scale with fixed vdimm say 1.5v dual rank
3866c14 vs 4266c16 vs 4533/4600c17
vs sr
4000cl14, 4400cl16 vs 4700/4800cl17

but thet wont. cause its all about views.
2nd they dont know how. its all about auto.

if they want to say ONLY for amd then be clear on it
start with gearing
and the
fclk bandwidth vs tcl

pretty simple.
 

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #14 ·
funny youtubers

y dont they choose say a 10700k?? it has no gearing. scales. so u can see actual impact of ram oc.

scale with fixed vdimm say 1.5v dual rank
3866c14 vs 4266c16 vs 4533/4600c17
vs sr
4000cl14, 4400cl16 vs 4700/4800cl17

but thet wont. cause its all about views.
2nd they dont know how. its all about auto.

if they want to say ONLY for amd then be clear on it
start with gearing
and the
fclk bandwidth vs tcl

pretty simple.
Why do I care about a Intel CPU? I'm here to talk about AMD Zen 3.

They tested 8 games with several memory speeds and latencies for a 5900X.

I don't understand your whole post, they literally custom tuned the memory modules themselves.
 

·
Not New to Overclock.net
Joined
·
5,373 Posts
Why do I care about a Intel CPU? I'm here to talk about AMD Zen 3.

They tested 8 games with several memory speeds and latencies for a 5900X.

I don't understand your whole post, they literally custom tuned the memory modules themselves.
its the discrepency in general when youtubers make (ram) effects

was stating a fact for awareness

1. have a baseline awareness if fclk wasnt a limit.. example use intel

2. zone to amd.. start with gearing
3. choose a gear with fix a vdimm and best case tcl
4. dual rank vs sr with fixed vdimm

example 1.5v fixed bdie.
so is 4000cl14 better than say 3733cl13?
then decide between performance vs say requirement of imc voltages and complexity

then choose the winner of than and redo a dr one

then sr vs dr

this is how data is useful to end user
once u conclude

then go vs say cheaper 32gb klevv djr dr kit
etc. usd 130 vs usd 400 dr bdie..

look at the review. can u even form any answer to the above

with this ppl can gauge..

they tested nothing and swayed ppl to their lack off understanding ram tuning. hence y u were in a state of confusion and just accepted theirs which is flawed.
 

·
Registered
Joined
·
1,683 Posts
Discussion Starter · #16 ·
its the discrepency in general when youtubers make (ram) effects

was stating a fact for awareness

1. have a baseline awareness if fclk wasnt a limit.. example use intel

2. zone to amd.. start with gearing
3. choose a fear fix a vdimm and best case tcl
4. dual rank vs sr with fixed vdimm

example 1.5v fixed bdie.
so is 4000cl14 better than say 3733cl13?
then decide between performance vs say requirement of imc voltages and complexity

then choose the winner of than and redo a dr one

then sr vs dr

this is how data is useful to end user
once u conclude

then go vs say cheaper 32gb klevv djr dr kit
etc. usd 130 vs usd 400 dr bdie..

look at the review. can u even form any answer to the above

with this ppl can gauge..
That wasn't the point of the test, this isn't about Memory on its own. This is a about memory and how it affects Zen 3 processors.

Sorry but your post is irrelevant.

We're on a AMD CPU specific sub here. How did you find this post? Did it pop up on your feed or are you aware this is a AMD sub post?

The point was to study how memory OC benefits Zen 3 and that's it. It wasn't to conclude anything about memory, it was to conclude about Zen 3.

Intel is irrelevant here.

The data is useful, they custom tuned memory modules, look at the latencies for the tuned modules. The data shows that there isn't a big benefit with 3800 MHz at 54.3ns of latency vs. 3600 MHz with 60ns of latency.
 

·
Registered
Joined
·
3,357 Posts
hardware reviewers do not test online games, due to standardisation. Taking their data as 100% truth is a mistake i will never repeat. Ryzen b-die 3400 cl14 tunned on 2600 at 4.2ghz and 8700k at 5ghz. Reviewer test in bf 1 campaign spits out 172fps for the ryzen and 175 for 8700k, results completely replicable and repeatable by any user, i know cus i did it with 2 pcs. You go into the online mode and after 10min of gameplay the 8700k, gives you 25% better framerate on avg.
CPUs make a huge impact, if you have the GPU power for it.
The Intel 8700K has both a massive IPC and clock speed advantage over the Ryzen 2600, so I am not surprised the framerates are higher. I agree that you can't trust mainstream online reviewers. However, this comparison does not tell you anything about memory speed considering the 8700K is so much faster.
 

·
Registered
Joined
·
2,099 Posts
I think memory OC is like sheep following one another. I think the majority of people that are doing it don't know why they're doing it or what it's going to do for them.

I am and was one, not going to deny that. At some point, I just wondered "why am I doing this? What is it actually doing for me?" And I found out it's literally doing nothing.

I bet a lot of people testing for several hours and buying $300+ kits probably don't know why they're doing what they're doing either.
Hey, maybe they dont, but some of us just enjoy pushing our hardware irrespective of the gains

:D :D
 
  • Rep+
Reactions: SirPerfluous

·
Not New to Overclock.net
Joined
·
5,373 Posts
The Intel 8700K has both a massive IPC and clock speed advantage over the Ryzen 2600, so I am not surprised the framerates are higher. I agree that you can't trust mainstream online reviewers. However, this comparison does not tell you anything about memory speed considering the 8700K is so much faster.
yeah but what i am pointing out

1. reviewers are stingy baskets thats not going to bin cpus. they will use whatever free **** they have and assume its the same for 99% of ppl out there. since they cant do fclk 2100 etc be it their cpu/ram/board/genetic brain. create a baseline where fclk is unlimited . use intel.
2. then zoned to amd gearing
3. choose one gear like gear 1 and then fix a vdimm at lowest tcl..
4. then do sr vs dr
5. then cost of chipset
6. efficiency.. the requirement from the imc voltages and long term stability

this encompasses all

with this for example a person can be well informed.
user who only can do 3600 but maybe c13 might be better than chasing for 3866c14 with dr kit.. his stability for 2 dimm dr is much better than say 4dimm sr kit..since his speed/imc related voltages are less of a strain to the cpu..

or a user can chase flck 2100

or boards.. what they can do be it board or bios limitation..

this is what the community needs.

that review is so skewed to a youtubers inherent genetic flaw.
 
1 - 20 of 34 Posts
Status
Not open for further replies.
Top