Overclock.net banner

10201 - 10217 of 10217 Posts

·
Premium Member
Joined
·
6,678 Posts
So right now I am at 3800 MHz CL16 and 54ns of latency with dual channel single rank ram. Is there any benefit in me getting two more sticks for four sticks total? Specially if I game at 1440P?
Reviewers do say running dual rank of each memory channel does improve performance by a good margin.
 

·
Registered
Joined
·
1,345 Posts
Reviewers do say running dual rank of each memory channel does improve performance by a good margin.
Most of those reviews don't test a dual rank at the same highest single rank set-up though, they usually can't achieve the OC and CL of their dual channel single ranks.

See, right now I have my single ranks running at 54ns latency, 3800 MHz CL16. If I put in another pair, I might not be able to get those exact settings correct? And therefore there's a high probability of me getting similar, slightly better or slightly worse performance?

This is what I understand of it. Of you watch the HU video, none of his dual rank memory could reach his single rank max tuned ones with 54.6ns of latency.
 

·
Premium Member
Joined
·
6,678 Posts
Most of those reviews don't test a dual rank at the same highest single rank set-up though, they usually can't achieve the OC and CL of their dual channel single ranks.

See, right now I have my single ranks running at 54ns latency, 3800 MHz CL16. If I put in another pair, I might not be able to get those exact settings correct? And therefore there's a high probability of me getting similar, slightly better or slightly worse performance?

This is what I understand of it. Of you watch the HU video, none of his dual rank memory could reach his single rank max tuned ones with 54.6ns of latency.
It is possible you will not be able to achieve the same speeds. However, I just built this new PC and have 4 sticks of single rank B-Die in it. So far I have not done any real tuning, just set a speed and booted. I am at 3800/3800 CL16 1T and have 58ns latency. My secondary and tertiary timings are horrible right now, so I think I should be able to get down to 54-55ns latency when I dig into the tuning. So while you might be able to get higher and tighter on single rank with 2 sticks only, that doesnt mean it is impossible to manage the same as what you have with 4 sticks.
 

·
Registered
Joined
·
1,345 Posts
It is possible you will not be able to achieve the same speeds. However, I just built this new PC and have 4 sticks of single rank B-Die in it. So far I have not done any real tuning, just set a speed and booted. I am at 3800/3800 CL16 1T and have 58ns latency. My secondary and tertiary timings are horrible right now, so I think I should be able to get down to 54-55ns latency when I dig into the tuning. So while you might be able to get higher and tighter on single rank with 2 sticks only, that doesnt mean it is impossible to manage the same as what you have with 4 sticks.
But what will I get out of it for spending $100 for 2 more sticks at 1440P is my main question. If I needed the extra 16GB of RAM, that would probably be more compelling. But if someone has data (since I haven't seen it yet) of performance going up 5% or so at 1440P, perhaps it will be worth while.
 

·
Old crazy guy
Joined
·
1,739 Posts
But what will I get out of it for spending $100 for 2 more sticks at 1440P is my main question. If I needed the extra 16GB of RAM, that would probably be more compelling. But if someone has data (since I haven't seen it yet) of performance going up 5% or so at 1440P, perhaps it will be worth while.
It really depends on what do you mean with performance increase.
You can have an average/max FPS increase with some games even higher than 5%, like SOTR or AC, because their game engines are quite peculiar and the speed is directly proportional to system RAM performances.
Those are the same engines which FPS will increase with better ram latency.

In general most game engines tend to be optimized for a working memory set that fits a 16GB system ram configuration.
Which means that average and max fps is mostly the same with a 16GB or 32GB memory configuration.
Not always, but very often, when the game is CPU bound there's a gap of about 5% or more between the 2 configurations.

The reason is the optimization itself; if you need to keep in check the memory usage you need to swap in and out a lot of data.
When you are CPU bound obviously this swapping is going to be slower cause the rendering pipeline has the maximum priority.
With the current GPUs it's very easy to be CPU limited at 1440p in many if not most games even with a mid/entry-level card.
If you have a good card and an average CPUI then is going to be very easy you'll be CPU bound most of the times.

There are a lot of videos on YouTube comparing 16GB vs 32GB like this one:

You can see which ones give you more avg/max FPS and which ones does not show any improvement.
But you can also see the most important performance gains with more system ram; min FPS and 0.1/1 percentile.

Because all this swapping in and out will likely at some point choke the rendering pipeline.
Which means lower fps at some point, usually when you really don't need it, and stuttering.

What these videos are failing to show is how big is the difference in an average PC.
These tests are ran on similar, clean, Windows installations.
Which is not how most people are using their PC.
Utilities, game panels, antivirus engines, it's pretty common to have a hell of apps running in background.
They all compete together with Windows for a share of the system ram.

When you have only 16GB there's not much left and the stuttering can be massive.
While with 32GB there's usually plenty of ram left for everything else.
Swapping ram to the pagefile is very expensive and if you only have one SSD it's going to compete with the game swapping its data.

Streaming, or just recording the session or the instant replays, it's another use case where only 16GB can kill your fps.

I have an old PC in Italy with an FX-8350 that I use only a few times per year and that one too I upgraded it to 32GB.
Games are on a RAID-5 array, which is just a tad better than an HDD.
Playing from HDD with 16GB was a truly horrible experience, massive stuttering in some games.
The difference with 32GB is like night vs day.

On top of all this Windows itself with every new release is increasing its memory usage.
Starting from 1909 notebooks with 4GB are almost unusable, 2004 is even worse.
I had to do 4 upgrades to 8GB, 2 from HDD to SSD, during last summer between family and friends, due to desperation.
So you have to expect in the next 1-2 years a constant decrease of gaming performances with Windows 10 and only 16GB.
 

·
Registered
Joined
·
1,345 Posts
It really depends on what do you mean with performance increase.
You can have an average/max FPS increase with some games even higher than 5%, like SOTR or AC, because their game engines are quite peculiar and the speed is directly proportional to system RAM performances.
Those are the same engines which FPS will increase with better ram latency.

In general most game engines tend to be optimized for a working memory set that fits a 16GB system ram configuration.
Which means that average and max fps is mostly the same with a 16GB or 32GB memory configuration.
Not always, but very often, when the game is CPU bound there's a gap of about 5% or more between the 2 configurations.

The reason is the optimization itself; if you need to keep in check the memory usage you need to swap in and out a lot of data.
When you are CPU bound obviously this swapping is going to be slower cause the rendering pipeline has the maximum priority.
With the current GPUs it's very easy to be CPU limited at 1440p in many if not most games even with a mid/entry-level card.
If you have a good card and an average CPUI then is going to be very easy you'll be CPU bound most of the times.

There are a lot of videos on YouTube comparing 16GB vs 32GB like this one:

You can see which ones give you more avg/max FPS and which ones does not show any improvement.
But you can also see the most important performance gains with more system ram; min FPS and 0.1/1 percentile.

Because all this swapping in and out will likely at some point choke the rendering pipeline.
Which means lower fps at some point, usually when you really don't need it, and stuttering.

What these videos are failing to show is how big is the difference in an average PC.
These tests are ran on similar, clean, Windows installations.
Which is not how most people are using their PC.
Utilities, game panels, antivirus engines, it's pretty common to have a hell of apps running in background.
They all compete together with Windows for a share of the system ram.

When you have only 16GB there's not much left and the stuttering can be massive.
While with 32GB there's usually plenty of ram left for everything else.
Swapping ram to the pagefile is very expensive and if you only have one SSD it's going to compete with the game swapping its data.

Streaming, or just recording the session or the instant replays, it's another use case where only 16GB can kill your fps.

I have an old PC in Italy with an FX-8350 that I use only a few times per year and that one too I upgraded it to 32GB.
Games are on a RAID-5 array, which is just a tad better than an HDD.
Playing from HDD with 16GB was a truly horrible experience, massive stuttering in some games.
The difference with 32GB is like night vs day.

On top of all this Windows itself with every new release is increasing its memory usage.
Starting from 1909 notebooks with 4GB are almost unusable, 2004 is even worse.
I had to do 4 upgrades to 8GB, 2 from HDD to SSD, during last summer between family and friends, due to desperation.
So you have to expect in the next 1-2 years a constant decrease of gaming performances with Windows 10 and only 16GB.
Damn it... Okay I ended up buying another b-die kit... Hoping to get it to the same speed timings as my existing kit... They're identical, so it'll be 4X 3600 CL16 going to 3800 CL16
 

·
Registered
Joined
·
16 Posts
Guys hi. I have B-Die 4133 MHZ CL19 ram. I used dram calculator 3800 mhz fast mode with manual and imported timings but I dont have stability on my pc. Right now I have strange issues on my windows. So I disabled XMP settings then all problems(random reset etc) gone. I added my typhoon html. Can you also help me find the best values? I used recommend values
But when I run into a problem, I don't know what to do. Should I raise the voltage? If yes, which one? Should I play with "Termination Block" alternative values or "Cad_Bus Block"? if yes, in what order? Thanks. Have a nice day and be safe.

G.Skill F4-4133C19-8GTZR
 

·
Registered
Joined
·
2,336 Posts
My second matching kit, 2x8 single rank B-die, will be coming in soon. I plan on taking out my current kit and just dropping them in and not touching the bios, in hopes it will work with the speed, volts, and timings, of my current kit. If they pass all stress tests, what needs to be adjusted going from a 2x8 setup to a 4x8 setup?

My current Zen Timings:
2467173
 

·
Registered
Joined
·
500 Posts
i rather have tight tuned 2x8 then 32gbb.. no way you gonna stay within 1.1 soc and reach same timings with that much more memory for your mem controller to handle.. Im running 3733 14-15-14-14 at 1.085 soc

BFV 64 player my mem was at 7.2 of 16gb.. lol.. if your game need much more then its just bad optimized.. like ..star citizen.. nothing happens around you.. you're alone ..yet wants 16 gb.. haha
 

·
Registered
Joined
·
2,336 Posts
@reqq if you are replying to me it'd be best to understand why people are adjusting their memory to a 256bit path via configuring their memory to dual rank.
 

·
Premium Member
Joined
·
6,678 Posts
@reqq I think you are greatly overestimating how hard it is to run 4x8 sticks on these platforms. Maybe way back when that would have been difficult, but not on modern systems. I am already on 4x8 with very simialr settings as @OCmember and even less voltage and it all boots up fine and hasnt even required any serious tweaking. It works so easily.
 

·
Registered
Joined
·
1,345 Posts
My second matching kit, 2x8 single rank B-die, will be coming in soon. I plan on taking out my current kit and just dropping them and not touching the bios, in hopes it will work with the speed, volts, and timings, of my current kit. If they pass all stress tests, what needs to be adjusted going from a 2x8 setup to a 4x8 setup?

My current Zen Timings:
View attachment 2467173
Me too, I'm doing exactly what you are doing. I bought a second 2x8GB and it should be here tomorrow, b-die. Wish me luck. I'm sure our post will cross in the near Future.

@reqq I think you are greatly overestimating how hard it is to run 4x8 sticks on these platforms. Maybe way back when that would have been difficult, but not on modern systems. I am already on 4x8 with very simialr settings as @OCmember and even less voltage and it all boots up fine and hasnt even required any serious tweaking. It works so easily.
Are you using b-die and if so, what are your Zen timings and latency?
 
  • Rep+
Reactions: OCmember

·
Registered
Joined
·
529 Posts
What's the general consensus on GDM enabled 1T vs. GDM disabled 2T ?
I've done a comparisons using AIDA64's cache & memory benchmark, they seemed almost identical (differences within margin of error). However, AIDA64's accuracy is questionable.
 

·
Registered
Joined
·
393 Posts
What's the general consensus on GDM enabled 1T vs. GDM disabled 2T ?
I've done a comparisons using AIDA64's cache & memory benchmark, they seemed almost identical (differences within margin of error). However, AIDA64's accuracy is questionable.
My test on 2x 16gb dual rank b-die showed similar results:

15 15 15 30 45 270 2T GDM-Off was the same as 16 16 16 32 48 264 1T GDM-On - both @3800Mhz

tested with SiSoSandra
 

·
Registered
Joined
·
529 Posts
My test on 2x 16gb dual rank b-die showed similar results:

15 15 15 30 45 270 2T GDM-Off was the same as 16 16 16 32 48 264 1T GDM-On - both @3800Mhz

tested with SiSoSandra
You should make the timing same except for GDM for an apple-to-apple comparison.
 

·
Registered
Joined
·
393 Posts
You should make the timing same except for GDM for an apple-to-apple comparison.
Yes, sure.

But not gonna bother if its already the same between such similar timings. I dont think you will see any beneficial gains. But i am ok with it...means you can use 2T to stabilise higher MCLK without losing too much performance, imho.
 

·
Overclock the World
Joined
·
1,984 Posts
You should make the timing same except for GDM for an apple-to-apple comparison.
Yesterday :)
Yes, sure.

But not gonna bother if its already the same between such similar timings. I dont think you will see any beneficial gains. But i am ok with it...means you can use 2T to stabilise higher MCLK without losing too much performance, imho.
Sadly every person has to decide for himself, because it has negative sides

GDM OFF 2T
Positives:
  • faster at the same speed with same timings, same voltage, no pbo, nothing auto, just clkdrvstrgth difference
  • MUX run at fullspeed , it ends up faster even with two roudtrips

Negatives:
  • because memory runs at fullspeed it will fail faster on higher speeds.
    As the strain is far higher on them
My apparently good A0s still are no exception from the PCB issues
GDM On can run these @ 4400, GDM off can just barely stabilize 4066, 4133 is a current ongoing issue again


GDM ON
Positive:
  • less strain on the memory
  • can run lower tRFC to compensate for being slower
  • if the PCB is a holding point, also higher maximum MCLK
  • lower voltage at the same speed and less mainboard depended (less clkdrvstr)

Negatives:
  • slower at the same speed, as MUX run at half speed internally
  • odd's rounding happens several times inside the chain, not only tCK | While tCK & tRP are only the "master timings" which will be read out rounded. Remain rounding happens down the chain internally
  • less timing finecontrol that way, as you should do the math with even numbers to skip double rounding/tripple rounding

At the ends its still for everyone to decide
If you have a weak PCB or already exceed 1.48vDIMM on this low end,
Pushing GDM OFF will not give you many positive results
If you are already at the FCLK limit, it is reasonable to get GDM away, as you get more out of the same speed and it leaves you more timing finecontrol

My current issues are that 1.46-1.48 is what these dimms work on on their limits
1.54 with high clkdrv are instant shutdowns and hard crashes. Sometimes a memory lane loss
1.5 are about the limit as high as i can go on them before instant reboots happen.
GDM off is hard to sustain beyond 4067MT/s so far. No voltage headroom left 😅😐
Ends up as a question between 4067MT/s low tings vs 4200 even timings and higher fabric
I'm playing with RTT values now to see if there is a possibility to filter the higher VDIMM and delay hardcrashes
 
10201 - 10217 of 10217 Posts
Top