Post your Zen 2/2+ memory speeds!

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
Should I work on bringing down the trfc2 and c4 values in a similar way?
The c2 and c4 seem to adjust automatically based on trfc, at least that was how my ASRock did it in Ryzen. Otherwise, just set at 256 and 192 or something.
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
The c2 and c4 seem to adjust automatically based on trfc, at least that was how my ASRock did it in Ryzen. Otherwise, just set at 256 and 192 or something.
Thank you for the tip. I am upgrading from a 2500K so all new O/cing world. Making good progress. Going to do some stability testing

New setup.PNG
 

tangoseal

[H]F Junkie
Joined
Dec 18, 2010
Messages
8,937
Can you run an adia mem benchmark. I am curious about your latency.

To me this is not impressive, however my 3900x, not pictured in this thread, is hitting around 67ns total latency which is from 3600bdie @ 14ns tweaked timings.

Here is the 3600 non x using Gskill Hynix CJR (C Die)

This is to be expected because as bandwidth increases upwards, if you are able to keep timing tight and mean, like 14-15ns at 3400+mhz ram then your final latency should also go down. So yes high bandwidth ram is superior on Zen 2 when using aggressive higher voltage and timings than just raw high bandwidth using XMP. XMP is chump settings. Guaranteed safe, guaranteed to under-perform.

ram.png
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
To me this is not impressive, however my 3900x, not pictured in this thread, is hitting around 67ns total latency which is from 3600bdie @ 14ns tweaked timings.

Here is the 3600 non x using Gskill Hynix CJR (C Die)

This is to be expected because as bandwidth increases upwards, if you are able to keep timing tight and mean, like 14-15ns at 3400+mhz ram then your final latency should also go down. So yes high bandwidth ram is superior on Zen 2 when using aggressive higher voltage and timings than just raw high bandwidth using XMP. XMP is chump settings. Guaranteed safe, guaranteed to under-perform.

View attachment 180097
I can see that for sure..
 

ccityinstaller

Supreme [H]ardness
Joined
Feb 23, 2007
Messages
4,241
Thank you for the tip. I am upgrading from a 2500K so all new O/cing world. Making good progress. Going to do some stability testing

View attachment 180096


I would highly suggest you download Typhoon burner (to confirm what ICs your ram kit uses) and then take that and plug it into Ryzen Ram Calc 1.6 is the new version...
,
Pick Zen2 for your CPU, the type of memory IC you have, if it's single or dual rank (again TB will tell you this but most 8GB DIMMs are single), and if you are on a 360/370,450/470/570 platform and the speed you want to attempt first ..

Then just click Fast, take a photo with your phone, and go into the bios and set alll the subtimings it gives you. I prefer the bios way but you can also modify them with Ryzen Master in windows if you want.
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
I would highly suggest you download Typhoon burner (to confirm what ICs your ram kit uses) and then take that and plug it into Ryzen Ram Calc 1.6 is the new version...
,
Pick Zen2 for your CPU, the type of memory IC you have, if it's single or dual rank (again TB will tell you this but most 8GB DIMMs are single), and if you are on a 360/370,450/470/570 platform and the speed you want to attempt first ..

Then just click Fast, take a photo with your phone, and go into the bios and set alll the subtimings it gives you. I prefer the bios way but you can also modify them with Ryzen Master in windows if you want.
already doing this. Had to tweak the voltage a bit but down to 68.2 NS and about 53000mb/s read on Aida. Not sure if it's worth tweaking much more. Gone from 72.3 to 68.2 NS just by tweaking my sub timing's. Pretty happy. I do have Samsung b-die 8gb single rank sticks.
 

ccityinstaller

Supreme [H]ardness
Joined
Feb 23, 2007
Messages
4,241
already doing this. Had to tweak the voltage a bit but down to 68.2 NS and about 53000mb/s read on Aida. Not sure if it's worth tweaking much more. Gone from 72.3 to 68.2 NS just by tweaking my sub timing's. Pretty happy. I do have Samsung b-die 8gb single rank sticks.


Sorry I am an idiot and did not zoom in on your AIDA64 screenshot (mobile for the loss)...

You are running the same thing I am with my dual rank sticks, 3466 timings. If you got a good B die kit, you should be able to just bump your ram speed up to 3600 and keep the same timings since you have SR sticks.


Hopefully you will get lucky, but even if you do not, 3466 with those numbers is nice! I'm happy with it.
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
Linus did some testing as well with tighter timings. Below are his settings and results:
Screenshot_20190811-091724_YouTube.jpg

Screenshot_20190811-091739_YouTube.jpg

Screenshot_20190811-091751_YouTube.jpg

Screenshot_20190811-091802_YouTube.jpg


Little to no rains on CS:GO or R6:Siege which is not surprising considering the ultra high fps on those games aleady.

>>
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
Here is where I am stopping. Pretty damn happy. 3600 is boosting to 3960-3990 on all core work-loads. 4060-4100Mhz for single or light loads. Memory is stable at 3533 and FI is at 1767 1:1 ratio and tight CL 14 timings. Getting a solid 68ns for latency.

Current Ram Setup.PNG
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
Here is where I am stopping. Pretty damn happy. 3600 is boosting to 3960-3990 on all core work-loads. 4060-4100Mhz for single or light loads. Memory is stable at 3533 and FI is at 1767 1:1 ratio and tight CL 14 timings. Getting a solid 68ns for latency.

Yep, your tRFC is about as low as it will go no matter how much voltage you throw at it. Your results are especially impressive considering you have an R5 3600 and a B-450.

Congrats on the poor man's 8700k and I mean that respectfully.
 

tangoseal

[H]F Junkie
Joined
Dec 18, 2010
Messages
8,937
Yep, your tRFC is about as low as it will go no matter how much voltage you throw at it. Your results are especially impressive considering you have an R5 3600 and a B-450.

Congrats on the poor man's 8700k and I mean that respectfully.

Yup 67 to 68ns is the sweet spot for top core performance on the new chips.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,736
Tighten your Tras up to 28...what is your Trfc set to? Try 288...That will really speed things up and you should be able to do it since I assume you have b die. You should be able to run those timings at 34/66/3600 if you have SR sticks.

Ok this is where my memory settings are at. TRFC is set at 256.

Memory settings.png
 
Last edited:

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
This is what I have so far.

View attachment 180489

Your timings are really good but your latency seems a little high. It could be because you are running an x370. Try a higher freq, evening you have to raise timings a bit. Also, ensure you are at 1t in the bios for command rate. The software will always show 1t even if you are 2t.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,736
Your timings are really good but your latency seems a little high. It could be because you are running an x370. Try a higher freq, evening you have to raise timings a bit. Also, ensure you are at 1t in the bios for command rate. The software will always show 1t even if you are 2t.

I figured maybe being Dual Ranked Memory might have a effect on it as well. I will give it a try a bit later and see if I can raise the frequency. I will double check and make sure in the bios I am at 1t.
 

jmnovak

Weaksauce
Joined
Sep 14, 2007
Messages
127
I dont know how you guys are getting such high bandwidth. Im only pulling 54,200 read /29,200 write @ 3666 with these timings
.
Screenshot (4001).png

Noticed its the 3900X guys with high bandwidth. I know about the write issue because of the chiplet design. But I'm guessing that your read bandwidth is higher also because of it also.
 
Last edited:

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
I dont know how you guys are getting such high bandwidth. Im only pulling 54,200 read /29,200 write @ 3666 with these timings
.
View attachment 180533

Noticed its the 3900X guys with high bandwidth. I know about the write issue because of the chiplet design. But I'm guessing that your read bandwidth is higher also because of it also.

Your numbers look right. Yep, the single ccx guys are going to have the lower write speeds. If you want faster read speeds, try hitting cas 15 or 14 at those speeds. Your latency is actually the best I have seen.
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
I dont know how you guys are getting such high bandwidth. Im only pulling 54,200 read /29,200 write @ 3666 with these timings
.
View attachment 180533

Noticed its the 3900X guys with high bandwidth. I know about the write issue because of the chiplet design. But I'm guessing that your read bandwidth is higher also because of it also.
Damn thats some super solid latency. Only pulling 53,500 on write so don't feel bad :)
 

jmnovak

Weaksauce
Joined
Sep 14, 2007
Messages
127
3666mhz is soild stable ran Memtest to 1200% no errors. I can pull 3733mhz but get around 3 errors at the 500% mark on Memtest most likely would have to bump settings down or voltage up to get it.... but at that speed I get
Screenshot (4002).png

3800mhz is a wont boot at any settings so this setup is maxed at the 3733mhz.

Also this memory wont run CAS 14 past 3466mhz. thinking its because its 2nd rank (32GB). to get CAS 15 I have to turn off GearDown but that just makes memory unstable past 3400mhz.
 
Last edited:

jbc029

Limp Gawd
Joined
Mar 26, 2013
Messages
484
Finished tweaking the memory a couple days ago. This is my finalized 24/7 run. Still horse around with some things, but I've explored most of the chip and memory at this point.
3700x
ASRock AB350 Pro4, Bios 6.0 (much better bios than the 5.9, but they dropped official support for all first and second gen Ryzen if you update to this version)
OC'd the memory from 3200 to 3733
4.2 all core @ 1.4v



aida64_bench.png
mem_kit_used.png
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
Finished tweaking the memory a couple days ago. This is my finalized 24/7 run. Still horse around with some things, but I've explored most of the chip and memory at this point.
3700x
ASRock AB350 Pro4, Bios 6.0 (much better bios than the 5.9, but they dropped official support for all first and second gen Ryzen if you update to this version)
OC'd the memory from 3200 to 3733
4.2 all core @ 1.4v



View attachment 180791View attachment 180792
What are your current timings? Thats damn good latency if you are still using XMP timings...
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
Finished tweaking the memory a couple days ago. This is my finalized 24/7 run. Still horse around with some things, but I've explored most of the chip and memory at this point.
3700x
ASRock AB350 Pro4, Bios 6.0 (much better bios than the 5.9, but they dropped official support for all first and second gen Ryzen if you update to this version)
OC'd the memory from 3200 to 3733
4.2 all core @ 1.4v



View attachment 180791View attachment 180792

Gets 4.2 ghz on the 3700x and 3733 mhz with 67ns latency using a $90 board and $80 worth of memory.

Intel fanboy has left the chat.
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
It's fun to play who can get the highest or lowest number, but I would really like to know what numbers even matter and if Aida really is that great of an indicator for gaming numbers.

Just look at Steve's tuned 3000mhz kit vs thr unturned 3600 and 3800 mhz setups. It was often equal or better performance yet had higher latency and way worse write/read speeds.

I think speed/(cas*trfc) is the ticket for performance. Just not sure if that shows up as some kind of Aida or other software indicator.
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,213
It's fun to play who can get the highest or lowest number, but I would really like to know what numbers even matter and if Aida really is that great of an indicator for gaming numbers.

Just look at Steve's tuned 3000mhz kit vs thr unturned 3600 and 3800 mhz setups. It was often equal or better performance yet had higher latency and way worse write/read speeds.

I think speed/(cas*trfc) is the ticket for performance. Just not sure if that shows up as some kind of Aida or other software indicator.
I think at 1440P and 4K its not gonna be much of a difference. Let me look through some online game benchmarks and see which are more memory sensitive with built in benchmarks and see if I can compare different timings and speeds with my B-Die and compare performance.
 
Joined
Oct 11, 2018
Messages
527
https://openbenchmarking.org/result/1908150-HV-1907047HV73

Not exactly apples to apples since it's someone else's computer running the various ram speeds and I'm just there at the end with the one labeled cpu 0.1uv ddr4-3200

But it is kinda interesting that my 3200mhz ram is trading blows with theirs at 3800mhz. And my ram is old dual rank 16GB modules (64gb total) Better timings? better motherboard? i know my cpu isn't overclocked (i assume we're both using pbo). Doubt undervolting is helping as you can see in the temps, these are mostly single thread tests that dont impact temps much at all (for most tests)

Definitely some odd things going on in a few of the tests where it looks like they're only running single threaded when they should be fully multi-threaded. I'll have to dive into why that is.

edit:
Looks like software versions may be playing a part in some of the more drastic test anomalies. Will have to find someone running a more current OS.

Still though. just goes to show how tiny changes in software can completely overwhelm the significance of expensive hardware choices.
 
Last edited:

jbc029

Limp Gawd
Joined
Mar 26, 2013
Messages
484

Attachments

  • 20190812_225719.jpg
    20190812_225719.jpg
    622.5 KB · Views: 0
Last edited:

Concillian

Limp Gawd
Joined
Jun 14, 2005
Messages
270
It's fun to play who can get the highest or lowest number, but I would really like to know what numbers even matter and if Aida really is that great of an indicator for gaming numbers.

Just look at Steve's tuned 3000mhz kit vs thr unturned 3600 and 3800 mhz setups. It was often equal or better performance yet had higher latency and way worse write/read speeds.

I think speed/(cas*trfc) is the ticket for performance. Just not sure if that shows up as some kind of Aida or other software indicator.

tRFC is big, for sure, and therefore tRC is too, and tRC is based on tRAS and tRP, and tRAS is based on tCL and tRCD... all of these are related in some way, so as you lower primaries, you can lower tRFC too.
tFAW is good and therefore tRRDS
tWR is important for writes, but writes aren't like defining for performance, but it helps.
tRDRDSCL and tWRWRSCL are a minor effect as well.

As far as AIDA goes, Ryzen is scaling with Read, Copy and Latency, but you're correct that AIDA is not a direct indicator at all. I think with a particular setup, AIDA will improve with gaming improvements, but you can't necessarily take AIDA numbers from one setup and extrapolate how another setup will perform based on it's AIDA numbers.

The subtimings aren't doing much for latency, but they help read and copy scores quite a lot, and we see performance benefit in all the various gaming benchmarks.

The other factor is that Dual Rank memory sticks should perform significantly better than Single Rank as long as you don't give up too much latency/frequency moving to harder to drive sticks (or 4 sticks instead of 2). For the most part at this point in time 8GB sticks are almost all Single Range and 16GB sticks are almost all Dual Rank. Some legacy 8GB sticks may be Dual Rank, but these are basically unavailable new.
 
Last edited:

Riccochet

Fully [H]
Joined
Apr 11, 2007
Messages
24,060
What were you using (timings wise) to get that 108 second run with the reported 3600 memory speed?

Just curious since that's a pretty big difference between 108 (and reported 3600ram) and the current 125 (reported 3600 ram as well)..

I have no idea, actually. lol No way XMP timings for 3600 would have done that. Not at 19-20-20-20-36.
 

jmnovak

Weaksauce
Joined
Sep 14, 2007
Messages
127
Well after tweaking timings I dont think getting much more than this out this Gskill dual rank stuff.


Screenshot (4037).png
 

jhatfie

[H]ard|Gawd
Joined
Mar 19, 2003
Messages
1,472
Have only messed with my memory a bit. Running 4x8GB Corsair RGB Pro DDR4-3466 (32GB). It is B die, but lower tier stuff. Totally cannot run the recommended 1T settings in the DRAM calculator, cannot even boot (maybe a issues with 4 modules?). Still cannot complain, got them almost 2 years ago for my previous build.


memory_3600.PNG

cinebench.PNG
 
Last edited:

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,801
Have only messed with my memory a bit. Running 4x8GB Corsair RGB Pro DDR4-3466 (32GB). It is B die, but lower tier stuff. Totally cannot run the recommended 1T settings in the DRAM calculator, cannot even boot (maybe a issues with 4 modules?). Still cannot complain, got them almost 2 years ago for my previous build.


View attachment 181681

Possibly the fastest 32gb memory setups here. Beast of a machine you have there.
 
Top