AMD Ryzen 3000 Builders Thread

Page 100 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
I think it's the infinity fabric that is the problem. It's like the coffins corner when flying above 47,000ft. It gets very touchy and somewhat unstable. I think @ 3733mhz you will be able to run those same timings without any instability.

There also could be a FCLK hole in memory controller/uEFI.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Does anyone see anything wrong with this as far as memory settings/timings??? Anything I should poke around with to improve?

Don't see anything obviously wrong. Also, your MT score is pretty good! Several 3950X chips I've seen benched in reviews won't do more than 9200 MT (or so).
 

YBS1

Golden Member
May 14, 2000
1,945
129
106
Don't see anything obviously wrong. Also, your MT score is pretty good! Several 3950X chips I've seen benched in reviews won't do more than 9200 MT (or so).
I've went over 10K already, but it was with manual overclocking. It would be nice if there was a way to allow boost while setting a manual minimum all core overclock.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
I think it's the infinity fabric that is the problem. It's like the coffins corner when flying above 47,000ft. It gets very touchy and somewhat unstable. I think @ 3733mhz you will be able to run those same timings without any instability.
Maybe I missed something in from the DRAM timing/voltage sheet. I'll redo 3733 Saturday. I'll shut up if I can hit that with good timings.
 

Hans Gruber

Platinum Member
Dec 23, 2006
2,092
1,065
136
Maybe I missed something in from the DRAM timing/voltage sheet. I'll redo 3733 Saturday. I'll shut up if I can hit that with good timings.
The other thing I am thinking. Because your kit is a 3200mhz kit with B-die ram. That could be an issue. I know it makes no sense but B-die is heavily binned. You would think they would run high clocks by slightly loosening the timings but that doesn't always work. Where as with Hynix stuff, the sky is the limit but with timings nowhere close to B-die. The only caveat I will add is the Hynix has to be binned 3600mhz. I am guessing B-die may be the same. The golden chips.

Two things, run your kit for a day or two at what has been stable. Then overclock again. The other is to try loosening the timings even to Cas 16 @ 3733mhz and then at 3800mhz. That can establish a stable computer and then OC again. I lack the patience myself for that. For me it got so bad with hard crashes that I had to find an old computer with a jumper to properly flash my CMOS. Lastly, do not use XMP when OCing. Manually type in your timings for the 14-14-14-32 or whatever you are running. Auto for the rest.
 

YBS1

Golden Member
May 14, 2000
1,945
129
106
Have you used Aida64 and checked latency and bandwidth? Pushing beyond the calculator settings can cause performance regression.

My 3200 cl14 kit is running at 3800 cl15 which is beyond the calculators suggested settings. Still playing around with it, but I'm probably tapped out already.
No, I don't have that. I thought I had read the free version doesn't have access to the memory benchmark. It seems the other somewhat popular memory bench is MaxxMem2, I ran that and it showed an overall score of just a hair under 38GBps (Sorry 37GBps). I don't remember the individual score breakdowns. Latency showed as 62ns.
 
Last edited:

YBS1

Golden Member
May 14, 2000
1,945
129
106
Another thing, on the other system (The 3900X with the Asus TUF X570), that has been an adventure after first dabbling with the ASRock Aqua. The Aqua was pretty much a breeze to relatively get working the way I wanted. The Asus however screamed right out of the gate performance wise (first Cinebench R20 run was over 7400), but then I noticed the ridiculous voltages it was pushing. The scariest of which was the SOC pushing like 1.36, took me a bit to figure out how to get that under control. It seems Ryzen Master and the Asus bios fight over settings in a way that the Aqua doesn't.

Next up and this may be common knowledge to the more veteran Ryzen crowd but I found odd....While attempting to curb the crazy voltage (~1.45) it was feeding the cpu while NOT under a load, I figured I'd just throw a negative offset on it to take care of that. That brought it down to a more reasonable 1.32ish, and the all core boost while under load remained the same, but the performance dropped dramatically. Down to about 6200 in CB R20. Is that a known bug with using offsets?
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
Next up and this may be common knowledge to the more veteran Ryzen crowd but I found odd....While attempting to curb the crazy voltage (~1.45) it was feeding the cpu while NOT under a load, I figured I'd just throw a negative offset on it to take care of that. That brought it down to a more reasonable 1.32ish, and the all core boost while under load remained the same, but the performance dropped dramatically. Down to about 6200 in CB R20. Is that a known bug with using offsets?

It's called clock stretching. Decrease your negative offset until the performance returns.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
I'm pretty sure my b-die is maxed out. I'm unable to boot 1:1 beyond 3800MHz. I haven't found a way to decouple and not drop into 1:2 mode on my motherboard. I can boot up 3866 CL16, but the divider issue made me revert back. I didn't try anything higher as I figured it's a waste of time running in 1:2 mode.

3800 MHz_Tweaked_5.png

Viewing this charts maximum Thereotical maximum bandwidth it looks like I'm sitting at roughly 95% read and 100% write speed. Latency looks to favor running a fixed all core overclock.


DDR4 Bandwidth.PNG

Chart is from the following link.

 

Zstream

Diamond Member
Oct 24, 2005
3,396
277
136
But what is the actual payback on memory oc?

There is a curve, but it’s pretty simple. If you’re at 3200 speeds at cl18, you get the best perf/dollar. If you want the best you need to go to higher, but the percentage of performance decreases.

It all depends on what persona you are. Money is no object? Perf/dollar kind of person, or budget building.
 

gdansk

Golden Member
Feb 8, 2011
1,988
2,357
136
Perhaps I am asking a dumb question.

I have two kits of memory:
4x 8GB 3600MHz Corsair Vengeance LPX (CL18)
4x 16GB 3600MHz Corsair Vengeance LPX (CL18)
My CPU is a Ryzen 3700X and my motherboard is an MSI B350M Mortar.

When I run Prime95 on the first kit (total 32GB) it works fine for hours.
On the second kit, it fails rather quickly until I lower the memory clock to 3200MHz.

Is it possible my motherboard cannot support 64GB at 3600MHz or is it more likely the memory is defective?
My brother has a B450M motherboard and a 3600X, I should have him test this memory too.
 

Det0x

Golden Member
Sep 11, 2014
1,027
2,953
136
I'm finally done with my build and everything is working great (i think)
Managed to squeeze my 3950x upto 4741mhz single core and get the memory latency down to ~65ms :)
In Cinebench R20 i'm getting 9550 multicore and 535 singlecore scores without manual OC

finale.png
 
Last edited:

Hi-Fi Man

Senior member
Oct 19, 2013
601
120
106
Does anybody have any experience with Micron 19nm "E-die" RAM on Zen 2? I'm trying to run 2x16GiB @ 3,400MT/s using timings from the DRAM calculator but I'm getting random crashes every once and a while. I'm not sure where to start on trying to stabilize this RAM. I've tried raising voltage but no dice. What's the first timing(s) I should start loosening?
 

sdifox

No Lifer
Sep 30, 2005
94,687
14,935
126
Does anybody have any experience with Micron 19nm "E-die" RAM on Zen 2? I'm trying to run 2x16GiB @ 3,400MT/s using timings from the DRAM calculator but I'm getting random crashes every once and a while. I'm not sure where to start on trying to stabilize this RAM. I've tried raising voltage but no dice. What's the first timing(s) I should start loosening?

I think my Ballistix Elite are E Die but I am just using stock xmp 3600 16-18-18
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Does anybody have any experience with Micron 19nm "E-die" RAM on Zen 2? I'm trying to run 2x16GiB @ 3,400MT/s using timings from the DRAM calculator but I'm getting random crashes every once and a while. I'm not sure where to start on trying to stabilize this RAM. I've tried raising voltage but no dice. What's the first timing(s) I should start loosening?
As shown in the screen shot above your post, get the Ryzen Dram calculator. You'll be able to plug in the what you want to hit (E-die should do 3600 MT/s, but with relaxed setting, especially with 32GiB.
 

Hi-Fi Man

Senior member
Oct 19, 2013
601
120
106
As shown in the screen shot above your post, get the Ryzen Dram calculator. You'll be able to plug in the what you want to hit (E-die should do 3600 MT/s, but with relaxed setting, especially with 32GiB.
As stated in my OP, I'm using the calculator. It doesn't seem particularly suited to this Micron RAM though.
 

Hi-Fi Man

Senior member
Oct 19, 2013
601
120
106
Timings.png

Currently I'm seeing if tFAW at 24 will prove stable. Termination Impedances are all on auto and so are the tertiary timings. V2 basically raises tRCDRD to 20 and tightens the other primaries to 16. I read that tRCDRD needs to be rather loose on these ICs; I figured 18 was loose as it is at 3,400 with 1.385v but maybe not.
 

sdifox

No Lifer
Sep 30, 2005
94,687
14,935
126
There is a curve, but it’s pretty simple. If you’re at 3200 speeds at cl18, you get the best perf/dollar. If you want the best you need to go to higher, but the percentage of performance decreases.

It all depends on what persona you are. Money is no object? Perf/dollar kind of person, or budget building.


Haven't had a desktop since Opteron 165. I doubt I'll notice any memory oc gain. It's working as speced. Good enough for me.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
The uEFI version I'm running now seems to be working much better than the versions I tried when I wanted to test all core overclocks on my 3700x.

I was trying to bust the 63ns latency barrier in Aida64 so I had to play around with core clocks to accomplish the task.

3800 MHz_Tweaked_6.png

Once I found what it took I decided to retry the all core overclocks which look to higher and more stable than the old versions of the uEFI that I tested.

Testing a 4300/4325 all core overclock as I didn't really want to run a higher loaded vcore.

4300_4325 all core.PNG

See anything scary in the above that would make you not want to use the clocks as a option? I forgot to set the fan curve on my AIO cooler, but it looks like the temps aren't too bad with what is shown.

Blender.PNG
 
  • Like
Reactions: lightmanek and Ajay

mopardude87

Diamond Member
Oct 22, 2018
3,348
1,575
96
Hia
I got the tomahawk and 3900x. First one died after 48 hours of handbraking.
Second holds the same without problems.
I am not ocing. Without oc its completely ok.
And got it for 95eur lol

Hardware unboxed got a stock 3950 working fine as well even if vrms hit like 84 cel or so. B450 wise this board may be the only one capable reliably.

Going with the 3700x unless a 3700 comes out by March when I intend on this upgrade. With luck 4000 series is drop in or 3950 is heavily discounted in 2021. The 3700x should be sufficient though if not overkill but I think next gen games will make it a very viable purchase.
 

Noid

Platinum Member
Sep 20, 2000
2,373
183
106
@ YBS1

There seems to be a trade off with Latency and bandwidth. ( for me )
Thats the 1st 60K result I have seen.
( I assume you have 2 sticks of 8GB )

have been breaking below 62ns consistently now.
However bandwidth stuck around 57K.

Then ,,, a random score of 58.3K with latency of 62.7

I find results of all benchmarks fluctuating.
( with no changes to BIOS or VC settings )
Making it very hard to " dial in " tweaks.

My AIDA64 latency dropped more with a CPU boost clock ~4615 Mhz
( and my best 3Dmark score )
However, I cant duplicate that manual OC over 4600+ Mhz
( not yet )

I'm puzzled that the AUTO setting is giving me a boost over 4600 Mhz ( randomly )
( i think throttling the CPU temp at 80c on air )
 

Attachments

  • 61.7ns_AIDA64.JPG
    61.7ns_AIDA64.JPG
    84.3 KB · Views: 9
  • 58.3K_AIDA64.JPG
    58.3K_AIDA64.JPG
    86.7 KB · Views: 9
Last edited: