My speed test of 3600MHz CL16 vs 3800MHz CL18 (need input)

[Nena Trinity]

DDR4-3600MHz CL16-18-19-37-55

 

vs

 

DDR4-3800MHz CL18-19-19-38-57

 

Its quite safe to say the 3600MHz is faster but the 3800MHz will gain more from FCLK so what would work better?  

[emosun]

i dont think either setting will result in any sort of noticeable gains outside of benchmark numbers

[Guest]

The timings are just too loose to make anything of it. 

As stated, daily use, barely any difference.

 

CL14's at least on 3600mt/s and drop some latency.

[Nena Trinity]

17 minutes ago, emosun said:

i dont think either setting will result in any sort of noticeable gains outside of benchmark numbers

I suspect that but I think the results matter more depending on the application...

12 minutes ago, ShrimpBrime said:

The timings are just too loose to make anything of it. 

As stated, daily use, barely any difference.

 

CL14's at least on 3600mt/s and drop some latency.

These are not Samsung B-die its Hynix CJRs in dual-rank... 

[Guest]

1 hour ago, Nena Trinity said:

I suspect that but I think the results matter more depending on the application...

These are not Samsung B-die its Hynix CJRs in dual-rank... 

Ah well then CL 15-17-17-17 maybe. Give a little soc voltage. Hhm 1.225v.

[Nena Trinity]

13 minutes ago, ShrimpBrime said:

Ah well then CL 15-17-17-17 maybe. Give a little soc voltage. Hhm 1.225v.

Gear down mode will make it 16-17-17

[Guest]

1 hour ago, Nena Trinity said:

Gear down mode will make it 16-17-17

Disable it in the timings menu to post odd number cas latency.

[Nena Trinity]

2 minutes ago, ShrimpBrime said:

Disable it in the timings menu to post odd number cas latency.

System refuses to boot with even XMP if its off...

[Guest]

39 minutes ago, Nena Trinity said:

System refuses to boot with even XMP if its off...

That Ryzen memory controller doesn't like them Hynix modules.

[Nena Trinity]

6 minutes ago, ShrimpBrime said:

That Ryzen memory controller doesn't like them Hynix modules.

Well I wanted sometihng that was dual-rank to replace my 3200MHz CL16 kits of Samsung C-die so I guess I should have tried 2x16GB Crucual? They were no way to check how many ranks they had... Could be either 16Gbit Rev B or 8Gbit Rev E! >w<

[Guest]

10 minutes ago, Nena Trinity said:

Well I wanted sometihng that was dual-rank to replace my 3200MHz CL16 kits of Samsung C-die so I guess I should have tried 2x16GB Crucual? They were no way to check how many ranks they had... Could be either 16Gbit Rev B or 8Gbit Rev E! >w<

See the thing is, the dual ranking doesn't make up for the performance if 16gb single ranked setups.

You see this more with competitive benchers that will often times run 2:1 but memory at 4000mhz CL12-12-12-12 and 12-11-11-11, on a set of B-die and very high voltage. For some reason, SK Hynix on any AMD, or even Intel is just not the way to go. Be it single or dual ranked. 

 

What you have probably doesn't scale with voltage. If I'm on my 8700K, I start my memory frequency and tweaking between 1.50v and 1.60v

 

Any how you can try a little more IMC voltage. If I'm not mistaken, it doesn't turn color until about 1.30v, so if you're under 1.2v, you have a little headroom yet  

[Guest]

Here's where Zen+ sweet spot is. 

2700x 3600mhz B-Die. 1.45v

1.225v soc

[Nena Trinity]

15 minutes ago, ShrimpBrime said:

See the thing is, the dual ranking doesn't make up for the performance if 16gb single ranked setups.

You see this more with competitive benchers that will often times run 2:1 but memory at 4000mhz CL12-12-12-12 and 12-11-11-11, on a set of B-die and very high voltage. For some reason, SK Hynix on any AMD, or even Intel is just not the way to go. Be it single or dual ranked. 

 

What you have probably doesn't scale with voltage. If I'm on my 8700K, I start my memory frequency and tweaking between 1.50v and 1.60v

 

Any how you can try a little more IMC voltage. If I'm not mistaken, it doesn't turn color until about 1.30v, so if you're under 1.2v, you have a little headroom yet  

Well Buildzoid seems to be saying 3800MHz in Dual Rank is best for a Ryzen 3000... 

9 minutes ago, ShrimpBrime said:

Here's where Zen+ sweet spot is. 

2700x 3600mhz B-Die. 1.45v

 

RAM and FCLK settings vs performance on Ryzen 3000 - YouTube

https://www.youtube.com/watch?v=nugwAOvijHQ

[Guest]

Just now, Nena Trinity said:

Well Buildzoid seems to be saying 3800MHz in Dual Rank is best for a Ryzen 3000... 

RAM and FCLK settings vs performance on Ryzen 3000 - YouTube

https://www.youtube.com/watch?v=nugwAOvijHQ

 

I seen that. Of course keeping 1:1 ratio. 

But your timings and latency is Maybe not bad, but not great.

 

That's where a lot of Intel's performance comes from 40ns latency and high frequency. 

I'm running 1933mhz (effective) 14-14-14-24-444 trfc 2Tcr 1.55v 

Would you like screen shots? 

[Nena Trinity]

3 minutes ago, ShrimpBrime said:

I seen that. Of course keeping 1:1 ratio. 

But your timings and latency is Maybe not bad, but not great.

 

That's where a lot of Intel's performance comes from 40ns latency and high frequency. 

I'm running 1933mhz (effective) 14-14-14-24-444 trfc 2Tcr 1.55v 

Would you like screen shots? 

If I understand this then 3800DR at 20-22-22-40-1T will be at 54k READ and CL14-15-15-32-1T 56k READ and similiar trends seems to be the case my kit should now be at 54k to 55k READ in that case maybe I should ignore latency and go all in on bandwidth? 

[Guest]

1 hour ago, Nena Trinity said:

If I understand this then 3800DR at 20-22-22-40-1T will be at 54k READ and CL14-15-15-32-1T 56k READ and similiar trends seems to be the case my kit should now be at 54k to 55k READ in that case maybe I should ignore latency and go all in on bandwidth? 

Depends on where you look for bandwidth. 100mhz cpu frequency increase will net more memory bandwidth, you don't always have to OC the memory to obtain it.

[Nena Trinity]

3 minutes ago, ShrimpBrime said:

Depends on where you look for bandwidth. 100mhz cpu frequency increase will net more memory bandwidth, you don't always have to OC the memory to obtain it.

Well I think I will leave it at 3600MHz for now, I got a feeling the IF clocks may affect temperatures more? If so going for faster RAM with more PBO headroom will gain more gaming FPS? :3 

 

The stock XMP profile is 17-19-19-39 but due to GEAR DOWN its 18 and since it was 18 I curiosity decided to see how far MHz went and it hit up to 3800MHz just fine anything over is a waste on Ryzen,,, I also after decided to see "How far down does timings go on stock?" Well result there is 16-18-19-37 the last diggit can probably be pushed more down but eh stability good.  

[Guest]

1 minute ago, Nena Trinity said:

Well I think I will leave it at 3600MHz for now, I got a feeling the IF clocks may affect temperatures more? If so going for faster RAM with more PBO headroom will gain more gaming FPS? :3 

 

The stock XMP profile is 17-19-19-39 but due to GEAR DOWN its 18 and since it was 18 I curiosity decided to see how far MHz went and it hit up to 3800MHz just fine anything over is a waste on Ryzen,,, I also after decided to see "How far down does timings go on stock?" Well result there is 16-18-19-37 the last diggit can probably be pushed more down but eh stability good.  

Will they run 16-16-16-16-36? It's a nice smooth timing set.

[Nena Trinity]

Just now, ShrimpBrime said:

Will they run 16-16-16-16-36? It's a nice smooth timing set.

Ummm to my understanding at 1.35v nope, maybe at 1.4v but I think Buildzoid at some point said timings arent scaling with voltage so I havnt bothered trying that also a bit worried about temps in the case at full CPU + GPU load. :3

[Guest]

Just now, Nena Trinity said:

Ummm to my understanding at 1.35v nope, maybe at 1.4v but I think Buildzoid at some point said timings arent scaling with voltage so I havnt bothered trying that also a bit worried about temps in the case at full CPU + GPU load. :3

Why not? You have a screen shot with 14-16-16-32 1T. 

Worth a shot, maybe try a command rate of 2T instead. You won't loose anything from it.

[Nena Trinity]

 

10600 with 2666MHz CL13-15-15-35 for comparison...