I have a pair of 16 gb samsung bdie ram rated up to 3200 that I have managed to clock up to 3800 CL16-15RDC-14RP with gear down mode on (seems like gear down mode is pretty much required for dual rank ram) at 1.42V. IF is running stable at 1900.
I have two questions, firstly what is the maximum safe voltage for Samsung BDie ram. I read conflicting stuff with one post claiming 1.6 is fine which I doubt. A few posts seem to indicate 1.5 is fine for day to day and the overall consensus is that bdie ram can handle more voltage. I am pretty sure that I could tighten up the timings a bit more with voltage if its safe.
Second question and perhaps most important what would benefit the 3900x more, slower speed but tighter timing or faster speed (including if) but slower timing.
Jag
Edit: Just to clarify, I did try searching for this but all I see is comparisons between 3200 and 3600 which is a much larger frequency delta.
According to LTT it’s a mixed bag. Sometimes 3600 CL14 is faster, sometimes 3800 CL16 is faster. For the most part though it seems that the difference is negligible.
Hi thanks for the excellent reply. One other question is that it looks like the voltage is fine for the ram but what about the IMC. Does dram voltage pass through it and if so how does it affect the IMC?
It looks like gskill will be releasing 3800 cl14 kits at 1.5v so I assume it would be safe as they wouldn’t release a kit that would damage a cpu? At least I would think that would be the case.
Awsome great so I could push the memory voltage to 1.5 if I was wanting to. Just outside of curiosity what is the max voltage for the soc if I need to bump that up. I have it set to 1.1 l. I believe the bios default is 1.2.
Doesn’t Ryzen 3000 series infinity fabric reduce its speed to 1/2 memory clock at or above DDR4-3733 or thereabouts?
If so, the gains you get from 3800 DDR4 clock may be outweighed by the loss in infinity fabric speed (which would presumably step down to 1900mhz or so).
edit:
re-read the OP, i see you have already mentioned the downclock to IF.
Given that Ryzen 3000 has huge caches (GAME CACHE!!!11), i’d be tempted to reduce memory speed to 3600 and run with tighter timings in order to maintain infinity fabric at 1:1 ratio.
As infinity fabric speed will impact core to core communication, and memory speed will be of diminishing returns in any case.
You’re only going to improve performance via higher memoryt clock for CPU cache-misses, which in theory with the huge caches should be “not often”. You’ll likely see improvement on memory benchmarks, but just be careful not to assume that a memory benchmark win will mean a real world use case win.
The big performance gain from faster memory with prior Ryzen CPUs is due to the boost to infinity fabric speed that was locked to memory. With Ryzen 3000 if you go beyond 3600, you’re going to kill IF speed.
If in doubt, benchmark of course. But AMD themselves list DDR3600 as the sweet spot, due to this specific issue (infinity fabric clock) IIRC.
Maybe this will change in future when we have memory that can reliably run at DDR4-7000 or whatever. IF that ever happens. But not today, i don’t think.
Not unless you have some massively niche workload that is very cache-unfriendly… with any normal workload you’re going to be hitting the cache a lot more often than not.
I had my memory at 3600 14-19-14-36-1T which gave a latency of 67.3ns, then i went to 3733 with 16-23-18-26-1T and that gave a latency of 65.3ns. The former was hard to get stable and required lots of voltage, the latter is pretty much stock except for ram which is at 1.42V and is rock solid. Both of these are with memory and IF synced up or 1:1.
I’d say it’s worth it to see how high you can go with your IF and then match your ram and work on timings. Some chips can even go 1900 on the IF but mine does not unfortunately…
This is with a cheap micro e-die kit though, but ryzen dram calculator makes it very easy to tighten timings.
I’d keep it as low as possible tbh since it eats into the power budget of your cpu which in turn will affect your boost speed. And with the dram speeds we’re seeing i don’t think we need something like 1.3v either, 1.1v is a good baseline for 3600 or 3800.
Sorry if I stated otherwise but I actually manually clocked the IF to 1900 as my 3900x can handle that no problem. So my IF is in sync with my ram 1900/1900
My set of G.Skill F4-3200C14D-32GTZR seems to be OCing beautifully. I was running stable with the 1900/1900 IF/Mem clock at 16-15-15-32 at 1.42 (Probably a tiny bit higher as my bios says one thing and HW Info says it’s higher a bit odly ryzen master reports what the bios does).
After m.meri pointed out that Samsung can work at 1.5 I up the voltage to 1.48 in the bios (1.5 according) to hwino and fingers crossed I have been running at 14-14-15-32. While higher voltage sounds like this will work fine?
Thank you again w.meri. you have bee most helpful. I think I will keep the soc voltage stock as it will impact thermal headroom as pointed out in the above post. Although it would be nice to see if I can use it to decrease the memory voltage from 1.5.
In my case my processor seems to want to hang out at 4.1 boost and I very rarely see it try to hit 4.6 and if it does it only does for a fraction of a second. So not sure if I want to add more heat into the mix.
Can someone recommend how to test my memory of? I have been using prime 95 with the largest fft and memtest 86.
My motherboard overvolts the memory as well. As for the 1.5v idk, i mean it’s in the specs but we’re not running the ram in spec. I read somewhere people recommend 1,45v max for daily use. Perhaps with the increased load from tighter timings and higher speeds it is a problem?
And in regards to the over volting that is interesting. I wonder if its intentional or just inaccuracies in the measurement of the voltage data. With that said I assume that could be a problem with ram kits like I listed above.