so i just upgraded my cpu…
currently its running on my old ram config of 4 mixed dims, 2x(2x8gig) 3000 same brand same model (differnet parts used). similar timings and its stable somehow.
im looking to increase to 64 gigs 4x16 would be my preferred config as ryzen 5000 benefits from having 4 dim configs. (increasing overall system performance by about 10% over 2 dim configs.).
i also see that ryzen 5000 can handle up to ddr 4000, depending on your cpu silicone quality, and can run the infinity fabric at 2000 t1 timings sometimes.
my question, is it worth the bother of trying 4000 over 3600 which is about 10bux/8 pounds cheaper.
but will run the infinity fabric at 1800 so an additional potential performance loss.
so 4000 and chance my luck with the silicone lottery or play it safe and go with 3600?.
anyone running a 4000 kit stable?..
how much fiddling?, gains worth it?.
Realistic expectations are 3733MHz/3800MHz IF, 1:1 sync for chiplet Zen 2/Zen 3 CPUs.
Anything above that is pure luck, so I won’t advise spending on higher speed kits if you’re just going to turn on xmp and call it a day, because it probably won’t work at 1:1 sync and you’ll lose performance if left untuned.
2 DIMMs per channel configuration can also reduce your max IF clock if your memory controller isn’t strong enough, so watch out for that if you’re going to stick 4 DIMMs onto your board. 2 sticks of dual rank 32GB sticks are a much safer bet.
5900x upgraded from a 2700x.
and at 140 for 64 gig… its not terrible.
@Mach3.2
yeah i heard 3733 was the max that the zen2 3000 cpu’s could handle as they top out at 1866 on the fabric…
im not considering 3733 as its latency spectrum is very narrow. meaning theres not much room for tightening timings.
@jode thanks for the heads up.
yeah it seems hit and miss as i said earlier but was hoping a couple of you guys had more luck.
You shouldn’t be aiming for 4000MT/s because your IMC might not be able to handle it with absolute stability all year round. So go to 3600MT/s DIMMs and then try your luck with overclocking to 4000. Performance difference is minimal unless your main applications are assorts of benchmarks.
Also go for two DIMMs of 32GB each, instead of four DIMMs of 16GB each. Much easier to overlock and rip the same benefit of four DIMMs of single-rank 16GB DIMMs.
In the past, Crucial with Micron Rev E is my go to DIMMs. But that’s gone. Some consumer brands still use Micron Rev E. Good luck hunting it down 'cos it’s hard as they never tell and even use the different make of chips in the same model of DIMMs.
So Hynix CJR might be more widely available and perhaps lightly better substitue for Micron Rev E. Very few DIMM vendors adverse CJR but mostly use some CJR parts. Anyway good luck hunting it down. As I’m on the same boat. That’s what I gathered so far.
I got a kit of 2*32GB Crucial 3200MHz 22-22-22 JEDEC Micron Rev.B kits, those are really troublesome to OC and I still can’t squeeze much out of it. Settled with 3533MHz 22-22-22 @ 1.4V y-cruncher all tests 8 iterations stable.
3600MHz refuses to post, so do any timings tighter than 22 ticks. It could very well be my dogshit IMC, so hard to say if I actually hit the ceiling with these green PCB DIMMs, but safe to say don’t expect much out of JEDEC sticks.
That’s a very big gamble if you ask me. I don’t think it’s worth fiddling with it on your main computer just for the sake of it.
It also highly depends on the IC in the RAM kit you’re looking at. If it’s anything but Samsung rev. B, Micron rev. E or SK Hynix DJR you’re gonna have a tough time. If I’m not mistaken these are the three “golden child” of overclockers and will give you the highest chance of getting to 4000 MT/s.
Absolutely not. It’s a big investment time wise and it may result in you having to tune the kit you bought to 3600 or 3800 MT/s because XMP won’t work. What’s worth spending on, in my opinion, is the lowest latency possible at 3600 or 3800MHz.
1800 MHz IF is already pushing it. To a safe limit, but still pushing it nonetheless. You’re gonna increase VSoc too and that’s always something that should be staying as in spec by AMD as possible. If you need 1.2V to run 4000 MT/s you’re already in the danger zone, which is not worth it.
thats generally the plan if 4000 isnt viable which by the sounds of it, it isnt.
i also looked a little deeper at the timings on offer. they werent great at cas20 t2…
it would effectively be slower than my current 3000…
so it looks like im going with 3600…
I’ve been losing my head around what RAM to use, speeds, latencies, voltages, tuning and so on. Really glad to have helped you not go down the same path of insanity!
tbh i dont mind tuning a pc…
always something ive done so fiddling with ram isnt new to me.
but yeah it is a rabbit hold i often send others down… while looking at the whole myself…
guess its time to dive in…
Oh no no, this is way before your thread! I was trying to figure out what 2x16GB kit to get for my system and even had to talk to r/overclockers. Ugh, so many bad people in there.
Samsung B dies are older tech, meaning lower density chips. If you want 64GB, then 4 sticks are a must. It might work at XMP or even be pushed a little beyond. Equally possible might not run at XMP at all. Depends on motherboards.
Dual rank Samsung B-dies are very hard to stabilise at 1T GDM off. 2T will be easier. I also think it might be close to impossible to run 4 sticks of it with reasonable frequency and latency (at least 3600MHz 15-15-15 1:1 sync).
I try to run the FLCK and memory in sync (coupled mode). So more then 3600Mhz isn’t really all that usefull.
for my 5950x I use a 4000 kit clocked to 3600 and IF to 1800. Works like a charm
Received my Hynix CJR 32GiB x 2 (rated 3600 18-22-22-42), replacing Micron Rev. E 8GiB x 2 (rated 3000 15-16-16-35). My little micro-ATX box is so much happier now.
My Micron Rev. E is one of the earlier production batch circa 2019. I mention this because I heard newer batches are apparently better quality. But based on what I have, the Hynix CJR performs better than the Micron Rev. E at the same clock, 3733. A huge surprise.
Rated spec wise the CJR sticks shouldn’t be able to beat the Rev E. The CJR is also denser die, 16Gbit vs 8Gbit (for Rev. E), dual ranks vs single rank (for Rev E), and quadruple the capacity (and harder to drive).
After overclocked to the optimal state…Latency is about 0.5ns (or 0.8%) higher for the CJR sticks. A pass for me since it’s harder to drive. Bandwidth is 1.9% higher in read for the CJR, 6.1% higher in copy. Thanks to true 1T command rate finally works with the same IMC and this is a huge surprise.
I also find Hynix CJR is easier to overclock. By that I mean error shows up very quickly when parameters are even minute distance away from the perfect point. This is a huge time saver. Micron Rev. E is perhaps resilient to abuse and errors take long time to show up, especially when nearing the perfect point.
I was originally afraid of trading clock for capacity. It turns out this upgrade is a win in every possible way. At the end of DDR4 life-cycle, something great and cheap is a sweet deal. This is the one I haven’t had for a long while. lol
