Worth a change in order to reduce latency?

[Firegod]

Hey all

So I currently have 32GB of Kingston HyperX Predator 3200Mhz (CAS16) (2x16GB). From a bit of research online this appears to have a latency of 10.08ns. I have overclocked the memory and by increasing the clock speed I happily run at 3600Mhz with CAS17, which has reduced the latency to 9.52ns.

Having a look online I could change to 32GB of Corsair Dominator Platinum 3600Mhz (CAS14) (2x16GB). This would decrease the latency (compared to my current memory) to 7.84ns. (I can't get the HyperX to run at that frequency with that low a CAS value).

Do we think it be something that could help with general system responsiveness and smoothness? I guess any reduction in the time it takes to fetch and retrieve data from the CPU / Memory is a good thing. If I can sell my current memory, it would hopefully only cost about £100 to make the switch.

£100 wasted or as enthusiasts.... a welcome gain?

Thanks.

 

[Firegod]

Hi all sorry for the delay in replying - interesting thoughts, and all welcomed.

I have been doing some more reading and there is a benefit to be had, albeit marginal gains. And as people have mentioned for gaming specifically, I'd struggle to see any benefit when gaming at 1440P and above. 1080P does see some nice improvements it must be said (in both average and lower percentile results). I'm thinking it is more of a luxury to at this moment in time, considering everyone's comments.

I mean for exmaple my main OS SSD is PCIE3.0 but the B550-F would accept a PCIE4.0 M.2 in the main slot. I wonder if PCIE4.0 M.2 would show better gains than doing the memory?

@Guest2 as per sig really, so an 8-core 5800X on a B550-F mobo with a 6700XT Fighter 12GB GPU. Gaming at 1440P. As mentioned I have a PCIE3.0 M.2 SSD for OS and my games are on a secondary PCIE3.0 M.2. Current memory is 3200Mhz (clocked at 3600Mhz) running CAS17 (17-20-20).

It was just when I saw the memory for £194 and thought hmm if I could get maybes 80-100 for my current memory, it might be a tempting idea and not too painful on the old wallet.

Cheers.

 

[Firegod]

Thanks @Minstadave and @Guest2. For info here is what I'm using at the moment which has been stable for many months now.

My timings are the second column of entries compared to the "safe calc". This is with my XMP import as DRAM calc doesn't have any option for Hynix J-Die that I can tell specifically.

I think what I have is setup pretty accurately - doubt I can squeeze much more out of it? DRAM Voltage is at 1.39v. I'm also running Load Line Calibration at Level 3 with 120% on the Current Capability with an Optimised phase.

 

[Firegod]

Stable just with general use & gaming or running stress tests? One would argue its 'stable' if you spend all your time only doing one task, like web browsing. Another person would only say it's stable if it passes rigerous stress tests. There's no right or wrong IMO

Personally I rigerously test with stress tests so my system doesnt crap out on me during a game which i'm playing in a team.

On Ryzen DRAM calc, select MEMbench and then change membench mode to default, then run it. It should take 200 odd seconds to complete.

This is 3800 C14

Yeah I did a few tests, I edit video and music as well as gaming. The video stuff is quite intensive and I've not had any wobbles or crashes. Just ran the MEMBench part from DRAM Calc:

I might see if I can up the voltage and tighten up any further timings. I'm set at 1.39v but I keep hearing that 1.45v is "safe" so that might help me out and then just stick with what I have for now.

@michty_me typical the Corsair memory I was looking at is now out of stock - must be a sign. Don't do it!

 

[Firegod]

Sorry this thread has slightly gone off the original topic, that's my fault. But just wanted to say I've bumped DRAM voltage from 1.39v to 1.4v in the BIOS and set CLDO VDDP to 0.95v. I have managed to then correct my tRC value (which I just noticed). Not sure why it was 83, but this is now correctly set to 77 (tRC=tRAS+tRP) and tCL is now 16 (down from CAS17). Woo. It is Windows stable at least (I've yet to try anything else yet) and I re-ran the DRAM Calc Membench. Time dropped from 247.23 pictured earlier to 241.96 and latency from 64.7ns to 63.1ns. A small improvement.