5600x Ram Overclocking

[KezzaMcFezza]

I recently upgraded from a i5 6600k to a 5600x and i had my 2400mhz corsair lpx running at 3000mhz with no issues in memtest and no crashes for years, and then once i upgraded to my 5600x i cant get my ram past 2666 even at 1.4 volts and i was wondering if anyone knew why the 5600x is so poor for ram overclocking?

 

I also noticed the latency on the intel system is a bit lower compared to the AMD one.

 

Intel System

 

AMD System

 

 

[Levent]

8 minutes ago, KezzaMcFezza said:

why the 5600x is so poor for ram overclocking?

It isnt. I got R3 3100, R5 3600, R7 5800X all able to handle 4 dimms at 3800mhz no problem.

[WereCat]

Did you change timings or just set the speed and left timings in Auto? 

 

I see you use looser primary timings with T2 on Intel while tighter with T1 on AMD.

 

Not that primary timings matter that much anyway besides stability and Aida64 latency test which is biased towards tight primaries. 

On AMD you'll always have more latency due to how it works with FCLK/MCLK/UCLK.

[RONOTHAN##]

Try a different BIOS revision (newer isn't always better), it's fairly common for there to be BIOS revisions with broken memory overclocking on AMD boards, especially if you're using older A0 based memory sticks like you're likely doing if the memory is carried forward from a Skylake based system. 

 

Another thing to try would be to set the SOC voltage to 1.15V. This is the voltage that the memory controller runs off of, and the vast majority of chips can achieve their max memory clocks with an SOC voltage between 1.1 and 1.2V, so 1.15V should be close enough for most CPUs. 

 

As for the latency, the memory system on these CPUs is completely different than older Intel. Even modern Intel is completely different than older Intel. The Aida memory latency test peaked with Kaby Lake and has been getting higher ever since due to the larger and slower ring clocks achievable by the newer chips. The Aida latency test is terrible for memory performance though, and generally should only be used to compare the same CPU clock speeds (this test is affected fairly heavily by CPU clocks) and the same CPU as cache design also influences this number. 

 

5 minutes ago, Levent said:

It isnt. I got R3 3100, R5 3600, R7 5800X all able to handle 4 dimms at 3800mhz no problem.

Can confirm, my 5900X will even do 4800MT/s if I desync the FCLK (only on 2 DIMMs though). You run into BIOS, FCLK, and motherboard limits well before you run into the IMC limits on those chips. 

[KezzaMcFezza]

35 minutes ago, WereCat said:

Did you change timings or just set the speed and left timings in Auto? 

 

I see you use looser primary timings with T2 on Intel while tighter with T1 on AMD.

 

Not that primary timings matter that much anyway besides stability and Aida64 latency test which is biased towards tight primaries. 

On AMD you'll always have more latency due to how it works with FCLK/MCLK/UCLK.

on the intel system i used auto timings as it responded well to just cranking the voltage and clock. but the amd system i have tried alot of timings and i have settled with 2666 with these timings which are tighter than the intel system for sure.

[KezzaMcFezza]

35 minutes ago, RONOTHAN## said:

Try a different BIOS revision (newer isn't always better), it's fairly common for there to be BIOS revisions with broken memory overclocking on AMD boards, especially if you're using older A0 based memory sticks like you're likely doing if the memory is carried forward from a Skylake based system. 

 

Another thing to try would be to set the SOC voltage to 1.15V. This is the voltage that the memory controller runs off of, and the vast majority of chips can achieve their max memory clocks with an SOC voltage between 1.1 and 1.2V, so 1.15V should be close enough for most CPUs. 

 

As for the latency, the memory system on these CPUs is completely different than older Intel. Even modern Intel is completely different than older Intel. The Aida memory latency test peaked with Kaby Lake and has been getting higher ever since due to the larger and slower ring clocks achievable by the newer chips. The Aida latency test is terrible for memory performance though, and generally should only be used to compare the same CPU clock speeds (this test is affected fairly heavily by CPU clocks) and the same CPU as cache design also influences this number. 

 

Can confirm, my 5900X will even do 4800MT/s if I desync the FCLK (only on 2 DIMMs though). You run into BIOS, FCLK, and motherboard limits well before you run into the IMC limits on those chips. 

ill try the tips you suggested and ill see if i can break the 3000mhz barrier

[WereCat]

Just now, KezzaMcFezza said:

on the intel system i used auto timings as it responded well to just cranking the voltage and clock. but the amd system i have tried alot of timings and i have settled with 2666 with these timings which are tighter than the intel system for sure.

Auto timings depend a lot on what motherboard and bios you have. You're better off to tune it manually

[SupaKomputa]

7 minutes ago, KezzaMcFezza said:

on the intel system i used auto timings as it responded well to just cranking the voltage and clock. but the amd system i have tried alot of timings and i have settled with 2666 with these timings which are tighter than the intel system for sure.

Ryzen 5000 is picky with memory chips, i can confirm that 2666mhz micron chips worked as using my previous 2133mhz samsung c die (3600mhz w/ 3600x) won't even boot.

So far i can go 4200mhz with the micron but i settled to 3600mhz for daily usage.

Updating firmware might support more chips.

[Somerandomtechyboi]

Spoiler

My ddr3 is faster than your ddr4 =)

 

Jokes aside maybe bios is just bugged so try updating the bios, the latency is inherent because of fclk multiplie chiplets and whatnot compared to a monolithic die, if updating to the latest bios doesnt work just keep changing bios revisions till it works

 

For overclocking the rams once you fix the bios bug send a thaiphoon burner pic to know what ics were dealing with, if its two 2x8 kits and not a proper 4x8 then send another pic with a different ic if the rams dont have matching ics