help me perform RAM memory overclocking because the niggers from /v/ chased me away because I went there because it’s dead here, the prodigal son has returnedI have an ASRock B650 LiveMixer, two sets of G.Skill Flare X5, DDR5, 32 GB, 6000 MHz, CL30, that is 4 sticks of 16 GB, and a Ryzen 7 7800X3D.Currently I have 3600 MHz CL40.On the ATX ASUS PRIME X670-P motherboard there was a ready preset that worked on all sticks at 6000 MHz CL30, so it is possible.well, I pasted this to them and that was enough, in English of course; I’ll add that I experimented with ready-made 6000 MHz factory presets from ASRock, but the computer doesn’t boot, I have to remove the battery to reset the BIOS.
I don’t even know myself what to change besides voltage, MHz, and CL — it’s too advanced for me.
>4 sticksit's over. you're not hitting 6000. Try 5600.
also you're mixing kits which is not good.
>>107889008>>107889023I need precise guidelines from a white man for every function, because when I set things myself it doesn’t boot and I have to remove the battery to reset the BIOS.I made a grid from 66 screenshots
>>107888952You've put yourself up for a world of hurt because 4 sticks of DDR5 is much harder to run than 2. You'll have to look shit up online, you will probably find a few people who have struggled with this same kind of setup but the vast majority of people would have listened to the advice to use 2 sticks on AM5.
>> 107889090 Well, that’s exactly what I wrote online — where else am I supposed to look for help, on Reddit? Give me a break.I wrote that on X670 they worked for me right away, and the AI wrote back to me that it’s a better motherboard topology or something, which is why I need help here, where someone knowledgeable understands every function in this advanced BIOS
>>107888952Memory overclocking is a very silly exercise unless you are trying to push for a world record on some epenis benchmark. It is waste of time. Running stuff at JEDEC-spec is more then sufficient. This is even true for silly gaming. You will have a lot less headaches.
>>107889008Konata... My little star...
I bought CL30 and 6000 MHz, so I’m not going to sit at 3600 MHz and CL40. I don’t want to set a world record, but what I have now is taking away a lot of FPS from me >>107889251#
>>107889339its not taking a lot of FPStake out one kit, enter bios, select XMP, reboot and retest. the difference will be impossible to measure in most games (provided one kit is enough RAM to run those games)
>>107889339That's speed and timing is not possible with 4 UDIMMs on AM5, LGA1700 and LGA1851. The memory controller and motherboard cannot handle it. You have to take a hit on clockspeed and latency if you want to drive 4xUDIMMs on those platforms. Lower portion of JEDEC-spec speed is the best it can do. >I have now is taking away a lot of FPS from me You will not notice the difference in practice outside of looking at numbers. You sure hell will notice the instability, crashes and lengthy memory training when you are trying to overclock the memory.
>>107889442how much harder is it realistically to run 4 single rank sticks vs 2 dual rank sticks? Cause my 14900k handles 6600 just fine on 96gb.
>>107889064remove the memory from DIMM_A1 and DIMM_B1. you will now be able to hit 6000MHz fine (probably even higher).The Ryzen 7800X3D is not capable to electrically drive 4 slots of memory at very high speeds. The memory controller is just garbage. You'll be lucky to hit 4800MHz that way. With a 8600G (monolithic apu) you may be able to hit 5600MHz because they have slightly better memory controllers, but that's it. They won't release chips with better memory controllers until Zen 6, and that's still at least a year away.Only way to hit 6000MHz or higher is to use 2 slots only.
>>107888952>4 sticks of 16 GBIf you care that much about memory speed, better sell them and buy a 2x 32 GB kit>>107889064Load the EXPO profileLower memory frequency until computer bootsRun some programs of your choice that heavily stress memory (e.g. y-cruncher, stressapptest, Memtest86+, Prime95, OCCT, HCI MemTest, Karhu)In case of errors/crashes, lower memory frequency furtherRinse and repeat until things are stableThen lower memory frequency slightly further still, just to be safeIgnore all other settings unless you're autistic
>>107889451its not hard, its luckyour 4 single rank sticks are statisitically less likely to match (meaning stable at the same clocks) than your two kitswhich is why you would be a 4-rank-kit if you want to run 4 ranks, because then the manufacturer will make sure the 4 bars will all run stable at the exact same clocks advertised.
>>107889451It is much more difficult. >Cause my 14900k handles 6600 just fine on 96gb.You are running them at Gear2 which is a 1:2 divider. AM5 is bottlenecked by the Infinity fabric.
>>107889251>Memory overclocking is a very silly exercise unless you are trying to push for a world record on some epenis benchmark.If you game on the iGPU then the FPS you get will rise 1:1 with the memory clock. Your best bet for that is the 8600G or 8700G which has a better memory controller and can hit 9000MHz on boards that come with 2 memory slots.
4 RAM sticks aren't ideal for DDR5, either stay at 5600mhz or go to 6400mhz while setting the RAM speed to gear 2.Gear 1 RAM is notoriously unstable on all AMD CPU's, idk why they still do that instead of copying Intel.In any case, Intel CPU's can OC RAM well beyond 6400mhz and stay stable.
>>107889150I mean more like trying to look for YT videos or already written forum posts / guides, made by people with this kind of uncommon setup. I think I remember Wendell from Level1Techs doing some shit with 4 sticks on AM5, he might have some relevant videos up and there might be relevant topics up on the forums.Also you might want to see if you have the latest BIOS version. If not check what has been updated, my AM5 mobo received shitloads of BIOS updates which claim improved memory compatibility since release.
>>107889451Intel chips have far better memory controllers.AM5 is complete utter fucking garbo when it comes to DDR5. The guaranteed speeds in the specs for 4 dimms is something like 4800MHz or even lower.
>>107889487>The memory controller is just garbage. No, the problem is that UDIMMs were never meant to go super-fast especially when driving 4 units. We are hitting the hard limits of it. That's why CAMM2 is coming down the pipe. Intel controller have to operate with dividers (1:2 or 1:4) to handle faster UDIMMs.
>>107889514>Intel chips have far better memory controllers.i guess, but ddr5 is a step down in all ways from ddr4. I used to run 8 ranks of ddr4 on rocket lake and it was plug n play 3600 no headaches. Trying to get 192gb of ddr5 stable on raptor lake at anything above 4800 was an unspeakable nightmare.
>>107889251It helps with some gaymes or with AI video generation (locally) if you have an hybrid PC for games + AI gens for fun.
>>107889499>4 RAM sticks aren't ideal for DDR5It works fine, you just have to take a hit on clockspeed/latency. This had never had changed for UDIMMs. The hit is simply larger as DDR5 is operating at very high clockspeeds. Running 4xUDIMMs on desktop sockets has always been a trade-off for capacity versus clockspeed/latency.
>>107889534You get a Threadripper/Xeon W for that using RDIMMs.
>>107889531DDR5 operates are much higher clockspeed DDR4. DDR4 had the same problem if you tried running them at extreme-end of the curve. DDR4-3600 is easy mode.
oh yeah and ddr5 memory controllers on intel are a lot more temperamental. This sweet spot voltage range shit is obnoxious. I miss Comet Lake where you just pump volts into it until it Just Fucken Works. Now you go above a certain range on vddq or sa and it just explodes. what?
>>107889513AM5 BIOS updates won't really help much as the problem comes for AMD's memory controller (on the CPU) and that it runs RAM at gear 1.
>>107889339You will not notice substantial difference, retard. You will not hit 6k MHz with 4 sticks, just overclock it to 4000 and call it a day.
>>107889565yeah ok and ddr4 ran at much higher clock speeds compared to ddr3 or ddr2 it doesn't justify the regression imo. Also this >>107889567
>>107889567>you put more voltage in it that it's rated for and it explodes>what?
>>107889552No you don't need to spend so much money on hardware if it's just for games and AI hobby.I have an i5 13600k + 32GB DDR5 6400mhz cl32 + an rtx 3070, the GPU will get changed next gen but that set up is more than enough to run modern games at 3440x1440p.A quick RAM OC on intel is very easy (plug and play generally), people OCing on AMD are just asking for trouble.
>>107889587i don't mean literally explodes, I mean it becomes unstable. Comet Lake didn't behave this way. I could go up to 1.45v sa if i wanted to. I'd have to oxidize the chip before it became unstable from more voltage.Also 10nm is supposed to have the same voltage tolerances as 14nm. It's an imc design issue im pretty sure than my cpu crashes at any imc voltage above 1.275v.
>>107889635>1.45vor even 1.6v. comet lake was built different.
>>107889438>>107889442>>107889487>>107889490Guys, I used to have the same knowledge as you — I don’t know where it comes from in the public space, probably Reddit. Until I saw with my own eyes that X670 on a 7800X3D boots four memory sticks at 6000 MHz CL30, all stable, using a ready-made preset. You learn something new every day.Now I have an ASRock and I want to repeat that
>>107889514>AM5 is complete utter fucking garbo when it comes to DDR5. The guaranteed speeds in the specs for 4 dimms is something like 4800MHz or even lower.It's 3600 MHz and until 14th gen Intel also only guaranteed 3600 MHz with 4x dual rank sticks (4000 MHz with 4x single rank sticks, but that's not very relevant unless you're a retard like OP)Even with Arrow Lake Intel only guarantees 4400 MHz (4800 MHz with 4x single rank CUDIMM)Kingston has a nice overview: https://www.kingston.com/en/memory/memory-population-rules
>>107889709intel always undersells their memory speeds by a wide margin. AMD perhaps not as much.
>>107889635>Also 10nm is supposed to have the same voltage tolerances as 14nmbut it's also more prone to thermal runawayalso it's not a design issue when something isnt stable when run beyond it's design limits
>>107889706>Until I saw with my own eyes that X670 on a 7800X3D boots four memory sticks at 6000 MHz CL30, all stable, using a ready-made preset.Doubt.avi.It was probably stable enough for a screenshot, but it will shit itself under any load on the memory controller.
