RAM -> "read-modify-write" / data masking causing this issue :-) (intel z690 z790 asrock only)

[ablemor]

Z790 without flashback??

[Kipperii]

If this method cannot explain why many people do not experience this problem without making any adjustments to their computers, then the solution is inconsistent.

[agendarsky]

i believe most ppl are clueless about this, like if i never touched oldschool pc era i would just think its game fault or some placebo, u look at fps counter and be like “yo its smooth its fine” … nah its not. i dont even call this input lag anymore, its some other beast. back in 2010 i was on old ass machine, only cs 1.6, nothing fancy, and i was top10 in cz/sk 1v1 aim league. then i swap hardware “better” on paper and boom, cant even kill random casual nonsteam kids. straight after hw change, so not emi, no voodoo, its either gpu or dimms.
gpu side there is not much… u can waste life tweaking nvidia inspector, digging into hidden registry flags, u gain 1 thing lose 2 things, fps is not problem here. counter shows high fps but feel is trash, and overtime degradation hits anyway.
ram side is way bigger mess. every mobo gives different bios options, some hide stuff, some expose it, each board got diff signal strength, diff training algos. i noticed turnaround timings not even same every boot lol. like wtf, one boot its tight, next boot its loose, feels different. turnaround timings are those gaps when mem bus switch direction read→write or write→read, basically dead cycles. not fixed, not stable, retrained every time. u feel it as weird desync but fps counter still flexing.
and this is not only thing. scheduler inside imc, dbi (data bus inversion), data masking, write bursts, refresh colliding, all that crap. invisible in benchmarks but u feel it in your aim.
and then u read about RMW (read-modify-write). wtf is this shit. instead of just write, dram do read first, modify bits, then write back. like why. its hardwired, not some bios toggle. no wonder it feels fucked, whole mem system doing gymnastics just to change few bits instead of straight write. Anyway , i reported this issue alongside with huge fps boost by disabling data masking (450 avg to 732 avg is no joke) to the Intel , got response back that they thank me for the feedback and investigation and it will be routed for specialized technical team to respond which means , from now they have 48 hours to respond. I doubt it will happen but atleast they responded. Also no response from AMD about this. If Intel actually start seriously investigate this and start caring about determinism and microlatencies and tend to fix this , they could easily take Amd from throne, i've got higher cs2 performance with "old" 14700k 33mb l3 than 7800x3d with 96 mb l3 just by tweaking my ram. Ram mode register set needs to reviewed carefully.

[agendarsky]

Well thats unlucky

https://i.ibb.co/GfJcCSWv/Screenshot-20 ... 172942.png

[MontyTheAverage]

i believe most ppl are clueless about this, like if i never touched oldschool pc era i would just think its game fault or some placebo, u look at fps counter and be like “yo its smooth its fine” … nah its not. i dont even call this input lag anymore, its some other beast. back in 2010 i was on old ass machine, only cs 1.6, nothing fancy, and i was top10 in cz/sk 1v1 aim league. then i swap hardware “better” on paper and boom, cant even kill random casual nonsteam kids. straight after hw change, so not emi, no voodoo, its either gpu or dimms.
gpu side there is not much… u can waste life tweaking nvidia inspector, digging into hidden registry flags, u gain 1 thing lose 2 things, fps is not problem here. counter shows high fps but feel is trash, and overtime degradation hits anyway.
ram side is way bigger mess. every mobo gives different bios options, some hide stuff, some expose it, each board got diff signal strength, diff training algos. i noticed turnaround timings not even same every boot lol. like wtf, one boot its tight, next boot its loose, feels different. turnaround timings are those gaps when mem bus switch direction read→write or write→read, basically dead cycles. not fixed, not stable, retrained every time. u feel it as weird desync but fps counter still flexing.
and this is not only thing. scheduler inside imc, dbi (data bus inversion), data masking, write bursts, refresh colliding, all that crap. invisible in benchmarks but u feel it in your aim.
and then u read about RMW (read-modify-write). wtf is this shit. instead of just write, dram do read first, modify bits, then write back. like why. its hardwired, not some bios toggle. no wonder it feels fucked, whole mem system doing gymnastics just to change few bits instead of straight write. Anyway , i reported this issue alongside with huge fps boost by disabling data masking (450 avg to 732 avg is no joke) to the Intel , got response back that they thank me for the feedback and investigation and it will be routed for specialized technical team to respond which means , from now they have 48 hours to respond. I doubt it will happen but atleast they responded. Also no response from AMD about this. If Intel actually start seriously investigate this and start caring about determinism and microlatencies and tend to fix this , they could easily take Amd from throne, i've got higher cs2 performance with "old" 14700k 33mb l3 than 7800x3d with 96 mb l3 just by tweaking my ram. Ram mode register set needs to reviewed carefully.

unfortunately I don't think I can tweak the ram I have on gaming laptop