buildzoid
patreon
Early tRAS re-testing results
🕑 Added 2025-01-27 09:41:40 +0000 UTCI've kinda mentioned that I want/need to retest tRAS on AM5 in a few videos livestreams. Well I've finally started on that. Here's the results I've gathered so from dual rank testing:
LOOK AT THE ATTACHED PNG
I also have data for SR but it's a mess and unfortunately I'm currently in the CZ for family reasons so it will take a while before I have the SR data in a format worthy of public viewing. Anyway the tRAS on AM5 doesn't affect performance "AT ALL"(I suspect the slow results at tRAS30 are the dual rank setup getting unstable). More intrestingly AMD's tRC implementation doesn't follow the JEDEC spec of tRC = tRAS+tRP and so setting tRC WAY BELLOW tRAS provides performance gains. I still need to test if setting tRC so extremely low can be/is stable. The main take away for now is that you can just set tRAS to 126 and not worry about it causing stability issues while using the "usual" tRC settings.
BTW I'm now even more confused about why AM5 even has a tRAS setting because IDK what AMD is actually using it for. Maybe some kind of manual row refresh? Becase ACT straight into a PRE would basically be a manual refresh of 1 row.
Comments
Rolly Carlos
Setting tRAS very low in my experience causes the iGPU to flicker the screen very badly especially when playing video.
Mario Valenzuela
I cannot get my tPHYRDL to match 😩 I have a 9800x3d, Asus crosshair x870e and G Skill ddr5 8000 CL 38 A-Die. I changed the PBU trainings from 0 to 5 and none of them match.
Actually Hardcore Overclocking (Buildzoid)
At the same timings and clocks all mem chips perform the same. The difference is that some mem chips just do higher clocks/better timings than other chips.
Joao Miguel
I failed to find answer on my searches so here comes a question. If we compare memory from different manufacturers but with same frequency and timings, would we see big differences?
Justin Korkiner
Link goes to a Discord CDN that 404's - https://cdn.discordapp.com/attachments/675738609390977029/1333024012829196350/image.png?ex=679762db&is=6796115b&hm=b5ed3ffc18685bcd4e9eea06ab9112f9b592e319c9a1f54377e29c41394d2763& Looked at devtools to verify. Not that it matters as the text is enough for testing. tRAS to 126 has zero negative performance impact on my DR AM5 setup so far. 2x32 Hynix A (6000 running at 6200 2133) on a 7950x3d. I am gradually testing lower tRC, but stable at 68 thus far. I actually see a bit of performance increase.
Yaroslav Furman
Could be cached, the direct link to it doesn't open either if I try to open the pic in the new window
Actually Hardcore Overclocking (Buildzoid)
try firefox because it works for me.
Justin Korkiner
Image doesn’t load for me either. App or Browser.
Yaroslav Furman
Yes, I'll double check what happened on the pc
Actually Hardcore Overclocking (Buildzoid)
did you find the working discord link?
Yaroslav Furman
The image doesn't seem to load anymore, the discord link is dead. Ok weird it loads fine from the mobile app.
ZarconDeeGrissom
I almost wonder if it's a name left over from SD66 days (Synchronous Dynamic, not DDR), especially with all the sub timings that are exceptions to tRAS for each possible read/write/refresh/idle combination, or maybe I'm thinking tRCD. /some-sarcasm AKA, have we got to the point that there are so many exceptions that the root timings don't matter anymore, lol.
Actually Hardcore Overclocking (Buildzoid)
by usually I mean the 60-70ish that I use for my OCs in 1:1 mode. So I guess set it like tRAS is 30 even when it's 126.
Xân
What do you mean with "usual" tRC setting? The usual way is as you say set tRC=tRAS+tRP but if setting it the usual way with a tRAS of 126, tRC would be really high. Do you mean finding a decent spot of tRP and tRAS, then set tRC accordingly and after that, set tRAS to 126? Or maybe just set tRAS to 126 and see how low tRC goes, regardless of the "rule"?
