I wish your results weren't so RNG across multiple reviews.. Tweaking sub timings without explicitly mentioning it.. some of these results just make no sense having used most dies on a 5800x.
@jaszy Hmm I'm not completely following you here. The only sub timings I'm adjusting is to match the XMP SPD.
Feels like theres no baseline and re-using results across multiple versions of AIDA should also be a no no. Last reviewer had the same problem that skewed certain kits into looking better/worse than they are.. IE: V6.20 >V6.30 being massively different in bandwidth results... yet posting new kits with older results of previous kits..
Not sure what "baseline" you are looking for. I generally do not update AIDA64 since the test system isn't connected to the internet. Re-using previous data is fine, Especially if its the same version AND the same CPU locked to a certain frequency as I do. Like do you want me to re-test every kit for every review? that seems a bit silly. I haven't personally noticed a difference in the past between versions. I understand your concern though.
AIDA64 is a bit out of context in itself. All it does is show the highest peak bandwidth. Which can change depending on what's running in the background. I do not think it directly correlates to real-world usage. But it does show you when a memory kit is "technically" superior for reads, writes, copy and latency.
Side note: I know AMD boards/BIOS/AGESA don't set TRC, but doesn't setting it manually defeat the purpose of XMP? I feel like using XMP (on AMD) should be 100% out of box settings with no manual adjustments .. Gives a false impression otherwise for users that just want 1 click performance. 2c.
Maybe an "optimal setting" result tab (what your doing now) + out of box?
Manually setting the TRC has its drawbacks in a review setting. However it is different per MB. Some actually do use all the SPD data. Others change it for ever boot. Therefore setting it is the only way to insure arcuate results for every boot, different MBs and future re-test. In this memory review, I actually had to set the tRC to 84 since the SPD of 72 would not post for the AMD system. Is it just this memory kit? Nah, I had this happen plenty of times before. Once again it just depends on the MB.
I also set all the other SPD data like tFAW, RRDS, RRDL and tRFC. Once again all the XMP SPD data available to insure actuate results. This is why two kits running the same primary timings can have widely different scores. CAS isn't the only important value. Part of it is based on the type of ICs, (Hynix, Samsung, Micron). In other ways, "poorly" binned kits can be passed off a great to the consumer because all they look at is the primary and nothing else.
I'll tell you a secret. Most memory benchmarks are redundant and out of context. AKA unless you have a 5800X + GTX 2080, you will get different results. It is the inherent problem with memory reviews. It is impossible to test all configurations and scenarios. Even core clock affects the results. I would do away with most benchmarks if I could get away with it. But generally most people just look at the charts and nothing else....
In my mind, memory reviews in general are more about the product itself and a general guideline of what is appropriate for different systems. Like how Ryzen benefits from 1:1 FCLK. Or the likelihood of actually running the XMP profile. I tossed out all the stuff from the last guy and started over because I had no idea what his settings was or if he did anything but just enable XMP. Like I said every MB is different in how it applies sub-timings. Sometimes the only reason it fails to boot at higher frequencies on cheaper MBs is because the training simply does not work. Type in the SPD data and off you go.
