Sheldon's wisdom

 Nice title, but it's mostly me and a video I saw about Aida64 (it totally sucks)
and a self-reminder to look at Linpack.

RRD_S, RRD_L and TFaw are related (I knew that) but long ago I had decided my RRD_L should be "10" and the buildzoid guy in the video 
is pushing 4-4-16 for rrds-rrdl-tfaw.
Anyway he used "Timespy" as a demo, and said it was better at the 4-4-16 thing. 
I don't know why but my TRRD_L should be 8, not 4, But I would just say
4-4-16 works but is slower than 4-8-16. (In timespy.)

My desk is a zoo of teeny bugs, my leg has red spots, so my concentration isn't what it should be, and I didn't follow every word of the video.

What's that? Should work on every memory ever but it might not work on some motherboards?
Whut, memory bandwidth/latency improves on slower settings?

I got Aida sucks, I got Timespy is a good test, but beyond that I can't agree the RRD_L should be the same as RRD_S (or why would they have two settings?) and I proved it. His examples should have been tighter, but he stretched them to make a point.
It's been a goal of mine for the last couple days to really find out about this whole Zen3 CPU-heat thing. So far it's the same or cooler.
Buildzoid ("Sheldon") seems to think Linpack will heat it up good, as well as break my PC (Why would I want to break...o nvm)
There's an Intel Linpack,

There's an AMD Linpack, but I get the distinct impression my poor little PC is underpowered for them. Plus it might be a Linux-thing, IDK. 

But I have Timespy! 

Cheap?!!

https://www.techpowerup.com/download/linpack-xtreme/
There's nothing to replace this with...
There is no store I can go back to (so's they can palm it off on someone else)
IDK, 🧐

I still won't know what my Latency is....I don't care how many nonsensical Gflops I can sling (I don't think)... Aida costs bucks but it's readable (IDK)

I like this benchmark, it says I'm fast, but will it love me
Tomorrow?

Point being, it would briefly jump into mid-sixties celsius-territory, 

go back to 50's...I was making popcorn so I wasn't watching every second.

Plus I *still* don't have a widely accepted result for latency.

The people buying my memory are either way overclocked or they aren't interested in Aida, maybe both.

PPT, TDC, EDC, They're just letters in a chart to me, and for several years I relied on PBO ("precision boot overdrive") to set them for me, and PBO would set them very high (like, "500" high)

But lately I've become more concerned with the values.

PPT=142, TDC=95, EDC=(mffm) say everyone, is a good default.

But at what point do those values slow stuff down?
Should PPT be 150, should EDC be higher?

The CPU never rises above 65C or so, am I going in the right direction?

Self-doubt never looks good in a blog, 

I won't hear the answers.

(Uhhhhhhh!) raising them extremely slightly from the defaults,

Religiously running CinebenchR20 afterwards, I long for the value I got when I first got this thing (everything seemed so fast then)

And I think I've got it, and slightly improved to boot.

(Well, I don't know, I'll ramble more about this in future entries)

A *horsefly* will land at the north pole and change the value.
(IDK)
For the record, "SMT (union-cores)" are "off"
so it really means "16" not "16 and friends"

*now* I'm itching (sort of) to download Cinebench r23, buttttttt....since I care little for the score and only worry about the heat, 

(since It's the lowest score on the planet beeeeeeecause "smt" is turned off),

Is R23 prettier? New-and-improved?
But some guy in the 90C-doldrums mentioned it along with his PPT-etc settings, so I will maybe look into it.

R23 forces you to stare at an obsessively clean and sterile room *forever* 

(8 eternally long minutes?) IDK, and I don't envy the person living there, hoping to brighten up the place with prints on the wall, a *casually* dropped magazine on the table.

It screams "prison" to me. I reached 67C.

https://vimeo.com/574354829

The  abnormally low-contrast of these pics worries me, my eyes have either gone weary from staring at the room or my monitor is dying...I'll go look at BRIGHT stuff for a while.

Imagine for a second you're in some math class, and some guy is struggling with his zillion-button programmable calculator, and your four-function gave you the answer a minute ago.

It's what I think about PBO-Curves, as opposed to ham-handed voltage-offsets which do the damn job so why tweak anything else (You'd have to see the picture in my head to understand)

---

https://www.anandtech.com/show/15137/amd-clarifies-best-cores-vs-preferred-cores (ble??) Variable benchmark scores might be from stuff mentioned in the link, or it could be dusty-tome folderol no one cares about anymore, 

(seriously, zzzzzzzzzzz) and THEN there are the sleep-inducing user comments.

(better, first)

But (Bottom lining it) 

there are ways for windows to share a thread between CPU-cores.

This would be great, if all the cores were equal and in the same spot, electrically, but they're not (I need to read another quaint-and curious lore-filled site about WHY cores are in different rooms, not all stuck together).

So anyway this "windows scheduler" balances heat and other things I was too sleepy to understand, and AMD needed to fudge figures to trick the windows-scheduler into emphasizing using cores from one house (ccx).

Otherwise, windows would inadvertently be slowing stuff down just trying to be helpful.

But then the article gets super bombastic and tries to explain preferred cores and (zzzz) wht? o. "Preferred cores" and better-performing cores, which should be the same but are not, to fool the windows scheduler.

And All of this was in 2019, so that gold star next to your favorite core in "Ryzen Master," might not mean what it used to mean.

(Did I mention "affinity?" NVM)

I'm not familiar with the new "curve" features, tba.

That's nice...Have a nice day
("unaware"...)

They can't (or won't) get mostly equal cores to live in a house, it's too hard, let the kids do it instead.