Some things that have great meaning signify nothing for others, and I get that, so I'll just keep writing and hope it helps someone.
OK there was no activation fee for my internet, not when I started, not now.
And when I started, before spectrum was a name covering three companies,
The old cable company I was on were greedy bastards who had two speeds, 60(ish) and 100(ish).
And if you wanted the 100-ish speed, you needed to pay (I can't remember, $100?? $200??)
NOW it's twice that piddly speed, $200 be damned (the upgrade was free).
So....were they asleep, are we under some sort of Covid-Dispensation?
I can't be sure.
Every damn time I try talking to these guys I fuck up royally, so You call, you're much sexier.
Peaches and Dulcinea: One (*gasp*) last duty to perform (orphanrocks.blogspot.com)
under the section "Chapter 4"
It was a remedial-math-student's nightmare, trying to figure out how much more I would pay.
Well just now I broke-out the calculator and subtracted last month's bill from this month's bill.
I think the answer was "32"
I remember how disappointed my remedial-math teacher
 |
A = 2 π r h + 2 π r 2 |
was in me, but I think she'd agree that this time, the price wasn't too bad.
-----
Chapter-the-second, WDDM (It's a fancy name for the video-card driver)
and "Direcx12 Ultimate" and you're wasting both our time if you think I'm speaking generally about Directx or directx 12.
No, you've heard the pontifications, that WDDM needs to be 2.9 to run Linux,
That 12.2 is needed.
What a load of crap!
I *Could* go into some forum, say that and get everyone self-righteously Pissed Off
but here, you judge:
 |
| Fuck'n A |
I've seen 12.2 listed before, and wddm 2.9 before, but *now* it's all wddm 2.7 and directx 12.1, and I don't know why.
I can guess I have the wrong driver, but it seems to work fine.
I nearly woke up this morning to add a note about how fantastically great this keyboard is, but then I saw the typos.
Well, whatever:
BUY a USB-STICK (I don't need one but risking my 500GB usb drive on bios updates seems foolish)Don't Buy a HeatsinkDon't buy a PC-Speaker (they're annoying, and I only wanted one because the old one joined my dentures in eternity or the 5th dimension)Buy a Thermostat!! (Geez, if I forget this one, call the home to send their chariot, comin' for to carry me home)I have enough fans (dammit) Ah, shit, OK just this once but I think it'll be useless mostly.- Buy two more 3600c14 sticks to complete the set! They're the only things cheaper than when I bought them, they are the only things worth buying that I can afford, CMK16GX4M2Z3600C14 (Waaaait, I remember writing this blog back then, they were the more expensive ones because I'd bought what seemed like the 1.35v version, which turned out to be a typo, and they're the same price now or slightly higher) (I'm getting used to living on air for no good reason, hmm)
don't really say if 3600c14 can be scaled down to 3200c12, they just piss at each other.
But I wonder, sort of.
So it'll remain to be seen and since I am no Gandalf (whoever your trendy wise-type person is, "Yoda"?)
of memory, I'll be happy at 3466c14.
It seems I'd never be able to buy natively clocked decent DDR4, only *overclocked*, stick-up the butt, fire-in-the-ass memory, that would be much happier running 2666-c18. O, well.
 |
| Uhm, ??? hmm. |
I'm looking for a math-wiz site with a current-setup but in forums, "3200c12" must seem so last year, so decrepit. He seems to be just throwing out numbers, not really saying "I ran/run 3200c12 using *this* setup"
 |
| A good quote from a blog (speed beats latency,) Choosing Your RAM: What is CAS Latency and When Does it Matter? - Logical Increments Blog |
Whether the above quote is true I don't know, but I seem to want to agree, based upon what little tiny benchmarks I've done.
Plus if this looks disjointed, I just edited-in the picture and this sentence.
But if a speed cannot be achieved, and everyone says you're overclocking anyway, lower-latency seems to me to be the next-best thing to do.
I am not optimistic about all four of my ram-sticks doing 3600-c14, and the whole latency argument-research thing is relevant to me.
calculating cas-latency isn't easy for me, not knowing math,
and I do not really trust those calculator sites,
although when comparing two numbers using that same calculator,
it might not matter.
My example is Ram Latency Calculator (notkyon.moe)
which says that the data rate (it's the only rate it accepts)
(uhm, OK the clock-rate is half that, right?)
Yeah but then people start throwing out letters like MT/s, etc.
Whatever; Using that calculator for whatever it's worth,
3600-whatevers at cas 14 is 7.7ns,
3200-whatevers at cas 12 is is 7.5 ns, which would mean that,
if they're already sticking a prod up the poor DDR's butt
to achieve cas 14 at 3600, the poor thing would fall down and die at 3200 c12.
to achieve cas 14 at 3600, the poor thing would fall down and die at 3200 c12.
c13 is definitely doable.
Now, if *only* all those sites took that CAS-latency/nanosecond thing to calculate all the rest of those numbers, you'd have a good overall (theoretical) picture, but you don't, they factor in tides and phases of the moon as well, and it's daunting, hopeless.
Regardless...7.7 to 7.5 is small, but it might mean the difference between
being stable and being nervous with a twitch.
It follows (sort of) that 3466 (my current speed) with a calculated nanosecond-CAS of 7.7ns, might be even more ideal.
Waitasec, I'll ask:
CAS 13 is too fast, unstably so.
CAS 14 is of course slower...
So OK messing with CAS (in my head,) has proved mostly useless,
If I can get it to run at 3600C14, I should.
(without twitches) ("sans contractions," google says)
But hold on there...
I'm running CAS 16@3466 *now*, what would 14@3466 look like?
9.23, vs 8.07
That's a Full nanosecond faster.
That's good, right??
Except that, it isn't the whole story.
Buried under the iceberg of the first four advertised numbers is a whale of a lot of really tiny arcane numbers no one really knows much about (deferring to calculators from Russia), and a memory stick from Macy's that looks faster might in fact be slower than one from Gimbels, because of those teeny numbers.
But since I already own two, the choice is obvious, buy two more from the same company and hope like hell they haven't messed with revisions.
I probably got some figures wrong, but bottom line it's a thing that will improve my mismatched memory, and I won't have to worry about which DIMM goes where, they'd all be alike.
(As far as I can tell, software tells me that the board uses alternate slots, slow-fast, not how they are physically installed, slow-slow-fast-fast)
...even if the memory does zilch, nothing,
it would in theory be much more stable.
So (for example) 3800cxx memory, much more trendy,
would be more, four sticks would be more, etc.
I'm being frugal!
Nothing with a computer that limps, vs a stable setup, assuming I'm relatively
lucky and the New DIMMs work as advertised.
The rest of this is gibberish-notes, not to be read by anyone. "Eyes only."
uh
3466c16 is 9.23 ns, heavily rounded.
2933(1466) =158.9 *somethings*, if you take the true-speed (half of bragging speed)
as figured out by ...wait, I'm mixing my speeds.
But 9.23latency being a constant, couldn't someone lower speed and get a terrific latency figure?
so anyway, the disembodied voices say, "Calculate CAS12."
Which isn't worth the trouble *now* but *would* be if my entire memory were rated at 3600c14.
waitasec...
3000c12 is extremely close to 3600c14, it is slower by a hair.
Then you get into the whole "speed vs Latency"
argument.
*I* happen to think that since 3600 is way-overclocked, the latency wins out.
Except there's a limit to the latency of a Ryzen CPU, which I don't know, but I'll find out.
Everyone seems to like "12" but that is strictly anecdotal, not factual.
But speaking strictly as a seat-of-my-pants, blowhard-pontificator, slower might be better, within the bounds of the IMC's rated speed, not pushing it.
For more blowhardiness and until I find better links, please see
Unfortunately, the link I'm about to quote, talks *and Writes* so fast it's very difficult to decipher,
but he says everything I just said but with way more text, way more audience support.
TLDR: The ranges are small but Latency matters.
I swear I must be off-my-feed today, I can't really unnerstand a word of this next picture.
He loves very-fast vs latency, but SEEMS to be saying, 3000c12 rocks, pretty much.
I'm thinking I'll try the XMP for five minutes to see that everything more-or-less works,
then downclock it to 3000c12.
Was I unclear?
Geez, the hypocrisy.
This site has not proven useful to me so far.
Remember that we're speaking of values so incredibly small that they could easily be missed as a difference and mistaken for each other.
WAY more sophisticated test equipment is needed besides a game (which game? Which VERSION of that game? Which SETTINGS in that game? BAH!)
But they've nearly, almost, touched on a subject I'm looking for, which is why I found it:
Yes you can run up the speed but what about lowering latency?
Is it a crock or what?
Like ok, maybe you're designing cars with gigantic wheels to get you somewhere faster,
or teeny-tiny wheels, to save gas,
on uncrowded highways, in the summer, on a Tuesday at 4PM
 |
| Early evening in the barn after the cows have gone to bed, I sit there jotting on my checkbook |
All I've proven so far is, you can get damn-close to fast memory using lower and lower latencies, but can you *surpass* faster memory, if for example your memory-controller has a limited maximum speed?
Well-paid guys figure this stuff out, they went to college and/or use a table in a book, or a DRAM-calculator.
What they don't do is check every single dimm's timings, and let's hope they err on the side of caution.
But whatever, I gotta go and see if my current memory can be run at 3000C14-1T, and I am not optimistic.
Nope.
But if I had it down, maybe, I could switch one-single setting (cas14 to cas 12) next week!)
A cosmic question: Why can't a person run the Prime95 error testing thingy on one single number or group of numbers, and know whatever they tweaked is going in the wrong direction, or relatively stable?
What, the question was too long for you?
Suppose Prime95 finally fails "1280" after an hour.
It's complicated; let's just say it's a problem that the program works out repeatedly, looking for an answer it already knows.
So OK, you shut down the PC, tweak something, maybe voltage, and rerun the same damned error, now that you know what to zoom in on, but this time it does not fail,
it waits the de rigueur 30 minutes to fail something else!!
Maddening.
Anyway, once the memory seems stable at 3000c14, it also seems slower than 3466-c16, at least according to timespy.
It isn't much of a drop, but it's definitely noticeable.
Did I just waste two hours of my life??
If the ram, the new stuff, seems flaky at its rated speed of 3600, there'll be no reason to despair, I'll run it at a slower speed at a lower CAS.
But actually, if it'll run at 3600, that seems best.
No comments:
Post a Comment