WTF Bites
-
@HardwareGeek said in WTF Bites:
@BernieTheBernie said in WTF Bites:
It works with my old Firefox.
@Lorne-Kates, is that you?
"Old firefox" just means that you didn't let it update in the last 5 minutes or so.
No, that’s Chrome.
And the “Google News” site displays a toaster that you need to reload it for a new version if it’s been open over the hour mark, maybe even every 15 minutes.
-
Looking up fans at Noctua. Needs a low-profile fan and I looked up the compatibility chart.
All the 105W TDP CPUs comes with a warning. Except for the 3800X. So despite being rated to do up to the same heat output as the other ones with a comment it's marked as being all fine.
Official TDP is a lie (see Ryzen 3600 v. 3600X - identical CPUs with slightly different clocks, but one is 65W and the other 95W). Maybe they looked at actual measured heat emission when making that compatibility chart.
-
I got a new car earlier this year. It came with a free trial of SiriusXM. I thought it sucked; never used it.
Since that trial expired I've been receiving nonstop emails, snail mails, and phone calls offering me deals to sign up. I've never seen someone so persistent.
Fuck off, already.
See also my own experience (well, the start of it)...
-
Looking up fans at Noctua. Needs a low-profile fan and I looked up the compatibility chart.
All the 105W TDP CPUs comes with a warning. Except for the 3800X. So despite being rated to do up to the same heat output as the other ones with a comment it's marked as being all fine.
Official TDP is a lie (see Ryzen 3600 v. 3600X - identical CPUs with slightly different clocks, but one is 65W and the other 95W). Maybe they looked at actual measured heat emission when making that compatibility chart.
I always thought that was "at idle", and not the actual TDP when you're actually doing something...
-
Official TDP is a lie (see Ryzen 3600 v. 3600X - identical CPUs with slightly different clocks, but one is 65W and the other 95W)
Slightly different clocks sometimes mean big swings in power. Gate leakage is non-linear black magic, but generally it shoots up quick, especially if the entire chip is a bit shit (which binned parts tend to be). Besides, OEMs probably get all the good chips anyway, because you don't fuck with big customers.
Consider 3900, which has 12 cores, but also is rated 65W. It is a lie, of course, because it peaks somewhere in the nineties. But while X will keep rocking over 4.0 as long as thermals are ok, non-X have a strict power limit and a very kneeling Boost, so the average is
:close-enough.jpg:.Anyway, I do believe they've tested it, but bench press of 3800X is still ~150W. If it's going to be a gaming device rather than for general use, you'll be pushing it.
But yes, TDP is
.
-
@Tsaukpaetra said in WTF Bites:
I always thought that was "at idle", and not the actual TDP when you're actually doing something...
It is for 14nm+++++++++
-
@Applied-Mediocrity said in WTF Bites:
Slightly different clocks sometimes mean big swings in power. Gate leakage is non-linear black magic, but generally it shoots up quick, especially if the entire chip is a bit shit
I won't bore you with all the details (which are a bit outside my wheelhouse, anyway), but voltage makes a bigger difference than clock frequency. There are two sources of power dissipation, static and dynamic.
Dynamic power dissipation happens when logic in the chip switches from 0 to 1 and vice versa. This is proportional to the rate at which the logic changes states, which is roughly proportional to the clock frequency. (Some logic switches only occasionally; some (DDR) switches at double the clock frequency.)
Static power is dissipated whether the logic is switching or not, and is independent of clock frequency. This increases as device sizes get smaller, and in advanced chips with small device geometries, (less than about 90nm, according to one source), it is the dominant type of power dissipation.
Voltage affects both, and the power is proportional to the square of the voltage. Modern chips not only turn off clocks to logic that isn't in use and slow down clocks to logic that doesn't need the performance at the moment to reduce dynamic power, they also turn off or reduce voltage to save static power. So increasing the CPU's workload will increase both kinds of power non-linearly. Logic that may not have been in use, or only intermittent use, will be switching, at a higher clock speed, and the voltage will be cranked up to support the higher speed.
-
@HardwareGeek Right. Let me backpedal on my fib about gate leakage, which is something I actually know less than nothing about, but I believe the frist part of my statement remains true.
If we put together
the voltage will be cranked up to support the higher speed.
Exhibit A: 7700K (all cores in sync)
I expect the shape of the curve will be similar for any modern CPU.
with
power is proportional to the square of the voltage
there we are.
***
Dynamic power dissipation happens when logic in the chip switches from 0 to 1 and vice versa.
I thought it was that zeroes were more round and fat?
-
@Applied-Mediocrity The Wikipedia article on TDP even mentions that peak power might be a fair bit higher.
From what I get, TDP is just the amount of heat that the system is expected to able to dissipate from the chip. Chips these days are heavily self-regulating (i.e., if the temperature goes too high, they will throttle themselves to generate less heat), and that's probably how they would expect to stay under the TDP typically. But if you provide better cooling, you could then stay over the TDP for long times, I guess?
The more interesting question is what the limits are on the power-supplying subsystem. How much is a CPU actually allowed to draw from the motherboard? (Tangentially related: there was some discussion about GPUs, where some higher-end AMD cards apparently were out-of-spec w.r.t. the PCIe and power supply stuff.)
-
@cvi It doesn't readily explain how, for example, 3900 (12C) can have the same TDP as 3600X (6C). Even their design is different. 3900 has two CCDs, i.e., larger die area. In theory it should be able to transfer heat faster and more evenly to the heatspreader. Their peak power is 90W and 80W, respectively. Now, 3900X has the same 12C construction, but is suddenly rated 105W and peaks at unholy 180W. Why? More Boost goes into the CPU, witchcraft happens and you go faster
:clarkson_powaaaaah.gif:So, it cannot even be used for comparison among SKUs, and its relation to peak power is rather unpredictable. I believe that's enough to declare TDP an asspulled number of no good.
Also, let's peruse the exact definition from Wikipedia.
maximum amount of heat that it would generate when running "real applications"
This is almost word-for-word what Intel is currently saying about their performance comparisons, isn't it? They conveniently leave out what they mean by "real applications" (because we might think it's bullshit).
where some higher-end AMD cards apparently were out-of-spec w.r.t. the PCIe
RTG has been the worst repeat offender, but so did GTX 690, so does RTX 2080 Ti and will do RTX 3090. I don't think it matters very much when you're one of the bigwigs of PCI-SIG.
I had a row about that with :@levicki back when he just started posting here. He claimed that his MSI 2080 Ti Uber-OC eats fuckton of power, because some software told it did whereas I claimed it didn't, because Tom's Hardware had measured it using science. But then it turned out I was looking at the wrong numbers myself. I never admitted that, obviously...
-
@cvi Iirc the current PCIe power spec should allow up to 375W total board power, with 2 8-pin power connectors. 75W from MB, and 150W each from the connectors (6-pin is 75W and 8-pin is 150W). The previous spec was 300W max (8+6-pin) but it was revised when a lot of GPUs started running 375W anyway. But with the next gen Nvidia GPUs running up to 3x8-pin... Hmm...
-
@Atazhaia Which must be why they "invented" the new 12-pin thingamajig which (despite having smaller terminals and wire gauge) is rated for 600W. Totally inspires confidence in the Ampere line. Heh, Ampere. Yes, there will be lots of Ampere there
***
Then again, I doubt it's possible to do worse than this:
-
@Applied-Mediocrity I really hope that's at least a dual-GPU board because that one board beats my dual HD5870 setup in power connectors.
-
@Atazhaia It's two R9 390s.
-
@Atazhaia Indeed. Two 290X, pulling over half a kilowatt in. There's no kill like mo-mo-MO-Monster kill-ll...
@loopback0 said in WTF Bites:
It's two R9 390s.
Or that.
-
@Applied-Mediocrity And then we need to run two of them in Crossfire too for maximum power!
And, hm, well, that's pretty much an older version of the maximum GPU setup in the new Mac Pro, which can do double dual-GPU Radeon Pro cards.
-
@Applied-Mediocrity said in WTF Bites:
Slightly different clocks sometimes mean big swings in power. Gate leakage is non-linear black magic, but generally it shoots up quick, especially if the entire chip is a bit shit (which binned parts tend to be).
It's even more complicated than that. One of the technology choices possible at the low level is exactly what variants of gates are used. In particular, things like the thickness of oxide layers can be controlled. Very very low level stuff. The key take-away from that is that gates can be configured to run at various switching speeds (for a given voltage) and the higher switching speeds come with higher leakage currents and more power dissipation. (I know a little bit about this because some of my team do hardware accelerator design, where this stuff is very very important. It's still nearly incomprehensible gibberish to me though; I'm a software guy, after all.
)The faster CPUs could well be minor redesigns to use faster gates on the critical paths. They'd look almost identical (especially without an electron microscope to do detailed teardown) but the faster CPU would use more power.
-
@Applied-Mediocrity said in WTF Bites:
They conveniently leave out what they mean by "real applications" (because we might think it's bullshit).
Yeah, kinda forgot about that while I was writing the post. I wanted to conclude that the power consumption is mostly controlled by "software" (firmware etc), so TDPs aren't really related to just the hardware, so they end up being kinda arbitrary.
Saw the "real applications" part too. That definitively smelled like BS. (After all, all the Electron crap that's keeping my CPU busy hardly qualifies as a "real applications".
)RTG has been the worst repeat offender, but so did GTX 690, so does RTX 2080 Ti and will do RTX 3090. I don't think it matters very much when you're one of the bigwigs of PCI-SIG.
IIRC, somebody had hooked up a power meter to the card after there were some reports of frying motherboards. (Probably lower-end motherboards with less tolerances.) Now that I think back, it might not even have been the higher-end, but more the low/middle-end, where manufacturers are less keen on adding the additional 6 or 8 pin connectors for more power, and instead want to rely on power from the PCIe slot only.
-
@Atazhaia Yep. The new 30x0 cards do sound rather toasty. Good thing they're releasing in fall/beginning of winter...
-
Now that I think back, it might not even have been the higher-end, but more the low/middle-end
It might have been GTX 750 Ti. Some models, including entirely passive, were without additional power. Probably to underscore how economical was Kepler's jump to 28nm (which it was, just not quite as much).
https://www.tomshardware.com/reviews/geforce-gtx-750-ti-review,3750-20.html.
The average looks good, but those spikes do not.
-
@Applied-Mediocrity said in WTF Bites:
if the entire chip is a bit shit (which binned parts tend to be)
Well yeah, why else would they throw em out?
-
On the Community Server forum I support, we've made the decision to "encourage" users to use non-free email addresses by limiting the participation of users with gmail et al domains. This sounds weird but there are valid
.When such a user tries to post, we show them a message asking them to change their email address. We link them to the form where they can change it. Just kidding. We tell them to look up how to change their email in the FAQs. We provide a link, of course, not to the specific FAQ where they can find that answer, but to the FAQs landing page so they can hunt down that FAQ themselves. The FAQ itself does not provide a link to the form, either, but instructions on how to navigate to the form.
You've got to keep the users engaged, you see.
-
When such a user tries to post, we show them a message asking them to change their email address. We link them to the form where they can change it. Just kidding. We tell them to look up how to change their email in the FAQs. We provide a link, of course, not to the specific FAQ where they can find that answer, but to the FAQs landing page so they can hunt down that FAQ themselves. The FAQ itself does not provide a link to the form, either, but instructions on how to navigate to the form.
The evil ideas thread is
-
On the Community Server forum I support
And for the last time, the BDSM thread is
!
-
Our build is so non-deterministic I'm seriously contemplating writing a "retry until it works" script.
-
@error If you also checked that the last error is not the same as previously (because if it is, it is probably legitimately broken), it would probably work. Err, I mean, evil ideas thread is
.
-
Skype for Linux is amusing. When I start it, I briefly (a few frames) get the "default" window that shows all my contacts, previous calls and so on. That window then immediately closes, and I get thrown back out to the sign-in window. This happens every time.
-
Edit: spot the typo in the answer to the third question. Way to go, fonts.com!
Late to the party, but that's not a typo, it's the inability of that page (or any of the things holding that document since the original) to - ironically in the context of the discussion - to store the ligature fi (U+FB01).
archive.wtf
Other nearby occurrences of fi not ligatured shown.
Quite why that's the only occurrence of the ligature in the whole document is it's own
.
-
Apple names their new iPad colors:
Rosé Gold
Sky Blue
and
GreenJust Green. No fancy kind of Green. It's just Green.
-
@Atazhaia Presumably this is also yet another shade of Space Grey that's very similar but not the same as the many other shades of Space Grey on previous devices.
-
Quite why that's the only occurrence of the ligature in the whole document is it's own
.Why it appears in the document is the
. The ligature codepoints are supposed to be only used in fonts for storing the ligature glyphs, but it should be left to the renderer to decide if and where it will use the ligature, because which ligatures are used depends on the font. It is similar with position-sensitive forms in Arabic and Devanagari and other similar properties.
-
The ligature codepoints
The issue is that they shouldn't be codepoints at all, but rather just features of the set of glyphs used. But in English there's not many ligatures used so quite a few pieces of software ended up describing places where they want ligatures as actual characters. Everything else follows on from that. It doesn't really work in very ligature-heavy
languagesscripts, where an entirely different mechanism is required in order to keep the mess of ligatures at least mostly under control.tl;dr: ligatures shouldn't have codepoints, but sometimes do
-
@dkf They shouldn't, but Unicode was actually created with ass-backwards compatidebility in mind, so for :hysterical:
they do. And there is also all the ARABIC LETTER * (INITIAL|FINAL|MEDIAL|ISOLATED) FORM for which the same applies.
-
If you also checked that the last error is not the same as previously (because if it is, it is probably legitimately broken), it would probably work.
The last error will be file locked, the next error will be file locked. Eventually the file will not be locked and the build will work.
-
Just Green. No fancy kind of Green. It's just Green.
Focus groups did not respond well to the proposed "vomit green."
-
Scientific advisor, just now:
: We now have an official Instagram page! Every single one of you should prepare enough material for an Instagram post once every two months!
: 
-
@aitap That lives up there with the scientific-breakthroughs-per-second metric.
-
If you also checked that the last error is not the same as previously (because if it is, it is probably legitimately broken), it would probably work.
The last error will be file locked, the next error will be file locked. Eventually the file will not be locked and the build will work.
Well, then you can just check for that message and restart the build if you got that one.
By the way, TeamCity has a Retry Build Trigger, which just restarts the build specified number of times if it is failing. I do not see a way to tell it to retry if it failed with specific error though.
-
Scientific advisor, just now:
: We now have an official Instagram page! Every single one of you should prepare enough material for an Instagram post once every two months!
: One selfie?
-
Scientific advisor, just now:
: We now have an official Instagram page! Every single one of you should prepare enough material for an Instagram post once every two months!
-
@cvi You must be using Linux hardware
-
One selfie?
If only. It has to be appealing to both students (who may as a result wish to come work in the lab) and "potential collaborators" (whatever it actually means. May have been business owners), but no one bothered to come up with actual guidelines on what to fill the void with.
: "You're the statistics guy. Maybe you'll give me some beautiful-looking fitted surfaces or something."
-
@cvi Apparently it tries to use one of the keyring services to store authentication data and expects everything to just work a bit too optimistically.
-
Edit: spot the typo in the answer to the third question. Way to go, fonts.com!
Late to the party, but that's not a typo, it's the inability of that page (or any of the things holding that document since the original) to - ironically in the context of the discussion - to store the ligature fi (U+FB01).
archive.wtf
Other nearby occurrences of fi not ligatured shown.
Quite why that's the only occurrence of the ligature in the whole document is it's own.Also, why is it
fiand notffi, (U+FB03)?
-
Rosé Gold
According to the text in that image, it's Rose Gold, not Rosé Gold. And rose gold is a thing, quite apart from Apple Dumb Names™.Gold is, of course, gold-colored. But pure gold is too soft for many purposes, so it is typically alloyed with other metals. (The amount of gold in the alloy is designated by "karat"; n karat gold is n * 1/24 gold. 18 karat (18k) gold is 18/24, 75%, gold. 14k is 14/24, 58.3%, gold. 10k (the lowest that can be sold as "gold" under US law) is 10/24, 41.7%, gold.) It can be alloyed with anything, but silver, copper, nickel and palladium are the most common, and the choice and proportion of alloy ingredients can change the color somewhat. An alloy of gold and nickel or palladium produces "white gold". An alloy of gold and silver will have a slightly greenish tint — "green gold". And an alloy of gold and copper will have a reddish color — "red gold" or "rose gold". Gold alloyed with silver and copper together in the right proportions balances out those color shifts to give a good imitation of pure gold. (There may also be other minor components in the alloy to produce desired metallurgical properties, such as zinc to lower the melting point to make solder, or [I don't remember what and
to look it up] to make the molten alloy flow better for casting.)
-
-
@error We need an ffs ligature…
-
-
Apple names their new iPad colors:
Rosé Gold
Sky Blue
and
GreenJust Green. No fancy kind of Green. It's just Green.
Boring. Needs more A.I.
New paint colors invented by neural network - AI WeirdnessCommentShareCommentShare
So if you’ve ever picked out paint, you know that every infinitesimally different shade of blue, beige, and gray has its own descriptive, attractive name. Tuscan sunrise, blushing pear, Tradewind, etc… There are in fact people who invent these names for a living
-
Just Green. No fancy kind of Green. It's just Green.
Focus groups did not respond well to the proposed "vomit green."
Was "soylent green" an option?
