I had one of my best driving lessons today, but what really irks me is the willingness of cyclists to blow past red lights. It happened three times during my 90-minute lesson today, including once where I was traversing an intersection at speed and had to hit the brakes hard (though not quite fully or to a full stop) to avoid a collision.
I mean, seriously, WTF. I did, however, manage to react before my instructor had to take action.
1:00 pm in the city. I'm not surprised that there's lots of food deliveries, but your safety has to come first.
If you need maximum security, use a TPM. A Trusted Platform Module can encrypt memory regions and decrypt the memory only for specific applications or the OS, without the encryption key ever leaving the security processor itself.
But while TPMs are commonplace in modern laptops, they're not universally available.
But IIRC there are other ways to protect sensitive data in memory, even if they're not as bulletproof as a TPM...
(though do note that vulnerabilities have been found in TPMs in the past; for example, Infineon modules were subject to a vulnerability called ROCA; Windows will not use an Infineon TPM whose firmware is not updated to fix this)
And ideally should lock it in RAM too, so it doesn't get paged out.
(In Windows, that'd be SecureZeroMemory which cannot get optimised out and runs immediately unlike ZeroMemory, and locking in RAM is possible too.)
But if the machine is hibernated, that'll still get written to disk and I don't know if there's much you can do about that. Maybe register for the suspend event and purge memory when it happens?
Two main applications: 1) DRM for high-value content (e.g. 4K video), and 2) guarding sensitive data like encryption keys (e.g. BitLocker, Windows Hello).
Yes, Windows Hello uses a TPM where one is available, e.g. to store the PIN.
And IMO it's a must for biometric auth, because you can't change your biometric identifiers.
(note that I'm limiting the content I'm posting in chat over the next several days, especially on Matrix, due to a mod warning; won't go into detail here)
It's just that NVIDIA is rather picky and might be testing edge cases that are not routinely encountered but may fail on cheaper FreeSync displays.
They're also trying to show that G-SYNC > FreeSync (though again, that's more a matter of the standards being higher for G-SYNC displays). There are premium FreeSync monitors, and those are unlikely to throw up the sorts of issues that NVIDIA claims might occur.
FWIW, I do have a G-SYNC monitor and the experience is practically flawless. I don't know much about FreeSync, but I've heard reports of issues especially during the early days after its introduction.
I suspect any properly-designed FreeSync display will not perform noticeably worse than an equivalent G-SYNC display under all reasonably foreseeable usage conditions.
Alright, I finally have the Radeon VII FP64 performance matter sorted out with AMD.
Contrary to earlier statements, it is being throttled. Radeon VII's rate will be 1:8, versus Vega 20's native 1:2 rate. Notably, this is still twice the native FP64 rate of all other Vegas.
For reference, Vega 10 and all variants of Polaris have FP64 at 1/16 of FP32, at the silicon level. AMD is trying to strike a balance between giving consumers more FP64 capability and not cannibalizing the Radeon Instinct cards.
@Bob while broadband speeds (at least for consumers) aren't too great here, support and customer care is amazing
It's never taken more than 15 seconds to reach a support agent, who either troubleshooted and fixed everything on phone/remotely, or dispatched someone to fix it on-site who arrived within a couple hours
Also, I find it amazing how internet costs (both mobile and broadband) here are exponentially cheaper in India. And "contracts" aren't a thing here.
You cancel when and as you want to. No cancellation charges, no nothing.
Here in New York, customer support for Internet service is generally good, but it still takes a few days to send out a technician. At least with Spectrum, you get a one-hour time window when you set up a service appointment, and IME they've reliably arrived inside that window.
...aargh. Why does my body just decide to tire out whenever I try to spend more than a few minutes on learning Docker?
As in, make me physically want to go to sleep.
Whatever. I'll see if I can continue after a bit of a nap, or if I should just close up shop (it's midnight here in New York) and start again tomorrow. Regardless, I can report that I did make some progress today.
@bwDraco don't the 'pro' Radeons have ECC RAM and things like genlock and more VRAM? seems unnecessary to artificially slow down the cards
but then I've long wondered if there's a silicon industry conspiracy to "stair step" the public with 20-50% improvements year over year rather than just going directly to the smallest node possible with the current science
I know it's not a valid argument on its own, but I'm skeptical of the fact that every 2-4 years we discover entirely new science/engineering that allows us to just about halve the size of transistors; couldn't we think of things in, say, 2008 that would let us go straight to, I dunno, 28 nm? and then FinFET would unlock the ability to go straight to 7nm?
if that truly is the rate of progress of the cutting edge, well, fine; but I do wonder if they sometimes by chance unlock a technology that would skip them ahead several halvings of die size and decide to "stair step" us to keep the economy stable
And FWIW it's the datacenter market that offers the highest margins, and therefore subsidies much of the cost of gaming hardware.
The GeForce RTX cards are already expensive enough. Imagine if there wasn't much differentiation between Quadro and GeForce. Without the extra profit margin on Quadro and Tesla, you might as well be paying close to $2,000 for an RTX 2080 Ti... (Bear in mind that NVIDIA spent huge amounts of manpower and money developing the RTX platform.)
As if $1,200 isn't already out of the reach of most consumers.
according to that, they make 3x more revenue from gaming than datacenters... maybe less profit per unit but I wonder which one gives them more net profit
Also, the TU102 GPU is very expensive to manufacture in the first place.
I mean, I'm playing devil's advocate here so what I wrote may seem radical, but what might be ideal for consumers may not be economically viable from a business standpoint.
Also, note that from a technical standpoint, FP64 capability is not free. It takes up space on the silicon, making the chip more expensive to manufacture, and can increase power consumption even when not being used.
This is why since Pascal, all but the top datacenter-oriented GPU variants (GP100, GV100) have FP64 at 1/32 of FP32 at the silicon level.
In fact, FP64 units are very expensive in terms of die area.
Case in point: GP102 (GeForce GTX 1080 Ti) is 471 mm² while GP100 (Tesla P100) is 610 mm². The main difference is the presence of full FP64 capability at 1/2 of FP32. In fact, the GP100 has no ROPs and can't render graphics. The HBM2 interface and NVLink capability would not explain that large a disparity in die size.
In other words, NVIDIA had to take out the ROPs and thereby sacrifice its ability to rasterize graphics to free up space for the extra FP64 cores and keep the die within a reasonably manufacturable size.
They later overcame that limitation with Volta GV100 and the TSMC 12FFN process designed specifically for NVIDIA to enable extremely large dies.
The need for this special process given the insane 775 mm² die size of TU102 (GeForce RTX 2080 Ti) explains at least part of the exorbitant cost of the RTX cards.
@allquixotic Semiconductor manufacturing is a very complex process, and things can go wrong in a lot of different places. While I'm not terribly familiar with the topic, from what I know, the last few process nodes have involved a significant increase in the number of individual steps needed to fab a wafer.
With more steps, there's more potential for a defect to come up. It takes a lot of work to refine the technology to a point where mass production is actually viable.
This is why EUV lithography is such a huge deal: because the much shorter wavelength of the laser enables much finer details to be drawn onto the silicon with a single photolithographic exposure. But issues like the lifespan of parts used in EUV lithography and high-output power sources needed to manufacture chips at a reasonable speed has made this transition very difficult.
It isn't really a matter of "can we make transistors and chips this small" more than it is "can we manufacture devices with this technology at industrial scale and at reasonable cost".
And that's why it takes so long to deploy a new semiconductor manufacturing process. You're not producing a few hundred chips at a time. You need to be able to ship millions of chips a month.
Milma milk comes in blue packet, yellow packet and orange packet. Each has different big words on it like full fat, toned and homogenized respectively.
I am a simple man, I go for the Pepsi Zero and pretend I'm Marty McFly here from 2025.
@Burgi I remember lists in python are heterogeneous and arrays in C/C++ are homogeneous. It has to do with the shrubbery of data types, yes.
@djsmiley2k My tally was 8 people, 7 trucks, a few dozen transport buses and three hundred scooters. That was all this morning.
@bwDraco how do you talk like this? Were these words ingrained in your brain in vitro? I mean, I could stare at PCPartPicker for several years and still not have this much tech jargon absorbed into my veins.
@rahuldottech don't you have something better to do?
Seems like my PXC 550 just crashed on me and had to power-cycle to recover.
- Connect device 1. - Connect device 2. - Play audio from device 2. - Play audio from device 1. - Disconnect device 1 while it is playing audio.
The headphones stop playing any audio, though the ANC still functions. Switching it off and back on is the only way to recover; notably, the headphones don't say "Power off" as they should when switched off.
Now this is interesting. Just got an Anker 30W USB PD wall charger - it will in fact charge and power the laptop (HP ENVY x360) while it's on, even though it's not enough for full-load operation.
I thought this machine needed 40W?
It was pulling ~25W from the adapter.
As for the Jackery SuperCharge 20000 power bank... it'll charge that at 30W, too. (the power bank supports 45W input)
Not really, it's just pulling power from the battery if the AC adapter isn't enough.
In fact, it'll do that on the stock 45W adapter (4.5mm barrel connector) if it's under full load. I have a separately-purchased HP 65W adapter and that will deliver adequte power to the laptop under full load.