Root Access

Bob

2:09 AM

quiet day today

Journeyman Geek

2:40 AM

Its barely started

bwDraco

roar

Does a GeForce GTX 580 hold up today?

Ave

3:01 AM

tldr: no

bwDraco

1.5 GB of framebuffer is the main limiting factor.

Also, power consumption is absurdly high compared to contemporary graphics cards. The GeForce GTX 1080 Ti has the same TDP but is an order of magnitude faster.

Journeyman Geek

also, its a funny point of time

didn't nvidia kinda have their a-hah moment in terms of efficiency/power with the 600s?

then again with the 800 series

bwDraco

Yeah. AMD even managed to get a brief process advantage over NVIDIA when they launched the 7000 series.

Journeyman Geek

@bwDraco I thought AMD literally had this strange situation were they were leagues ahead in non gaming workloads

bwDraco

@JourneymanGeek Yeah. The 7970 had close to fully-enabled FP64 compute at a time when NVIDIA was gradually cutting down on FP64 units in successive generation.

Nov 23 '16 at 8:13, by bwDraco

I mean, if you look at the Radeon HD 7970, FP64 performance was 1/4 of FP32.

(fully-enabled FP64 was 1/2 of FP32 on Tahiti; this was reserved for FirePro cards IIRC)

GCN was a total architectural rewrite for AMD.

AMD was betting on GPU compute becoming widespread. This never really materialized, though the compute-oriented nature of the GCN architecture does give it an advantage in some gaming workloads to this day.

NVIDIA disabling (up to Kepler) and eventually removing (since Maxwell) excess FP64 capability from its GPUs helped its power efficiency and performance in gaming workloads while boosting margins by letting it sell more expensive Quadro and Tesla cards to users that actually require it.

AMD has kinda gone the same route since Polaris, which has FP64 in hardware at 1/16 of FP32, unlike in earlier generations where FP64 capability up to 1/2 of FP32 was available in silicon but artificially limited to 1/8 or 1/16.

99.9% of consumers won't use GPU FP64, so why waste GPU die space and power on it?

Journeyman Geek

3:24 AM

well GPU compute is widespread

just not on the consumer level

bwDraco

Very few, if any, consumer applications use FP64 functionality. Having full FP64 on consumer cards would drive up manufacturing cost and power consumption because a significant amount of die space is used for functionality which is useless to most consumers. It would also mean lower performance for games; today's high-end GPUs are often power- and heat-constrained and wasting power on unused functionality reduces the power available for useful work. — bwDraco Feb 13 '17 at 2:14

Hence NVIDIA making specialized compute GPUs.

Journeyman Geek

erf. I seem to have misplaced my soldering iron

I know what its in, just not where that bag is ._.

bwDraco

AMD seems to lack the resources to come out with a specialized GPU with more than 1/16 FP64. The Vega 64 chip used in Radeon Pro and Radeon Instinct products is the same as the one used in consumer Radeon RX Vega products, only validated differently and supplied with different driver stacks. They don't have a special chip with extra FP64 capability, like NVIDIA has with GP100 and GV100. (Other recent NVIDIA GPUs have FP64 at 1/32 of FP32, at the silicon level.)

NVIDIA went so far as to get a bespoke 12nm process at TSMC (12FFN) for the GV100. AMD just does not have the money to do the same.

CUDA's market penetration is making AMD graphics cards an extremely hard sell for HPC applications where OpenCL is often not an option. For AMD to do the same sort of thing with a custom process, even at their preferred manufacturer GlobalFoundries, requires them to have massive profit margins because it's extremely expensive. Due to lack of market acceptance of their GPUs in the HPC world, AMD simply cannot command the sorts of margins NVIDIA is getting.

AMD simply does not have the money to develop niche or specialized devices like NVIDIA's tensor cores or DRIVE PX modules, let alone get a custom semiconductor fabrication process to manufacture a chip as large as GV100.

It's a money game, plain and simple.

allquixotic

5:26 AM

@bwDraco the only money they're winning lately is the custom dealios with Intel and the big console makers, but Nvidia won for the most popular console of 2017-2018, the Nintendo Switch... it has a Tegra X1

bwDraco

@allquixotic Yeah.

AMD has tight CPU-GPU integration, which is why they got the design wins for the PS4 and Xbox One, but that's about it.

Journeyman Geek

Community Promotion Ads for 2018

4

Cause I wanted a community promotion ad, for community production ads on meta, cause its meta...

Journeyman Geek

7:23 AM

Mk iii

►

New bias lighting setup

using 5050s, and a remote controlled relay

bwDraco

Wow. You spend so much time into projects like this.

I seriously need to be more motivated to act on my own initiative.

There is so much I want to do but I keep running into issues like "I need to spend more money on servers" (why not use existing infrastructure or perhaps just Astaroth?) or "I'm tied up on photos" (staying focused would speed this up and free up time).

yesterday, by bwDraco

There is absolutely no excuse for me to be wasting time doing meaningless things when I could be working on the same sorts of projects, learning JavaScript/PowerShell/Docker/LXD/ZFS/whatever while simultaneously meeting all of my personal and professional obligations.

So... what's the best way to just "reset" my mind so I can break free of this and get productive again?

allquixotic

@bwDraco As a point of analogy, consider if someone today asked the question "How can I best ask good questions on Super User?" while having about 36 reputation on the mainsite. Now, consider if that same user had asked hundreds of questions on meta.superuser, answered a few, but whose actual goal was to get rep on the mainsite. Yet every day they just ask more and more meta questiosn without actually contributing almost at all to the mainsite. What would you tell them?

if you need a concrete number, say they've asked, oh I don't know, 750 questions on meta. and they've answered maybe 120.

on the mainsite they have like one answer and one question

bwDraco

I'm trying to trace the history behind this problem and get to the bottom of this. Why am I stuck here in the first place?

allquixotic

So apparently your advice would be "Ask more questions on meta"? ;p

bwDraco

I simply cannot summon up the courage to actually do it.

allquixotic

7:38 AM

Shia LaBeouf "Just Do It" Motivational Speech (Original Video by LaBeouf, Rönkkö & Turner)

►

bwDraco

...this raises a more serious question. Would this need psychiatric treatment, like therapy or medication?

I seriously hope that in a few days/weeks/months from now, I can actually say "I set up a working LXD container!" or "I wrote this webapp, check it out!"

Journeyman Geek

snort

@bwDraco took an afternoon

You quite literally find something that annoys you

then design a solution

then build and iterate

bwDraco

@JourneymanGeek That reminds me of that secure erase script I wrote the other day. I was repeatedly erasing a particular drive and didn't want to memorize the commands to do it so I wound up trying to automate it.

Journeyman Geek

and you made a thing!

I mean here, it was literally "It would be nice if I could turn it off" ... "with less wires" "maybe individually" "and with a remote" "and OOH RELAYS"

I order a bunch of stuff, mess with it, fail, go on vacation, suddenly realise I missed something stupid, go for CCNA classes, test, iterate, test more.

bwDraco

So... I think I may have a better idea of what's going on. I have certain professional obligations (namely, post-processing photos) which require significant time and effort. Often, I cannot get myself to do this without running close to the deadline (due to a lack of willpower).

At the same time, I have other projects in mind like LXD, Docker, or JS, and I feel I cannot start working on them until I get more important stuff out of the way. As a result, I wind up being unable to assign any significant amount of time to these projects.

A corollary of this is that if I had more willpower, I'd feel freer to do what I truly wanted to do.

gist.github.com/bwDraco/11fbf52336b2ff1d67445a8e32a2d2d5

Journeyman Geek

7:55 AM

(also, my documentation skills suck, but....ehhhhh)

@bwDraco also, I'd point out, building stuff isn't always practical or even useful. ;p

bwDraco

@JourneymanGeek What's written for personal use is different from something that's published. I felt it would be useful to share it and this led me to add error checks and prompts to make it user-friendlier.

Journeyman Geek

and... that's exactly how and why you build things ;p

bwDraco

I was taught to do extensive commenting in college, and it's just how I code.

e.g. if this wasn't going on GitHub, I wouldn't have bothered with checking if the argument is a block device.

Journeyman Geek

which is also fine

everyone occationally does a dirty hack

@Bob that happens when the dirty hack fails, and something catches fire.

Bob

8:11 AM

@JourneymanGeek I should probably tape my lights back up sometime, come to think of it

(the adhesive failed...)

Journeyman Geek

._.

ODDLY ENOUGH!

I used a ton of double sided tape on one of these

the other is sitting atop my monitor cause the attachment point I made of wire failed.

bwDraco

@JourneymanGeek USB-powered Delta fan!

Delta case fan connected to USB via voltage regulator

►

Bob

@bwDraco ahhh, typical school code commenting :P

Journeyman Geek

boost converter?

Bob

somehow manages to teach people to spam comments everywhere except where you actually need to

or, as one of my classmates called it, MORE GREEN SALAD!

Journeyman Geek

8:13 AM

I approve of the use of crocodile clips

Bob

@JourneymanGeek daiso? :P

Journeyman Geek

@Bob oddly... no

Bob

@JourneymanGeek sprung for the 3M stuff? :P

Journeyman Geek

nope

was lying around

I quite literally ordered everything but the wood off of DX. and then realised I had no good way of fixing the back board to the VESA plate I was using for attachment

bwDraco

@JourneymanGeek Adjustable-voltage SEPIC boost-buck voltage regulator.

Journeyman Geek

8:15 AM

close enough

I'm not an EE. I just know which end of a soldering iron not to hold

Bob

@JourneymanGeek you're obviously not qualified for stock photos

bwDraco

pololu.com/product/2572

Bob

@bwDraco protip: a comment like # Print header is largely useless (unless you're using it for some kind of modularisation, which is usually better served by splitting into functions, though small scripts can be an exception). However, the hdparm command, the meat of your code, is something you do want to comment

did you pull the hdparm command from somewhere? link to the reference so if you come back later you know why you'd use that particular command

why do you need to --security-set-pass? (in this case, the link to the reference would explain that too, assuming you used the ata wiki)

those would probably be the comments that are of more value if you come back to it in 3 years

(I'm not saying you need to edit that file - it's quite well written and probably better documented than anything I've done recently - but the commenting of everything but that last command struck me as funny :P)

bwDraco

Updated.

Bob

@bwDraco heh, that actually shows something interesting: locality of information. I had completely missed that you had the link up the top

which reminds me I need to add some doc-comments -_-

Transcript for