#robotics Logs
May 10 2017
#robotics Calendar
12:06 AM anniepoo: now setting up snail rig to animate
12:09 AM solol: kinda funny that you are making a snail
12:09 AM solol: there any purpose for it?
12:09 AM orlock: any purpose for it?
12:09 AM orlock: you are new here i see
12:09 AM orlock: :)
12:09 AM anniepoo: 8cd World Domination? Very slowly...
12:11 AM z64555: well, at least the denzens will have plenty of time to prepare/get used to the new ruler
12:11 AM Snert: The Snail That Ate Manhattan
12:14 AM anniepoo: burp
12:22 AM Snert: of all the robotics projects I've seen so far, your snail has about the most commercial idea and doable and lovable.
12:22 AM anniepoo: 8cD
12:22 AM anniepoo: Thanks!
12:22 AM solol: what's your snail do
12:22 AM anniepoo: loves
12:22 AM anniepoo: >8@_
12:25 AM anniepoo: ok, more seriously-
12:25 AM anniepoo: I have a long term project to make robots that interact with children commercially
12:25 AM anniepoo: the snail is essentially my learning project
12:25 AM anniepoo: and a research tool
01:11 AM rue_house: I took a picture of my fibrator in the dark
01:11 AM rue_house: but right now I cant find the camera
05:38 AM Tom_L: orlock, next time get his full ip
08:02 AM solol: https://www.theregister.co.uk/2017/05/10/why_ai_means_the_end_of_cpudominated_computing/
08:49 AM tsglove: https://www.youtube.com/watch?v=3rqGJatwkyQ
08:49 AM tsglove: looks interesting!
09:06 AM solol: why can a computer only handle thousands to millions of brute force attempts per second if it is supposed to be rated at like 500GFLOP to 9TFLOP?
09:06 AM solol: what are sockets? FLOPS = sockets x cores/socket x cycles/sec x FLOPS/cycle... https://en.wikipedia.org/wiki/FLOPS
09:07 AM deshipu: sockets are the sockets that you plug the cpu into
09:07 AM deshipu: back in the day
09:08 AM deshipu: today, when they are all soldered to the pcb, you can just count the chips
09:08 AM solol: what do you mean
09:09 AM solol: but why can't you only really handle thousands of operations or so per second with hundreds of gigaflops?
09:09 AM solol: why can you*
09:10 AM solol: or even like 9TFLOP can brute force a million or two entries per second... that's nowhere near a multiple of trillions
09:12 AM tsglove: solol, go code it up in assembly language, and you'll understand.
09:12 AM robopal: in parallel?
09:13 AM solol: it's not just a feed through process?
09:14 AM solol: i assume it's all done in parallel, but aren't you limited by cores at that point?
09:14 AM z64555: I have no clue what you're trying to get at
09:14 AM solol: take for example brute forcing
09:15 AM solol: you could write a simple bitshift algorithm * n for entry, right?
09:15 AM solol: then just run it at whatever ghz it goes
09:16 AM deshipu: the modern hashing algorithm are chosen so that they can't be calculated too fast
09:16 AM deshipu: certainly not in a single cpu instruction
09:16 AM solol: so what if there is no hashing
09:16 AM solol: like just creating entries in a text file
09:17 AM deshipu: what are you trying to brute force then?
09:17 AM solol: it was just an example
09:17 AM solol: i want to see output matching the limits of a cpu/gpu
09:17 AM deshipu: the point of brute forcing is to find a value that hashes to the same as what you have
09:17 AM deshipu: writing to a text file is going to be limited by the speed of the disk and the data bus
09:18 AM z64555: What's the point of trying to see output that matches the limits of the cpu? stress testing?
09:18 AM solol: to run at realtime?
09:18 AM z64555: boi
09:18 AM z64555: that's not how realtime processing works
09:18 AM solol: so disk size would be RAM.. so could you fill up 100gb/sec?
09:19 AM deshipu: ram is connected through a data bus to the cpu too
09:19 AM deshipu: and it has limited speed too
09:19 AM deshipu: that's why cpus have l1 and l2 caches
09:19 AM solol: so how do you get a cpu/gpu to run at it's maximum speed?
09:20 AM solol: just recursion?
09:20 AM solol: or overwrites?
09:20 AM deshipu: no idea what you mean
09:20 AM z64555: you're missing the point, solol
09:20 AM solol: n + 1
09:20 AM deshipu: just clock it at its maximum speed
09:20 AM deshipu: it doesn't matter what it does
09:20 AM solol: how do i get realtime values of 9TFLOP out of my gpu?
09:20 AM deshipu: values of what?
09:21 AM z64555: how fast's your buss
09:21 AM solol: speed
09:21 AM rue_house: its called latency
09:21 AM z64555: *fast is
09:21 AM rue_house: and nobody but me gives it a second thought
09:21 AM deshipu: rue_house: latency is a spearate issue
09:21 AM rue_house: ^^ see?
09:21 AM rue_house: :)
09:22 AM solol: they rate CPU's/GPUs these days at hundreds of GFLOP to almost 10 TFLOP
09:22 AM deshipu: since we don't know about any kind of *input*, it's meaningless to talk about latency
09:22 AM rue_house: today people like really fast digital equipment that just happens to take a long time to produce results
09:22 AM solol: how would you get data up and running at those speeds
09:22 AM z64555: you dont
09:22 AM rue_house: serial
09:22 AM deshipu: solol: datas don't run
09:22 AM rue_house: processors themselfs are gonna become serial buss
09:22 AM solol: what do you mean... how do they calculate those #'s and slap it on the box?
09:22 AM rue_house: so each core will have a/few serial busses
09:23 AM solol: you guys don't know what im talking about?
09:23 AM z64555: you're difficult to understand, yes
09:23 AM rue_house: everything is serial today
09:23 AM deshipu: I doubt you yourself know
09:23 AM deshipu: rue_house: I am pretty sure the memory buses are parallel in modern cpus
09:23 AM rue_house: they are, but everything else is serial
09:24 AM solol: http://www.gamersnexus.net/hwreviews/2050-msi-radeon-r9-390x-gaming-review-and-benchmark?showall=1
09:24 AM rue_house: memory will go serial
09:24 AM rue_house: just wait
09:24 AM deshipu: I sense a no true scottsman fallacy
09:24 AM solol: scroll down it'll show TFLOPs
09:24 AM z64555: explain TFLOPS to me, in your own language
09:24 AM solol: 5.9TFLOPs
09:25 AM solol: trillion single precision floating point operations per second
09:25 AM solol: single precision = 32-bit numbers
09:26 AM solol: Examples of integer operation include data movement (A to B) or value testing (If A = B, then C).
09:26 AM z64555: mkay. Does this rating include SIMD or similar instruction sets?
09:27 AM solol: i assume it's anything that has a register value
09:27 AM z64555: that's not what I asked
09:27 AM solol: im not familiar with SIMD
09:27 AM solol: so yes it should include instruction sets
09:28 AM deshipu: solol: so you want to run a benchmark to measure how much your graphics card can calculate?
09:28 AM z64555: -_-
09:28 AM solol: yeah basically, but not just a benchmark, making it into a work environment
09:28 AM deshipu: solol: how?
09:28 AM rue_house: vector processors are different
09:28 AM z64555: SIMD< SSE, SSE2, MMX, are all sets that work on small vectors of floating points
09:29 AM deshipu: solol: not sure what you mean by "work environment"
09:29 AM solol: yeah it should include those z64555
09:29 AM z64555: such as a dim 4 vector
09:29 AM solol: any realtime use deshipu
09:29 AM z64555: If so, then divide that number by 4
09:29 AM z64555: Next, are this operations including all cores?
09:29 AM rue_house: mmx instruction: if this value is larger than that value and this bit is set then increment that number
09:30 AM solol: it'd have to be including all cores to determine a maximum value
09:30 AM deshipu: rue_house: gpus don't have mmx instructions :)
09:30 AM rue_house: I forget the name of the instruction
09:30 AM z64555: then divide that by the number of cores again
09:30 AM z64555: 5.9 T / 4 / # of cores
09:31 AM solol: why would you be dividing by cores?
09:31 AM rue_house: and why do you prefer flops from ips
09:32 AM solol: FLOPS = sockets x cores/socket x cycles/sec x FLOPS/cycle
09:32 AM solol: FLOPS = sockets * cores/socket * cycles/sec * FLOPS/cycle
09:32 AM rue_house: but most stuff isn't math
09:32 AM z64555: that's the same equation
09:32 AM rue_house: admit it
09:32 AM z64555: benchmarking is ideal-case testing
09:32 AM rue_house: most of what processors do is shovel data
09:32 AM rue_house: things like graphics cards do the math work
09:32 AM solol: so they should still be shoveling data at ghz speeds
09:33 AM rue_house: flops wont tell you that
09:33 AM z64555: Those numbers you are reading are the "red line," that CPU is not going to get any better than that
09:33 AM rue_house: ips / mips will
09:33 AM solol: you don't get down to thousands very easily from billions
09:33 AM rue_house: front side bug speed
09:33 AM rue_house: heh
09:33 AM * rue_house tries to type and eat breakfast more carefully
09:34 AM rue_house: and for the longest time, intel decided that a fast core was more important than a fast front side bus
09:34 AM solol: they don't use buses anymore
09:34 AM solol: haven't for 10 years
09:34 AM rue_house: no, they put he gpu on the cpu die
09:35 AM rue_house: when I get a motherboard I look at how fast the memory<->cpu path is
09:36 AM rue_house: so far AMD alwasy wins
09:37 AM solol: so how would you be working at speeds that max out the memory->cpu path since obviously hundreds of GFLOP should be able to do that
09:37 AM solol: be/get/
09:38 AM solol: pretty sure the memory->cpu path is usually < 100GB/sec
09:38 AM rue_house: heh, not when intel was using a 533Mhz FSB
09:38 AM rue_house: and AMD was 1.3Ghz
09:38 AM solol: im talking about nowadays
09:39 AM solol: quad 4ghz processors
09:39 AM rue_house: its too bad there aren't good wait state monitors
09:40 AM solol: you're straying from the point here
09:40 AM rue_house: I'd be willing to bet that current processors spend almost 50% of their time waiting for cache reloads, fetches and writes
09:40 AM z64555: pretty much
09:40 AM rue_house: DOS fits in the L1 cache of any new processor
09:40 AM solol: what does that mean rue_house ?
09:40 AM rue_house: its non-multitasking and WHOLY (*&^*&% IS IT FAST
09:41 AM solol: so it's how fast?
09:41 AM rue_house: (*&^*&%
09:41 AM solol: i didn't think dos was ever slow though
09:41 AM rue_house: not when its running from L1 cache
09:42 AM solol: it could scroll across the screen much faster than you could see it even on like a <100mhz computer
09:42 AM z64555: I'd ask about relevance, but there'd be no point to
09:42 AM rue_house: Its usually fulyl booted before an lcd screen has worked out what the refresh rate is
09:43 AM solol: idk. i want to get from GFLOP to RAM and back or something in realtime
09:43 AM solol: what are my limitations?
09:43 AM rue_house: io, and bus speeds
09:44 AM rue_house: do new x86 processors have a cycle count register in them?
09:45 AM rue_house: it would be intersting to see how many wait states are in something like a memory read/writeback
09:47 AM rue_house: Memory Type :
09:47 AM rue_house: DDR3
09:47 AM rue_house: Memory Speed (Non-OC) :
09:47 AM rue_house: 2133/1866/1600/1333
09:47 AM rue_house: hmmmm
09:48 AM rue_house: the DDR3 will be 64 bit?
09:48 AM rue_house: now, you cant go by 2.1Ghz either
09:49 AM rue_house: it takes multiple states to just set up a read/write
09:51 AM rue_house: huh, so typical is 1 pcie16, 1 peicex1, and a pci slot
09:51 AM z64555: that's typical?
09:51 AM rue_house: apparently
09:52 AM rue_house: ok I have to be at the worksite by 8
09:52 AM z64555: I guess they don't like the on-board ethernet. Or maybe that's tied into the pci bus
09:52 AM rue_house: it'll take 10 mins traval
09:52 AM rue_house: hmm
09:52 AM rue_house: z64555, nono, I just mean for those 3 thigns
09:53 AM rue_house: everyone has 1G/ 10G onboard ethernet
09:53 AM z64555: a GPU card explains the pcie16, and maybe the x1, too
09:54 AM z64555: that last pci slot has me clueless, tho
09:54 AM rue_house: things like a video capture card
09:54 AM rue_house: or a 56k modem
09:54 AM rue_house: or an oscilloscope card
09:54 AM z64555: 56K modem is not "typical" these days
09:54 AM rue_house: or an industrial robot controller
09:55 AM z64555: everyone uses ethernet, or *maybe* a wifi card
09:55 AM rue_house: hmm, I dont remember how old my hardware is
09:56 AM rue_house: still behaving tho
09:56 AM z64555: not quite older than dirt
09:56 AM z64555: :)
09:56 AM rue_house: ooo I think the drives are newer than the rest
09:57 AM rue_house: wow 'supermicro' makes a LOT of boards
09:58 AM rue_house: hah 2TB for $100
09:59 AM rue_house: oo 6TB, geez, install windows and have room left
09:59 AM z64555: lol
10:00 AM rue_house: oooo WD 5 year warranty
10:02 AM -!- #robotics mode set to +o by ChanServ
10:03 AM rue_house changed topic of #robotics to: If you ask a question, you must wait for the answer. | http://tom-itx.no-ip.biz:81/~webpage/index.php | K is for kilobite (which megahurts!)
10:03 AM -!- #robotics mode set to -o by rue_house
10:17 AM z64555: gotta remember to grab a copy of FreeRTOS
10:18 AM * z64555 writes it on the baord
01:18 PM tttb2 is now known as tttb
08:29 PM zhanx: great i am too big for this skydive place
08:30 PM zhanx: they need a bigger plane
09:41 PM Tom_L: why the hell would you wanna do that again anyway?
09:42 PM Tom_L: just to get over it?
09:48 PM mun_ is now known as mun
10:00 PM rue_bed: I need to work on the 132'
10:00 PM rue_bed: 12
10:00 PM rue_bed: 12' mecha stuff
10:00 PM rue_bed: any my typing
11:02 PM rue_shop3: so, I can measure the force on a servo
11:02 PM rue_shop3: the question is, can I use that to regulate it
11:03 PM orlock: via current draw?
11:03 PM rue_shop3: no no no
11:03 PM rue_shop3: have you not bveen listening to my babbaling for the last 6 months?
11:04 PM rue_shop3: you use a current detector (not meter, but a digital one) and you measure the drive time on the motor
11:05 PM orlock: ahhh, so two values
11:05 PM rue_shop3: ?
11:05 PM orlock: which would make sense, as you get power and time to complete the task
11:06 PM orlock: ?
11:06 PM Tom_L: 132' holy crap batman!!
11:06 PM rue_shop3: you dont need to know what the current is
11:06 PM rue_shop3: you just need to know if the motor is on or not
11:09 PM SpeedEvil: you don't know which way it's active
11:09 PM orlock: so you measure how long the motor is on for?
11:09 PM orlock: and you get a time
11:09 PM orlock: there has to be some other known value though?
11:14 PM SpeedEvil: Also, knowing it's on doesn't tell you torque, unless it's zero.
11:14 PM SpeedEvil: the speed is zero
11:27 PM rue_shop3: hahah
11:28 PM rue_shop3: SpeedEvil, so you lot have NOT been listening to me
11:28 PM rue_shop3: ok
11:28 PM rue_shop3: anyhow,
11:28 PM rue_shop3: this is funny, using this absolute-reading piezo pressure sensor, its picking up my heartbeat
11:32 PM rue_shop3: I'm finding that the biggest problem is setting up the forces right on the peizos
11:32 PM rue_shop3: its like you have to be flexing them sideways
11:35 PM Tom_L: how do they differ from pressure transducers?
11:37 PM rue_shop3: still learning
11:38 PM rue_shop3: the 2nd sensor I set up seem to have a lot of leakage, which wrecks the output
11:38 PM rue_shop3: the first one I set up holds its value for minutes at a time
11:38 PM rue_shop3: the 2nd one, about 4 seconds
11:38 PM rue_shop3: so, hmm
11:38 PM rue_shop3: oh
11:38 PM rue_shop3: these are cheap as all hell
11:38 PM rue_shop3: they are just piezo elements
11:39 PM rue_shop3: if you put a capacitor across a peizo, the charge on it will be equiv to the pressure
11:39 PM rue_shop3: if the capacitor is the gate of a fet, you can use the source/drain to tell you the charge on the cap with 0 leakage current
11:40 PM rue_shop3: even 10M is enough to cause these to drift off in a second
11:40 PM rue_shop3: so that first peizo setup must be in the 100M ohm range or so
11:41 PM rue_shop3: if you put a peizo across an led and tap it, the led flashes :)
11:42 PM rue_shop3: part of the challange to getting it to be usefull is the bias for the transistor
11:42 PM rue_shop3: a 100M ohm bias circuit is a challange
11:49 PM rue_house: ah wait a sec
11:49 PM rue_house: this isn't a force sensor
11:49 PM rue_house: this is a deflection sensor
11:49 PM rue_house: so is a magnet and coil
11:49 PM rue_house: so
11:50 PM rue_house: if I take a speaker and put a series cap on it, and use the mosfet gate as a second sense series cap, then that will work too
11:50 PM rue_house: damnit I dont want supper
11:51 PM rue_house: the gate capacitance of a 2n7000 is ...
11:52 PM rue_house: about 20pf
11:52 PM rue_house: so even 200pf would be ok