<py-bb>
Hi there I'm looking to try something on an ARM CPU - obviously I work mainly with x86 stuff which I've come to know. Mainly microbenchmarks to make sure my theory is good (what I can get out of them compared to what I think I can and such). This takes very little CPU time (compiling takes more) would anyone be willing to help me with this? Depending on the chip it can be done virtualised, otherwise a live boot and "sudo apt-get install o
<py-bb>
penssh-sever" you run a command to let me in that's it
_whitelogger has joined #arm-netbook
<py-bb>
The A77s look worth considering
_whitelogger has joined #arm-netbook
lkcl has quit [Ping timeout: 258 seconds]
_whitelogger has joined #arm-netbook
<buZz>
py-bb: anything is a good place to start :)
<buZz>
maybe get a raspberry pi?
<py-bb>
buZz: yeah I do - I figured a chunk here would be Raspberry Pi users but I was hoping to find some people who work with different chips and such. Help me find my feet
<buZz>
sure
<buZz>
but you arent asking anything specific really?
<py-bb>
I'd love to ditch the x86-64 tax alone (takes a lot of energy to decode those variable length instructions and the A77 looks mean as fuck)
<buZz>
'i want to play with ARM' <- then just go do that? :D
<py-bb>
Helpful?
<buZz>
what do you mean?
<buZz>
ARM has been commonly available to everyone that wants to play with it for many years now
<py-bb>
Yeah I really want to buy like 25 different beagle boards, faff around, then bin them. Not at all wasteful or anything
<buZz>
why not just use them?
<py-bb>
I thought maybe others would have bought them and maybe have one spare they could let me SSH into for a few hours. This isn't unusual .... it's not like Smeagle and the ring.
<buZz>
you sounded like you wanted to use them
<py-bb>
Not if they're trash *cough rPi*
<buZz>
some VPS providers also rent out ARM shells
<buZz>
also, qemu can emulate a ARM cpu on your x86
<py-bb>
Yeah I can read....
<py-bb>
And emulating will indeed help me evaluate the performance of a CPU
<py-bb>
You're being really unhelpful buZz I'm not sure if you're doing it deliberately though, if you're not, forgive my sarcasm there.
<buZz>
gee, sorry for not offering you access to a system you didnt pay for
<buZz>
which are readily available and very cheap
<buZz>
i can imagine i look really unhelpful but i'm mostly just suprised you'd even ask this
<py-bb>
buZz: around me right now are a bunch of PCs which are mostly idle - if someone can get use out of them (reasonably, eg "oh yeah I'm benchmarking how fast it can mine Shitecoin" <-- bad) I have no objections to setting up a VM or booting onto a memory stick for them (depending on trust, or use case, I don't mind pulling out a hard disk so they can't poke around)
<py-bb>
I don't think that's unreasonable. As you do I don't think you could help.
<buZz>
well, stick around, maybe someone gifts you access
<py-bb>
Rather grandiose term for "doesn't bother collecting the pennies it'd be worth"
<buZz>
well, as said, raspberry pi is cheap and available
<py-bb>
And trash
<buZz>
why?
<py-bb>
It used to have no DMA so any I/O crippled the CPU with wait time, the CPU is also trash grade
<buZz>
even pi1 has DMA
<py-bb>
Then it has no excuse for being that shit
<py-bb>
I assumed - my bad
<buZz>
you dont need 'good IO' for learning
<py-bb>
Depends what you're learning
<py-bb>
If you're learning "the feasibility of running any sort of workload on this thing that accesses storage" - then you do learn
<buZz>
well, you said you're wanting to just try opcodes on the cpu
<py-bb>
FS cache still has to warm up
<buZz>
ARM isnt really a high IO workhorse in most setups
<py-bb>
Yeah for throughput and shit, Fog doesn't cover ARM
<buZz>
Fog? whats that
<py-bb>
I'm sorry, I should have been clear: most programs take input.
<py-bb>
Agner Fog?
<buZz>
never heard of
<py-bb>
You could learn...
<py-bb>
It's an uncommon name
<buZz>
its not googleable it seems
<buZz>
unless you ment a person
<buZz>
> Agner Fog currently works at Technical University of Denmark. Agner does research in Evolutionary Anthropology, Cultural Anthropology, Social systems and Computer Science. His current project is 'Theory of cultural change based on evolutionary psychology, evolutionary biology, and cultural evolution.'
<py-bb>
Jesus
<py-bb>
Just "google" Agner Fog CPU guide" or something - they're legendary.
<buZz>
well, get one? :) 96boards.org has cool machines
<py-bb>
We went through this already.
<py-bb>
Seriously are you trolling me? I really can't tell
<buZz>
i dont see why you wouldnt just get one?
<buZz>
you clearly sound motivated to play with it
<py-bb>
Okay
<py-bb>
So rather than say what I've already said, in good faith, I'll expand it - please don't /deliberately/ waste my time though
<buZz>
ok, your sundaymorning time is valuable, i get it
<py-bb>
Most people don't actually like Intel. They charge a lot. x86-64 is very complex to decode and this costs you (rule of thumb) 40% in something, eg power, transistors, performance, this is why atoms suck
<py-bb>
If arm are not trash there might be people - perhaps here - who also have no love for Intel and would help me evaluate by letting me use their idle hardware and the 0.005 pence of power it'd use (max) for a few hours sometime to see if Arm is actually a real contender.
<buZz>
my atom server takes 25w total with ~30 containers running, pretty decent for me
<buZz>
that C3000 line is really quite decent, imho
<py-bb>
If I were to try and buy even a third of different ARM family chips - I'd become a hoarder, and I'm taking the risk here - I'd love to (just for power savings) switch - but right now it uses a lot less power yes, but it also does fuck all useful work. So...
<py-bb>
Your atom is shit, don't be an idiot by calling it a "server" and thinking it does useful work, they're all shit.
<buZz>
ok
<py-bb>
Also 30 idle containers is not that big of a deal for Intel any more - it's not bad but not a big deal. It's AMD for which it's not a big deal. They have much better support for shit like this
<buZz>
they arent idle
<py-bb>
Awesome thanks man you've given me a new purpose: "do shit on atoms" - also they still pay the tax and suck BTW. It will get trounced by an old sandy bridge chip any day of the week no matter how modern then Atom
<py-bb>
It'll be as one sided as a gang rape.
<buZz>
ok?
<buZz>
if you want to play with toys of other kids, you may want to try not being insulting ;)
<buZz>
else you'll just have to buy them yourself
<py-bb>
You're too cheap to let me by your own volition and I'm sorry if my education that I'm gifting you isn't polite enough.
<buZz>
for over 2 years now, and its not '30 idle containers' like you imagines
<buZz>
s/s$/d/
<py-bb>
What an amazing feat.
<py-bb>
30 containers running top? Damn
<buZz>
i'm happy thats the limit of your imagination :P
<py-bb>
Or 30 containers running something you have to run for people you hate
<buZz>
maybe once your irc account is -5- days old you stop behaving like a spoiled kid?
<py-bb>
Also you can run 25,000 if you like (should have enough PIDs) - it'll do it, just not if they're idle.
<py-bb>
Sorry you are what you eat, and I eat a lot of spoiled little bitches, if you were younger ....
<py-bb>
Also I have other nicknames....
<buZz>
i get that, if you stay so offensive, its hard to see you getting any assistance from anyone
<buZz>
but maybe your school loves it, what do i know
<buZz>
¯\_(ツ)_/¯
<py-bb>
You started it :)
<py-bb>
You wasted my time and I still tried to help you, even though you find it unthinkable humans want to help each other.
<py-bb>
I wont feel bad for trying
<buZz>
help -me- ?
<buZz>
help me doing what
<py-bb>
not being a noob. I told you about stuff, linked you pictures of the A77 cores, explained why historically arm chips have been trash and how "running 30 containers" is a useless metric, when machines start trashing badly with paging it's not uncommon to see load averages of several thousand (seeing them is uncommon :P)
<py-bb>
Anyway good luck in life.
houkime has joined #arm-netbook
<oh2ftg>
Speaking of ARM and netbooks, what do you guys think of the PineBook Pro?
<py-bb>
Probably shit
<oh2ftg>
For the price it looks interesting
<oh2ftg>
and the Rockchip SoC they are using is decent. Could be a fun toy
<py-bb>
What do you mean by decent?
<oh2ftg>
But considering that I have a pile of x86 toylaptops I'm unsure if I want more :P
<py-bb>
I'd use the benchmark of a 2011 Sandy Bridge i3
<oh2ftg>
If there was some exposed GPIO inside it would be more interesting
<oh2ftg>
Honestly depends on what you are after.
<oh2ftg>
But yeah that's likely a decent benchmark to compare against for a laptop usable for todays websites
<oh2ftg>
Like for other use than just IRC and SSH terminal use :P
<py-bb>
Then I'd still use power
<oh2ftg>
Then again, most of my ARM stuff is either microcontrollers or headless stuff doing some rather limited tasks like serving me samples over the wired ethernet
<py-bb>
Sandy Bridge was a great arch btw, Nehalem was alright but a good benchmark as it worked very well. A slightly later one may be better though (SNB was rushed out a bit Ivy Bridge cleaned it up if you know this stuff)
<py-bb>
BTW
<oh2ftg>
In all honesty I have not been that much performance oriented ever.
<py-bb>
https://i.imgur.com/PL9FheF.png this is a dual core Sandy Bridge laptop (9 years.... I've replace the keyboard 6 or 7 times) battery 5....
<oh2ftg>
At most Price/performance and that usually means being a bit trailing edge
<oh2ftg>
Especially when I have ~never bought a computer new.
<py-bb>
oh2ftg: it has it though... which is why I'm bringing it up that's a dual core (4 thread) 2.2ghz with no boost, it uses 0.14 watts for the GPU (I recorded that to) and the package uses 2 watts more for L3 and DRAM refresh on top of that.
<py-bb>
Now performance...yeah it has some but it's also power efficient
<oh2ftg>
I think one of my fat clients has some recent-ish i3
<oh2ftg>
I really should boot it up, throw ubuntu or debian on it and take a look at making it in to a websdr with my limesdr
<oh2ftg>
But that's not really the topic for this channel
<py-bb>
I'd wanna see if you have an ARM chip that can compete with SNB the A77 that prompted me to join looks mean, but the super low power ARMs end up using more power simply because they need so long to do the work.
<oh2ftg>
I think the "server class" ARM stuff was marketed on power efficiency
<oh2ftg>
But who knows
<oh2ftg>
You could find some vps provider which offers A77 based stuff.
<oh2ftg>
As I don't have anything with A77 and can't really make the effort for setting up a shellaccount on one anyway.
<py-bb>
oh2ftg: I have yet to see one that isn't trash, that's why I'm asking around here for someone to help me (maybe in return for something) ssh into some ARM stuff and test it out.
<py-bb>
I'm not looking for server gear I just want to see if it qualifies as "trash" or "not trash" - last time I tried (circa 2014) it wasn't feasible.
<oh2ftg>
Well all the truly high power stuff is gona be in the server bracket
<oh2ftg>
As most other's are not targetting a "desktop" market
<oh2ftg>
at most you see stuff like Nvidia Tegra series in nintendo consoles and Nvidia Shield
<oh2ftg>
And then stuff in arm based HTPC/set top box stuff
<oh2ftg>
and the later one is not really that powerful
<py-bb>
It depends on the design, there's only 3 major ARM designs out there
<py-bb>
Adding more cache or a better branch predictor, more cores, that doesn't change the micro-arch (much)
<oh2ftg>
And then there's Nvidia Jetsons, but those ride mostly on the GPU cores on them for CUDA and other acceleration in embedded systems.
<oh2ftg>
There really aren't any ARM cpu's/soc's targettign "desktop"
<oh2ftg>
Maybe the chromebooks, but those are more about optimizing batterylife and they likely throttle on sustained loads due to thermal design.
<py-bb>
GPU doesn't really count, like the "RAspberry Pi is as powerful as a PS3 bullshit" - in FLOPS and theoretically - as a general purpose thing it's as one sided as ... something really one sided
<py-bb>
Yeah thermal again isn't architectural. And that's not a very fast process either. Lots and lots of CPU cycles happen before the temperature registered by any of the sensors changes much
<oh2ftg>
And the raspberry pi GPU is locked down tighter than satans asshole
<oh2ftg>
And that's why I mentioned sustained loads.
<oh2ftg>
And thermal is the main reason for throtling on cellphones, one of the main target markets of higher performance ARM cores currently.
<py-bb>
Sustained loads is an unfair benchmark. I want to know what the architecture can do. I wouldn't expect a netbook to be able to represent decently cooled things with heatsinks that weight more than they do. C'mon man that's unfair
<oh2ftg>
Reportedly the latest ARM SoC's by Apple are fairly beefy, but I don't recall any data off hand.
<oh2ftg>
And telecoms stuff can be hilariously powerful. Like the Texas instruments telecoms stuff with some ARM cores and then some REALLY beefy DSP with switch blocks
<oh2ftg>
like 100Gbit switching fabric and tons of FAST DSP
<oh2ftg>
And some 1-2GHz ARM cores for running an OS for directing the show.
<py-bb>
I don't care about GPU as I said and yeah that's a different thing though the ARM chips are not doing much heavy lifting in those things
<oh2ftg>
Like said, most ARM stuff targets a different market with different interests.
<oh2ftg>
The most performance oriented is the server market.
<py-bb>
There's like 3 designs like I said.... if you get one of them, it's the same IP core (modulo device specific tweaks) and will behave like the others, clock for clock, modulo cache alterations (you can configure L2 and below cache almost always)
<py-bb>
They'll use the same page walkers, TLBs, cache coherency protocol, scheduler, decoders, ect ect
<py-bb>
Make sense? It's just number, process, power and frequency that vary. So if you have one of a model on a really modern process for a phone, it may thermally throttle and have a higher clock, but clock-for-clock it behaves identically to an older one of the same core (except for cache stuff if it has different cache sizes, but it's still very close)
_whitelogger has joined #arm-netbook
oh2ftg has quit [Ping timeout: 245 seconds]
attah has quit [Remote host closed the connection]
isacdaavid has joined #arm-netbook
peterM has joined #arm-netbook
linmob has quit [Quit: No Ping reply in 180 seconds.]
lkcl has joined #arm-netbook
houkime has quit [Remote host closed the connection]