<muurkha>
sorear: yeah, I didn't mean the 6502 would be a good DSP, quite the contrary
<muurkha>
the reason for a 6502 is that you can get a whole CPU with sort of reasonable cycle efficiency into 3000 transistors
<sorear>
kind of fudging the numbers by not counting memory
<muurkha>
(as long as you have enough fast RAM backing it up, and 3000 transistors of 6T SRAM is only 500 bits, not even enough for a whole 6502 zero page)
<muurkha>
heh, yeah
<muurkha>
a few years back I sketched out a simple minimal 12-bit stack CPU, and one of the things that surprised me about the exercise was that going bit-serial only cut it from 1000 NANDs down to 600 or so. I was expecting a bigger improvement in size for the order-of-magnitude reduction in speed
<muurkha>
but that's about 6502-sized
<muurkha>
(and, again, fudging by not counting memory)
<sorear>
the 6502 was severely pin constrained, not just transistor constrained, whatever you're doing now with on-chip memory you can probably make it twice as fast by going harvard
KombuchaKip has joined #riscv
jacklsw has joined #riscv
shicz2 has quit [Ping timeout: 265 seconds]
shicz2 has joined #riscv
vagrantc has joined #riscv
shicz2 has quit [Ping timeout: 252 seconds]
shicz2 has joined #riscv
shicz2 has quit [Ping timeout: 256 seconds]
shicz2 has joined #riscv
shicz2 has quit [Ping timeout: 265 seconds]
shicz2 has joined #riscv
ntwk has quit [Ping timeout: 265 seconds]
jwillikers has quit [Remote host closed the connection]
jwillikers has joined #riscv
<muurkha>
true
ntwk has joined #riscv
jwillikers has quit [Remote host closed the connection]
vagrantc has quit [Quit: leaving]
frost has quit [Ping timeout: 252 seconds]
jacklsw has quit [Read error: Connection reset by peer]
shicz2 has quit [Ping timeout: 265 seconds]
shicz2 has joined #riscv
jacklsw has joined #riscv
kailo has joined #riscv
PyroPeter has quit [Ping timeout: 240 seconds]
shicz2 has quit [Ping timeout: 240 seconds]
shicz2 has joined #riscv
PyroPeter has joined #riscv
BOKALDO has joined #riscv
riff-IRC has quit [Remote host closed the connection]
<dh`>
I've often wondered what a system made up of a few thousand 6502s would be like to use
<enthusi>
slow :)
<enthusi>
but there was that propeller SoC with a many-small-units aproach a few years back?
<enthusi>
I coded ALOT of 6502 assembler and which I enjoy the ISAs structure alot, it is not very well suited for data-exchange.
voltron has joined #riscv
jacklsw has quit [Read error: Connection reset by peer]
shicz2 has quit [Ping timeout: 265 seconds]
shicz2 has joined #riscv
<gordonDrogon>
and here am I looking to replace some 6502 systems with RISC-V ;-)
<sorear>
you'd clearly need support chip(s) of some kind, and the programming model depends a bit on exactly what those support chips are
<gordonDrogon>
for my thing? not really - it's (now) more a retro software hobby thing. he highest level language is bcpl. I just need a serial port although video would be nice at some point ...
<gordonDrogon>
support chips are the bane of the "retro new" world though.
winterflaw has quit [Ping timeout: 276 seconds]
Andre_H has joined #riscv
<sorear>
"I've often wondered what a system made up of a few thousand 6502s would be like to use"
vagrantc has joined #riscv
chripo has quit [Quit: ""]
chripo has joined #riscv
bauruine has quit [Remote host closed the connection]
cousteau has joined #riscv
dlan has quit [Ping timeout: 256 seconds]
dlan has joined #riscv
yongxiang has quit [Ping timeout: 245 seconds]
jjido has joined #riscv
englishm has quit [Ping timeout: 245 seconds]
englishm has joined #riscv
yongxiang has joined #riscv
<muurkha>
gordonDrogon: oh?
<muurkha>
sorear: yeah, you can't really just hook up two 6502s to shared address and data buses
<sorear>
4320x might be fun since those *did* have a bus arbiter
<muurkha>
not familiar
RNeese has joined #riscv
<RNeese>
hey guys
<jemarch>
hola
<sorear>
o/
<RNeese>
so I got the new starfive visionfive but waiting on a fix for it
<RNeese>
nice layout
<RNeese>
the usb hub chip did not get its rom flashed to it
<RNeese>
so waiting on a replacement chip
<RNeese>
then I hope to have a debian img on it soon
<muurkha>
cool!
<RNeese>
hard part will be the bl/firmware file
<RNeese>
once I have that working it will be ok
<RNeese>
as debian does not have grub/grub2 yet in pkgs so it has to be made to boot like the beaglev without grub
<RNeese>
they have the fedora running on it but its slow and bloated on boot
cousteau has quit [Quit: ♫ I can't forget the day I shot that network down ♫]
voltron has quit [Remote host closed the connection]
<Esmil>
..and then my /efi/boot/bootriscv64.efi is systemd-boot rather than grub since it just creates a menu to choose which (efi-enabled) kernel to load
<Esmil>
..and doesn't implement filesystems etc.
<Esmil>
and I can just press Q to exit to the grub console
<RNeese>
ok
<Esmil>
*exit to the u-boot console of course
<RNeese>
have to figure how to add it to debian build
<RNeese>
I need to get a img going
<Esmil>
No, on the starlight/visionfive u-boot is in spi flash, so you need to update that
<RNeese>
ok
<RNeese>
currently we dont have a working build on debian yet . but reading
<RNeese>
they changed things from what beaglev was to this and its making me relearn
<Esmil>
what did they change?
<RNeese>
no efi on the beaglev
<RNeese>
efi is a change
<Esmil>
yes, you just need to update to the latest u-boot
<Esmil>
that should work the same on the beaglev and visionfive
<Esmil>
it's just that your visionfive just shipped with a newer u-boot
<RNeese>
there is no visionfive in the build system
<Esmil>
which build system?
<RNeese>
that I find other then what the did for fedora
<RNeese>
buildroot
<Esmil>
no, the visionfive board isn't out yet, so it would be surprising if someone added it to buildroot already
jjido has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
<RNeese>
I have started but I have to get some changes goign
<RNeese>
the hashes need changing
<RNeese>
but ok
<RNeese>
I am relearning what I was not taught in beaglev
<RNeese>
I have a visionfive I am working on now
<RNeese>
I have 1 of the dev boards
<RNeese>
thats why I am trying to get it working with my current working img
<Esmil>
well, the kernel in the visionfive(-5.15.y) branches at github.com/starfive-tech/linux is working. you just need to use the jh7100-starfive-visionfive-v1.dtb device tree
<RNeese>
ok
<RNeese>
then get the blfixed for it
<Esmil>
i don't know what the blfixed is
<RNeese>
the bootl loader
<RNeese>
ok brb door
<RNeese>
but I will get a kernel file
<Esmil>
as I said. the bootloader in their u-boot repo should work perfectly fine. that's the one i forked at added my hack to above
BOKALDO has quit [Quit: Leaving]
mahmutov has joined #riscv
<RNeese>
ok
<RNeese>
police dept going door to door for toys for tots program
<gordonDrogon>
muurkha, actually you can hook up 2 x 6502 to the same bus/memory and get them working - the 6502 only accesses ram,etc. on half the clock leaving the other half free - in the day this was used for video...
jjido has joined #riscv
winterflaw has joined #riscv
clandmeter3 has joined #riscv
rektide has joined #riscv
<muurkha>
oh really? I didn't realize that!
<muurkha>
I thought it asserted stuff on the address bus all the time, just that half the time it didn't care what came back
clandmeter has quit [Quit: Ping timeout (120 seconds)]
chripo has quit [Quit: ""]
rektide_ has quit [Ping timeout: 260 seconds]
clandmeter3 is now known as clandmeter
chripo has joined #riscv
EchelonX has joined #riscv
adjtm has quit [Remote host closed the connection]
adjtm has joined #riscv
RNeese has quit [Read error: Connection reset by peer]
smartin has joined #riscv
jjido has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
riff_IRC has joined #riscv
riff-IRC has quit [Read error: Connection reset by peer]
<gordonDrogon>
I think the early computers that did that used the nmos 6502 which didn't tristate the busses so the apple II, etc. had separate tristate buffers.
<gordonDrogon>
however going more OT, but winding back a bit, I worked for a company who put what was essentially 128 6809's on the same die although they made them 32-bit versions - it was an early GPU type system later turned into a super computer acellerator sort of thing. This was about 20 years back.
<gordonDrogon>
so smallish cpus, fast internal buses and magic software...
<sorear>
the manycore dream never dies, does it
<gordonDrogon>
just the software that then becomes the issue ...
rlittl01 has quit [Read error: Connection reset by peer]
riff_IRC has quit [Read error: Connection reset by peer]
riff_IRC has joined #riscv
EchelonX has quit [Remote host closed the connection]
EchelonX has joined #riscv
<drmpeg>
I'll say that 99 and 44/100 percent of Parallella boards are in a closet gathering dust.
<gordonDrogon>
hm. that's a new one to me. intersting, but ..
<drmpeg>
Just a comment to "the manycore dream never dies, does it"
<sorear>
gordonDrogon: parallela?
<gordonDrogon>
sorear, yes - although a lot of these little board seem to have passed me by in recent years.
<sorear>
(if you did into that at all, be aware that "Epiphany-V" is entirely unrelated to risc-v)
<gordonDrogon>
I worked for a supercomputer company back then - always the same problem: I have this 30 year old FORTRAN program - make it go faster and we'll buy your system ...
mahmutov has quit [Ping timeout: 256 seconds]
<sorear>
decomposing the 30 year old FORTRAN program into pieces which can run on 128 6809s while minimizing NoC traffic is always the challenge
<muurkha>
128 6809s, wow
<muurkha>
sorear: I think the manycore dream has become reality in the form of GPUs
<gordonDrogon>
and GPUs are for the most part very application specific.
<gordonDrogon>
although leveraging them now for certian compute/vector operations is a thing..
<muurkha>
GPUs are ... not very application specific
<gordonDrogon>
it's all a blur now - we did code in the i860 all those years ago which (AIUI) was more or less turned into the MMX, etc. instructions in later Pentiums.
<muurkha>
I mean the whole deep learning revolution was running on GPUs before TPUs were created
<sorear>
I think the real question is how active the programmer is in orchestrating data movement through physical space, as opposed to mere addresses
<muurkha>
similarly lots of cryptocurrency mining has been done on GPUs, to the point that gamers are whining up a storm about "scalping"
<sorear>
which has happened to _some extent_ on GPUs but I don't have a sense of how rich the models are and how often they're used
<muurkha>
apparently the actual majority of GPUs sold are going to coin mining, though it's no longer profitable for the most popular coins due to ASIC mining
<gordonDrogon>
I've more or less left that world behind - I just want to play with a nice cpu and some ram ;-)
<muurkha>
applications using OpenCL include Blender, OpenCV, and Octave, and there's at least one OpenCL implementation of BLAS
<muurkha>
also I've worked with people using GPUs for software-defined radio and image compression
<muurkha>
Hashcat is the standard password cracking tool
<gordonDrogon>
the trick to solving the 30-year old FORTRAN was to see if they used standard librarliesl ike BLAS/ Linpack, etc. then implement those on our hardware then re-link their code and hope for the best...
<muurkha>
heh
<muurkha>
that was before OpenMP?
<gordonDrogon>
late 80's/early 90's ..
<gordonDrogon>
before open almost anything.
<muurkha>
yeah
<muurkha>
well, nowadays a lot of stuff is written in Octave (or MATLAB anyway) or TensorFlow so you can get a GPU speedup
<muurkha>
so, that's why I don't think GPUs are very application-specific. just a pain is all
* gordonDrogon
nods.
<muurkha>
also, R, and Numpy via CuPy
Noisytoot has quit [Remote host closed the connection]