freakazoid343 has quit [Read error: Connection reset by peer]
freakazoid343 has joined #riscv
vagrantc has quit [Quit: leaving]
freakazoid12345 has joined #riscv
freakazoid343 has quit [Ping timeout: 252 seconds]
brettgilio has joined #riscv
jacklsw has quit [Read error: Connection reset by peer]
mrus has quit [Quit: Surfing the great wave off Kanagawa]
riff-IRC has quit [Remote host closed the connection]
mrus has joined #riscv
riff-IRC has joined #riscv
Jmabsd has joined #riscv
mrus has quit [Quit: Surfing the great wave off Kanagawa]
mrus has joined #riscv
mrus has quit [Client Quit]
mrus has joined #riscv
mrus has quit [Client Quit]
loki_val is now known as crabbedhaloablut
devcpu has quit [Ping timeout: 245 seconds]
Jmabsd has quit [Changing host]
Jmabsd has joined #riscv
<Jmabsd>
Hi all! Just curious, do you believe there will be any faster RISCV with PCIe, than the SiFive Unmatched, anytime soon - when do you expect the next one really? Smaller litography and support for out of order execution would accomplish this
<Jmabsd>
(_whitelogger,abelvesa,ats,carlosedp,crest,dh`,geist,jcm,rjek,sorear,zaps_: your great insight would be much appreciated;apologies for ping)
BOKALDO has joined #riscv
<Jmabsd>
palmer1: hi
<dh`>
why me? I do software
<Jmabsd>
apologies for the pinging
<Jmabsd>
i saw an announcement that SiFive will make some faster CPU next year using PCIe logics from Intel erh
<Jmabsd>
the SiFive is a great historical achievement as it's a fully fledged computer. however it has no Out of Order Execution
<pierce>
I thought this chat room was compromised for a minute
<Jmabsd>
=o ?
<Jmabsd>
..and the SiFive board is is it.. 28nm, 22nm
<pierce>
Google is your friend, my friend
<Jmabsd>
and 256GB of virtual memory. so it's a stellar baseline computing environment however it is not yet competing in performance, neither with laptops nor with Pi 4 (BCM2711) / ARM72
<Jmabsd>
SiFive did announce that they manufactured their first 5nm CPU this May, however that was internal
peepsalot has quit [Read error: Connection reset by peer]
peepsalot has joined #riscv
<Jmabsd2>
pierce: apart from that Intel Horse Creek, are you aware of any more upcoming RISCV64 with PCI?
<Jmabsd2>
none of the RISCV SBC:s in the pipe, have PCI. they're really low-end. so SiFive Unmatched appears to be the highest end RISCV until 2022-2023 and then that Intel thing would come
hendursa1 has joined #riscv
hendursaga has quit [Ping timeout: 276 seconds]
<Jmabsd2>
ok i googled around a bit. the SiFive Unmatched is still a bit slower than a Raspberry Pi 4. There are no interesting news circulating in this moment for upcoming 2022 releases.
<Jmabsd2>
The V8 Javascript engine was ported to RISCV just recently, great
<Jmabsd2>
The V8 port is "for any RISC-V ISA that supports RV64I, M, A, F, D and Zifencei extensions". Just curious, does the U74 (unmatched) have them?
<Jmabsd2>
pierce,*: ? ^
mahmutov has joined #riscv
drmpeg has joined #riscv
Jmabsd2 has quit [Quit: Leaving]
jjido has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
pecastro has joined #riscv
BOKALDO has quit [Quit: Leaving]
Jmabsd has joined #riscv
Jmabsd has joined #riscv
Jmabsd has quit [Changing host]
<AhmedCharles[m]>
Yes, it's rv64imafdc+Zifencei, with perhaps a few others.
<AhmedCharles[m]>
Though, it's weird, I'm not sure how you're supposed to list the Z instructions.
<AhmedCharles[m]>
Ah, so the U74 cores are rv64gc and the spec says: We have also defined an abbreviation “G” to represent the “IMAFDZicsr Zifencei” base and extensions,
<AhmedCharles[m]>
as this is intended to represent our standard general-purpose ISA.
jjido has joined #riscv
jjido has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
<Jmabsd>
AhmedCharles[m]: Thanks a lot for the insights!! Curious where is there a spec document that says that SiFive U74 supports those said extensions???
<Jmabsd>
Also just curious, this means that Chrome with that RISCV-supporting V8 Javascript engine will just simply build and work out of the box right? If so, neat!
<Jmabsd>
ah, i looked at the fu74 page. obv look at the SBC's page. good. thanks.
<Jmabsd>
jrtc27: so these are just totally conventional extensions, this means running that Chrome with that V8 Javascript engine should just work out of the box right - this is totally uncontroversia
<Jmabsd>
l
<Jmabsd>
most mundane typical class of CPU ops
<jrtc27>
the U74 page only has the abbreviated form (because that's really intended for vendors who know about RISC-V so they really should know what G means) but the manual linked on that page does list the extensions in full
<jrtc27>
yes, not having any of those extensions on a linux system would be noteworthy and a bad idea
<jrtc27>
floating-point is about the only one you could conceivably leave out for some use cases
hendursa1 has quit [Quit: hendursa1]
hendursaga has joined #riscv
<Jmabsd>
jrtc27,AhmedCharles[m]: one more question - the U74 in SiFive Unmatched, does it have any meaningful virtualization support?
<jrtc27>
no
<Jmabsd>
i talked to someone who should be an expert and he said the hardware does *not* support virtualization
<Jmabsd>
mhm ic.
<jrtc27>
that's the H extension, which is not yet ratified
<Jmabsd>
so because it wasn't ratified, SiFive didn't make any silicon for it yet - i see.
<Jmabsd>
There is a Linux KVM RISCV fork. does it work on any real world hardware? ha ha
<jrtc27>
the Rocket design has H support
<jrtc27>
people have used that on FPGA for testing
<jrtc27>
(as well as QEMU)
<Jmabsd>
cool! aha but, it's not in the silicon on the SiFive Unmatched U74. I see. thanks for confirming!
<Jmabsd>
jrtc27: last question, are you aware of any other high/higher-performance RISCV64 chip on the way currently? (Apart from that Intel Horse)
<jrtc27>
the SiFive cores, which Intel said they want to use in their Horse Creek, are all I know of
<jrtc27>
in the general-purpose space
<jrtc27>
Esperanto have their $many core AI/ML HPC thing
<jrtc27>
and presumably Alibaba/T-HEAD are working on their next generation of cores
<Jmabsd>
jrtc27: cool! Esperanto, are they fast enough, could one run a normal OS on it? also same with Alibaba/T-HEAD, may those become available on the general market
<Jmabsd>
i recall reading the news that the guy who was chief architect for Sun/Oracle's SPARC CPU:s, started a RISCV business, but I have not heard anything more from them since.
jacklsw has joined #riscv
<jrtc27>
most of the cores are really just compute units for accelerating AI/ML workloads
<jrtc27>
I highly doubt they have anything other than M-mode or M-mode+U-mode
<jrtc27>
supporting virtual memory would be silly
<jrtc27>
but the SoC does have four general-purpose OS-capable cores on it to manage the thing
<jrtc27>
(1088 accelerator cores and 4 general-purpose cores per SoC)
<sorear>
Jmabsd: pinging a ton of people like that is considered rude in a non-emergency
devcpu has joined #riscv
devcpu has quit [Client Quit]
<sorear>
Jmabsd: regarding your actual question commercial chipmakers are about as secretive as militaries and nobody who actually knows anything will be able to answer
<pierce>
Is this what I sound like? Questions that have no purpose. Questions that are way too premature. Questions that are very Google-able?
<Jmabsd>
pierce: i liked to understand where RISCV is at now. the reference hardware is the Unmatched. i did not know what questions are premature.
<Jmabsd>
google is nice for indication but still
mahmutov has quit [Ping timeout: 245 seconds]
jjido has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
jacklsw has quit [Quit: Back to the real life]
<Jmabsd>
(sorear: sorry)
<Jmabsd>
sorear: your allegory is good.
<Jmabsd>
thanks all for helping me understand this space currently. interesting that Unmatched is indeed.. unmatched! lol
<Jmabsd>
the 16GB RAM is very helpful. that plus incredible IO counterweight its CPU speed which is not very great. i guess the L2 cache of 2MB helps a bit too.
<Jmabsd>
its performance should approximate a 4-core high-end desktop computer of approx year 2008, 2006? :-)
<Jmabsd>
ok here off
Jmabsd has quit [Quit: Leaving]
mrus has joined #riscv
jacklsw has joined #riscv
gordonDrogon has quit [Ping timeout: 240 seconds]
gordonDrogon has joined #riscv
awordnot has quit [Ping timeout: 240 seconds]
<GenTooMan>
sorear, no one wants to be a target as they say.
awordnot has joined #riscv
mahmutov has joined #riscv
Ivii has joined #riscv
jacklsw has quit [Quit: Back to the real life]
jjido has joined #riscv
<dh`>
I'm just curious what prompted them to pick 11 people to ping
<sorear>
they pinged everyone in the OFTC channel
<dh`>
I suppose probably it was just the first line of people in the names list
<dh`>
(on whatever server they were on)
freakazoid343 has joined #riscv
freakazoid12345 has quit [Ping timeout: 256 seconds]
BOKALDO has quit [Quit: Leaving]
adjtm_ has quit [Ping timeout: 240 seconds]
djdelorie has quit [Quit: Leaving]
adjtm has joined #riscv
vagrantc has joined #riscv
jimwilson_ has joined #riscv
<meowray>
jrtc27: is your setup -DBUILD_SHARED_LIBS=on or -DLLVM_LINK_LLVM_DYLIB=on ?
jimwilson has quit [Ping timeout: 250 seconds]
<jrtc27>
-DBUILD_SHARED_LIBS=ON
<jrtc27>
(along with split dwarf and gold)
<jrtc27>
(Ubuntu 18.04's LLD is too old to be useful...)
jjido has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]