05:08 <azonenberg> o/ CyberDre

05:44 <CyberDre> o/ azonenberg, sorry stepped away for pizza

05:45 <CyberDre> I have had quite a time installing OS's and building glscopeclient to test the OpenCL setup and finally got it working with my little Athlon/Vega APU

05:45 <CyberDre> Most people here are using Debian?

05:50 <azonenberg> That's my dev platform. I think several arch users here too

05:51 <azonenberg> I have two systems I test on: main workstation is dual socket skylake xeon (8 core/16 thread per socket) + RTX 2080 Ti

05:51 <azonenberg> then the lab box is a quad core haswell i5 (no hyperthreading) + Quadro K5200

05:53 <azonenberg> Which i use to make sure performance is still decent on older system

05:57 <azonenberg> And both of them run debian

06:06 <CyberDre> The Arch wiki on GPGPU had some good info on OpenCL that helped me understand the dependencies better on my systems. I tried Ubuntu 16.04 LTS, 18.04 LTS, and 20.04 LTS and have some updates I'm writing into the getting started to help with some of it. Catch2 install isn't quite straightword (have to switch to v2.x branch)

06:07 <CyberDre> 16.04 LTS is pretty unusable unless someone figures out how to get liblxi-dev and libglm for that distro, because they aren't in packages or correct version for that OS

06:08 <CyberDre> 18.X and 20.X worked well enough, the AMD drivers are problamatic and I had to remove a package after installing it, but seems to work now :) Also so libclfft added to the docs, which triggers the OpenCL flag in the cmake step, so you have to have OpenCL headers installed if you include that package

06:08 <CyberDre> I tried installing Debian and got really stuck/lost on Grub bootloader/network config but I might try sideloading it into a different drive for fun later

06:09 <azonenberg> Ok so... I didnt realize 16.04 was still supported

06:09 <azonenberg> I thought 18.04 was the current LTS

06:10 <azonenberg> Generally speaking I target the current ubuntu LTS and stable debian release, not oldstable

06:10 <azonenberg> clfft depends on opencl but if you don't have opencl it should not try using clfft

06:11 <CyberDre> 16.04 will tell you to upgrade to 18.04 when you run Software Updater first time, and 18.04 will tell you to upgrade to 20.04, which I think is the official LTS currently so no problems there. I was mostly trying it for fun/memories, it was my first Linux install when I first built a PC

06:11 <azonenberg> ah, ok

06:12 <azonenberg> Ok so technically 16.04.7 is not end-of-support until april 2021

06:12 <azonenberg> and extended support is still available until april 2024

06:12 <azonenberg> But i don't think it's unreasonable to say we don't support platforms quite that old

06:13 <azonenberg> I hope to support 18.x for a long time though. There were a lot of libraries and tools that came out around that time that would be difficult to do without or backport

06:13 <azonenberg> but moving forward i don't see needing that many more recent stuff

06:27 <d1b2> <you_snus_you_lose> that eye is really bad if it's only like 1Gb/s

06:27 <d1b2> <you_snus_you_lose> you probably really have a 5dB return loss like suggested in that pci-e paper someone linked

06:31 <d1b2> <you_snus_you_lose> you'll be able to find the fault with TDR

06:32 <_whitenotifier> [scopehal-docs] CyberpunkDre opened pull request #24: Updates Ubuntu and AMD OpenCL driver installation steps - https://git.io/JLQuY

06:33 <CyberDre> Cool, I added note about not supporting 16.04 and split Ubuntu section from Debian in getting started, added pics

06:38 <CyberDre> \o hasta luego

14:24 <azonenberg> you_snus_you_lose: yes the eye is awful. It's a cheap SAS cable from amazon, i suspect that's the primary issue

14:24 <azonenberg> The LA pod boards already exist and i'm going to do the best job i can on the host

14:24 <azonenberg> realistically, although the eye is awful it is within spec for the receiver

14:24 <azonenberg> and the RX is a logic analyzer sampling at 2.5 or (maybe) 5 Gsps

14:25 <azonenberg> Basically what it comes down to is, i've spent enough on hardware at this point i'm committed to using SFF-8087 at least for the first two units

14:25 <azonenberg> and it's gonna work as fast as it works

14:25 <azonenberg> and we're gonna have to live with that