Webinar - Fine Tuning Llama3 on AMD Radeon GPUs

[@OriginalPhillipCarroll]

I literally have 2 x ASRock 7900 XTX Creator blower style cards coming in today for pytorch ROCm things on WSL2. Good timing.

[@SirMo]

I've been using my rx7900xtx for inference, but I really wanted to give fine tuning a shot. This is awesome. Thank you guys!

[@jasper123t]

AMD YES!

[@HeavenlyPalace-Y2K]

AMD gonna beat RTX 5090 or 5080 in next gen GPU?

[@Hajdew]

when ROCm on 6700xt on Windows?

[@OG_Gamers_Live]

now if you'd fix al lthe bugs in AMD Adrenalin

[@mzamroni]

Bad microphone. They should simply use wired microphone. Why amd people keeps ignoring their audio quality

[@mohammadbagherbidram-d8h]

3tg

[@mohammadbagherbidram-d8h]

I mean 3rth 😅

[@SirMo]

@@mohammadbagherbidram-d8h you mean 3rd?

[@mytech6779]

Oh this is ROCm… sorry burned that bridge with 8 years of zero support commitments and fragile dumpster fire implementations. Which upcoming models will receive support? What is the minimum end-of-support date for the current supported hardware list? I'm not about to gamble thousands on GPUs rolling the dice that ROCm support isn't deprecated next month. Why does AMD seem to have severe OCD about hosting older versions of its software stack? The moment a new version comes out the old stuff gets pulled from the website, even if the new version has breaking changes and doesn't support old hardware. Maybe AMD should just focus on a simple universal SyCL-compatible interface for all future GPUs and APUs instead of spinning your wheels on this desperate ROCm nonsense? nVidia CUDA has at least passive support (hosting old versions of software, drivers, and manuals) for 12+ year old mid-range GPUs, so any student or hobbiest can get started with GPU compute acceleration. And ROCm is not even built for Debian, which is on about 60% of all servers and acts the upstream for numerous other Linux distributions. I'll check back after the switch to UDNA. But I'm not holding my breath considering that AMD's terrible undependable GPGPU compute support goes back at least to GCN1 professional workstations cards, years before ROCm.

[@Unclesam404]

Why so pessimistic? Amd just spent so much money on rocm as part of the AI push. It’s a small company vs nvda.

[@mytech6779]

​@@Unclesam404 The _management_ of the AMD GPU compute stack has always been too unreliable for any serious consideration, nVidia and Ati were similar size when this issue started. This is not about technical limitation. This is a matter of perpetual bad management and failure to make or fulfill the most basic support commitments expected of any reputable product. Like a car built with non-standard parts and sold with no warranty from a company with a history of low dependability and no parts availability for out of production models.

[@lolonim]

Amd stock has not raised even 1% this year while NVDA went up far far away. Together we advance? Together we drown😢

[@holthuizenoemoet591]

AMD "advanced money destroyer"

[@LeoBdeF]

💥💥AMD, please release one last AM4 CPU, a 5950 xtX with 5.5GHz (12c/24t) ... And fix the RAID configuration, make it simple, like Intel, I don't want to switch to INTEL CPUs ... and don't fotget to make this also for HDDs.

[@holthuizenoemoet591]

AMD took to long, switched to apple

[@Stanislav7180]

Male new adrenalin software and fix trashnite problem with amd