Anush Elangovan

1.1K posts

Anush Elangovan

@AnushElangovan

GPU SW @ AMD

Katılım Aralık 2024

212 Takip Edilen12.4K Takipçiler

Anush Elangovan@AnushElangovan·1d

Yes come try ROCm

Yannick Nick@keennay

I wonder if AMD would've been better suited selling PCIe versions of MI300X & MI355X GPUs (similar to H100 / H200 NVLs), & gaming cards comparable to 5090s, then you'd probably have way more of us in that eco Also the collab with George & the team at The Tiny Corp helped a bunch too, as well as HotAisle's AMD compute availability. Gradually more of us will play with ROCm

Eesti

5.3K

Anush Elangovan@AnushElangovan·3d

@terminoid_ wait does the latest upstream release not work ? A lot of our CI fleet is MI300x.

English

Terminoid@terminoid_·3d

@AnushElangovan why are your mi300xs stuck on such an old version of vllm? can't use qwen 3.5 or 3.6, or gemma 4, or fp8, or int8, etc.

English

129

Anush Elangovan@AnushElangovan·5d

Tell us how we can make ROCm better. We are listening and we will grind it down.

Aaryaman "Jam" Vasishta@adyaman

@AnushElangovan @wild_zones @AIatAMD Build commands shared here reddit.com/r/ROCm/comment… much simpler than before.

English

108

8.6K

Anush Elangovan@AnushElangovan·3d

@Elon_fanboy77 Too kind. I'm just the... Agent

English

121

NathanZ@Elon_fanboy77·4d

@AnushElangovan and Anush is the king!

English

151

Anush Elangovan@AnushElangovan·4d

Open is a Philosophy !!! Speed is the moat

AI at AMD@AIatAMD

“Open is more than a license to us. It’s a philosophy.” That approach is showing up in the real world, with ROCm powering 8 of the top 10 AI labs. @AnushElangovan at AMD #AIDevDay.

English

3.1K

Anush Elangovan@AnushElangovan·4d

Need more coffee to keep up with the agents

AI at AMD@AIatAMD

What are your agents doing while you’re at AMD #AIDevDay? Here’s what @anushelangovan said.

English

2.5K

Anush Elangovan retweetledi

AI at AMD@AIatAMD·4d

developers developers developers and t-shirts

English

1.9K

Anush Elangovan retweetledi

AI at AMD@AIatAMD·4d

Roll call! Who's here for AMD #AIDevDay?

English

7.6K

Anush Elangovan@AnushElangovan·4d

See I told you @jackhuynh had Strix Halos to give out. Great to meet our developers.

Jack Huynh@jackhuynh

🚀 Inspired to be at @AMD AI Dev Day in San Francisco today where the AI community is converging to shape what’s next. The energy is exceptional. Every keynote, demo, and conversation reinforces a deeper truth: AI has moved beyond advancing computing. It is now redefining human potential and possibility at global scale. At AMD, we’re proud to be at the center of this transformation, accelerating it with open, powerful, and accessible technology. A personal highlight was spending time with visionary founders and CEOs including @ComfyUI, Midjourney, @AnythingLLM, @Kog__AI, Starcluster and many more. These builders are turning bold ideas into real-world impact, and it was an honor to put Ryzen AI Max powered @HP ZBook mobile workstations in their hands so they can move even faster. This is leadership in the AI era: empowering creators, fueling ambition, and turning visionary ideas into scalable reality. The future isn’t coming. It’s being built right here, right now. Excited to keep pushing the boundaries alongside this incredible community.

English

Anush Elangovan@AnushElangovan·4d

@woctordho There is a machine readable ISA spec already published

English

woctordho@woctordho·4d

@AnushElangovan Also I hope your PDF docs can be published in an AI-readable format like this github.com/woct0rdho/rdna… Any format like html, tex, rst would be helpful

English

151

Anush Elangovan@AnushElangovan·4d

@ruoshui217 Understood. Yes 26.04 and later will have it by default

English

Ruoshui217@ruoshui217·4d

@AnushElangovan thank you，I have finished 24.04 installation and finished all setting. 0day support for major system branch is crucial for AMD and ROCm

English

Anush Elangovan@AnushElangovan·4d

@intothemild @codecovenant Right after 7.13 branch

English

here for the end times@intothemild·4d

@AnushElangovan @codecovenant I'm assuming thats landing in ROCm 8.0?

English

Anush Elangovan@AnushElangovan·4d

@bet0x Thanks for the feedback. I assume this is for AIMs that are tested for a longer period to ensure enterprise deployments. If you point me to the specific issues I'll follow up with that team.

English

102

Alberto Ferrer@bet0x·4d

@AnushElangovan On my side, as a note, your images are too primitives or old, even for enterprise, you lack serious support on several fronts, ie: aim-build your own build system uses already outdated software. I optimized myself and I got on your platform 35%+ speed.

English

113

Anush Elangovan@AnushElangovan·4d

@lursmani9 Give the latest The Rock releases a try

English

140

Davit Lursmanashvili@lursmani9·4d

@AnushElangovan Enable support for older GPUs that are arbitrarily excluded from the supported list. For example, RX 6600 is supported, yet RX 6650 XT for some reason is not, even though with a hacky override it works perfectly.

English

199

Anush Elangovan@AnushElangovan·4d

@ruoshui217 The dkms driver is optional the OSS driver is fully functional and should auto load in Ubuntu. The userspace is just a venv whl. Can help on discord or here

English

177

Ruoshui217@ruoshui217·4d

@AnushElangovan Just waste my entire day for ROCm. As A Linux rookie, installed Ubuntu 26.04 for my desktop, found AMD drivers and ROCm not adapted for new kernel. After hours frustrating attempts with Codex to fix DKMS error. Degrade to 24.04. PS. AMD drivers downloading speed is nightmare

English

241

Anush Elangovan@AnushElangovan·4d

@codecovenant stay tuned for HRX.

English

123

Code and Covenant🇨🇦@codecovenant·4d

@AnushElangovan Make it actually faster than vulkan? Vulkan is nearly 50% faster compared to rocm.

English

144

Anush Elangovan@AnushElangovan·4d

@hhhppp09317668 Tell me more please. GitHub issue ?

English

120

hhh ppp@hhhppp09317668·4d

@AnushElangovan fix the memory access issue when using multi-gpu which is plaguing llamacpp

English

123

Anush Elangovan@AnushElangovan·5d

@GaemaAI DM me the address and what you want to do with it

English

277

Gaema AI@GaemaAI·5d

@AnushElangovan Send us a Halo or two. We are already faster on R9700.

English

330

Anush Elangovan@AnushElangovan·5d

I have to rep a bit more to get biceps like that. But I do know there are T-shirts and maybe @jackhuynh will have some Strix Halo laptops for you. See you at Dev Day tomorrow 💪🏾💪🏾💪🏾

Ramine Roane@roaner

Who did this?

English

2.8K

Anush Elangovan@AnushElangovan·5d

@wild_zones @AIatAMD github.com/pypa/distutils… had the fix for Windows linker failing. Let me see if we can push that through.

English

235

wild_zones@wild_zones·6d

@AnushElangovan @AIatAMD This is how people talk about your products: “There are legitimately only a handful of people in the world that have provided evidence online of successfully installing an actually usable Flash Attention 2 directly on Windows 10/11 with RDNA4/GFX1201.” reddit.com/r/ROCm/s/qz4hV…

English

253

wild_zones@wild_zones·6d

I've been shouting from the rooftops that there's a 9000 series AMD GPU with 32GB of VRAM for $1299 sitting on the shelf of every Micro Center in North America. Who do I have to bug to at @AIatAMD to provide real support or let people know it exists??? 32GB wasting away!!!

Lotto@LottoLabs

The first users of localmaxxing.com I owe my life to You are amazing

English

1.3K

Keşfet

@terminoid_ @Elon_fanboy77 @jackhuynh @woctordho @ruoshui217 @intothemild @codecovenant @bet0x