Emad Barsoum
@EmadBarsoumPi
Followers
738
Following
925
Media
14
Statuses
821
Corporate Vice President, AI at AMD.
Joined October 2014
A lightweight high quality diffusion model with only 304M parameters and fully open!!! @AIatAMD
https://t.co/KicZIxabdO
rocm.blogs.amd.com
Nitro-E is an extremely lightweight diffusion transformer model for high-quality image generation with only 304M paramters.
0
10
56
One week. One community. All in on open AI. From AMD AI DevDay to PyTorch Conference, here's a look back at all the things we did during #OpenSourceAIWeek: https://t.co/Pkc7xc52k4
1
3
43
Strix Halo is speed. Strix Halo is the best "local supercomputer" - And it could be a laptop that can even build ROCm, or to play Battlefield 6 like a desktop.
22s (Halo) vs 56s (Spark) #Strix Halo beats #DGX Spark on Llama 3.3 70B: 4.9 vs 4.67 t/s. x86 + ROCm > ARM + CUDA https://t.co/lTuZIUoOVK
8
13
152
Will be fully supported by AMD ROCm Day-0
Announcing torchcomms – a new PyTorch API for distributed programming designed to support scalability, fault tolerance, heterogeneity, and extensibility. Torchcomms makes it easier for developers to leverage innovations in existing and new collective communications backends like
0
10
140
Overheard at AMD AI Dev Day @AnushElangovan takes on a quick round of “This or That." Supervised vs. unsupervised? Heap sort vs. merge sort? You might be surprised at his picks.
2
18
199
ROCm everywhere!!!
Building ROCm for the future.
2
2
40
Proud of the team and the great work they are doing in PyTorch!!! From contribution, test coverage, performance improvement and many others. @AIatAMD
https://t.co/0srlK7APOw
rocm.blogs.amd.com
At AMD, the PyTorch ecosystem team is committed to delivering an exceptional out-of-the-box experience for developers. Over the past year, the team has made significant progress in expanding PyTorch...
0
0
20
AI Powered by AMD [Hardware + Software + Solutions]
AMD is focusing on developing comprehensive compute platforms to support AI advancements. This includes enhancing both hardware and software capabilities, with a strong emphasis on creating a pervasive software layer that complements its https://t.co/0sSmfqlAby
1
8
96
Variant Wheels to install just what you need (no need to pull in cuda deps when not needed during a pip install torch) https://t.co/U7uGNlDKVt
amd.com
Learn how PyTorch 2.9 wheel variant support simplifies PyTorch installation for ROCm software.
2
2
43
More to come!!! @AIatAMD
We’re already seeing a 6–10% performance improvement on vLLM MI355X FP8 for medium-dense models since the launch of SemiAnalysis InferenceMAX™ last week. These optimizations were contributed by an AMD principal engineer, building upon the existing InferenceMAX MI355X
2
5
92
In PyTorch 2.9, Python wheel variant now support ROCm!!! @AIatAMD
https://t.co/SV7j2zicEK
amd.com
Learn how PyTorch 2.9 wheel variant support simplifies PyTorch installation for ROCm software.
0
0
6
A thing of beauty ☺️ - please meet Helios at #OCPSummit25 our new open, rack-scale AI platform built on the Open Rack Wide standard introduced by @Meta - MI450 Series Instinct GPUs, EPYC CPUs, and Pensando networking - super excited to bring Helios to market.
56
158
1K
Checkout the APRIL paper on RL with partial rollout by @thu_yushengsu @AIatAMD @lmsysorg
@EmadBarsoumPi Already Integrated to SLIME for both AMD & NVDA GPUs. https://t.co/k4cXT1ZrTQ
Seems like one of the key infra updates that frontier labs do for RL, helps mitigate the long tail problem of gpus working on just 1 completion
0
1
1
Imagine deploying 8,192 MI325x GPUs just months before the MI355x launched, and thinking that was a genius move. That’s a couple hundred million dollars’ worth of compute, and there’s almost no way investors are ever seeing that money back now that these bad boys have dropped...
6
11
131
OpenAI & AMD are partnering to deploy 6GW of AMD GPUs. This enables OpenAI to scale next-gen AI infrastructure powered by AMD Instinct GPUs, beginning with a 1GW deployment in the second half of 2026. https://t.co/0A5hqs0c4z
openai.com
AMD and OpenAI have announced a multi-year partnership to deploy 6 gigawatts of AMD Instinct GPUs, beginning with 1 gigawatt in 2026, to power OpenAI’s next-generation AI infrastructure and acceler...
81
207
2K
Thanks to AMD's Memory capacity and BW advantage
Batch size 64 x 8 num_generations, 3B model, and still ~90GB VRAM free. AMD is awesome!
4
10
229
We are proud to support DeepSeek-V3.2-Exp, the latest release from DeepSeek AI, optimized to run on our flagship AMD Instinct MI355X GPUs. Follow the simple steps in this article to launch the model on your AMD system: https://t.co/68MGpMgyU4
3
9
93