AMD strengthens its AI stack with Taalas acquisition
AMD plans to integrate Taalas tech into the company’s accelerator roadmap
AMD has officially closed its acquisition ot Taalas, a Toronto-based AI startup that builds specialised AI silicon. With this acquisition, AMD plans to utilise Taalas’ technology to strengthen its AI accelerator roadmap and create new system-level AI solutions.
Taalas has created specialised AI silicon that hardwires neural network weights and architectures directly into its chip’s circuitry. Instead of loading data from memory, Taalas hardwires it into its chips, enabling higher AI performance than CPUs or GPUs. The trade-off is that these chips are model-specific. Talaas claims on its website that its HC1 chip can run a Llama 3.1 8B model 48x faster than an Nvidia B200 accelerator. That’s 16,960 Tokens/s vs 353 Tokens/s.
AMD sees Taalas’ hardware as something that can complement its Instinct GPUs. AMD’s Instinct GPUs are more flexible than Talaas’ model-specific designs, but Talaas’ products offer extreme levels of speed and efficiency for users running specific models. This acquisition gives AMD access to a new class of AI hardware, and reinforces AMD’s engineering presence in Canada.
AMD today announced the completion of its acquisition of Taalas, a pioneer in specialized AI inference silicon, further strengthening the company’s full-stack AI portfolio and long-term inference roadmap.
As demand for AI inference accelerates, Taalas brings differentiated technology and world-class engineering expertise designed to deliver breakthrough performance and efficiency for increasingly specialized AI workloads.
“AI inference is evolving rapidly, and customers need a broad range of compute solutions optimized for their workloads,” said Vamsi Boppana, senior vice president of the Artificial Intelligence Group at AMD. “With Taalas now part of AMD, we’re expanding our ability to deliver differentiated inference performance and efficiency, while strengthening our full-stack AI platform.”
AMD plans to integrate Taalas’ technology into its accelerator roadmap and develop system-level solutions alongside AMD Instinct™ GPUs, complementing its broader portfolio of AMD EPYC™ CPUs, AMD ROCm™ software and the AMD Helios™ rackscale solution.
The acquisition also expands AMD’s engineering presence in Canada and reinforces its commitment to advancing AI innovation and growing world-class technical talent.
– AMD Press Release
While Talaas hardware is interesting, it remains to be seen how well it will perform long-term. Yes, it delivers world-class performance, but its model-specific silicon has downsides. What happens when AI models need updates? Once a model’s weights are etched, a Taalas chip’s model is set. The hardware cannot be repurposed if a user wants to change to a different model. That said, if a client wants to run a single model long-term, Talaas chips are a compelling option.
Join the discussion on AMD completing its acquisition of Taalas on the OC3D Forums.
