Spectral Compute Secures $6M to Eliminate Vendor Lock-In in HPC

SCALE, Spectral Compute’s flagship framework, lets CUDA functions run on any GPU with out efficiency loss or utility rewrites

Spectral Compute, the corporate behind SCALE, a breakthrough software program framework that allows CUDA (Compute Unified Device Architecture) functions to run seamlessly on any GPU, right now introduced a $6 million seed spherical led by Costanoa, with participation from Crucible and distinguished angel buyers. The funding will speed up Spectral’s go-to-market growth and deepen product growth because it brings true {hardware} flexibility to the high-performance computing (HPC) and AI ecosystems.

The world of AI and HPC has been dominated by Nvidia’s software program stack and CUDA programming mannequin. This dominance has created what many take into account to be premium {hardware} prices and vendor lock-in that make total provide chains susceptible to disruption. Now, new tariffs and export restrictions have solely heightened this publicity, leaving companies throughout analysis, finance, and AI highly-dependent on GPUs with 18-month lead occasions.

Spectral’s flagship resolution SCALE is a software program platform that permits CUDA-based code, the inspiration of most fashionable AI and HPC functions, to run natively on any AI accelerated chip, beginning with AMD GPUs. This means organizations can use the identical codebase throughout a number of {hardware} distributors and permits organizations to lastly make vendor agnostic cluster procurement selections. Spectral’s know-how ensures clients stay future-proofed and may faucet into whichever {hardware} presents the most effective efficiency, effectivity, or availability at any given time.

“Companies shouldn’t have to guess their total compute infrastructure on one chipmaker,” mentioned Michael Søndergaard, co-founder and CEO of Spectral Compute. “The imaginative and prescient of SCALE is that groups can lastly use their current CUDA code on any GPU, whether or not that’s Nvidia, AMD, Intel, or others, with out efficiency loss or expensive rewrites. We’re giving builders and enterprises the liberty they’ve by no means had earlier than.”

Unlike conventional translation or porting instruments, which may take 18–24 months and introduce bugs or degraded efficiency, SCALE permits near-instant interoperability. Early clients in microprocessor design and high-performance motorsports are already operating workloads utilizing SCALE to diversify their {hardware} stacks.

“Spectral has solved certainly one of computing’s hardest challenges: letting groups write CUDA as soon as however run it anyplace,’” mentioned Tony Liu, Partner at Costanoa. “Their breakthrough know-how eliminates one of many largest bottlenecks in AI infrastructure and HPC right now, unlocking selection and innovation throughout the compute panorama.”

Founded in 2018 by engineers Michael Søndergaard, Chris Kitching, Nicholas Tomlinson, and Francois Souchay, Spectral brings over 60 years of mixed expertise in HPC, GPU programming, and compiler design. The group’s roots in efficiency optimization span industries from computational fluid dynamics and high-frequency buying and selling to AI and digital broadcasting, all unified by a mission to make high-performance computing open, transportable, and environment friendly.

Today, SCALE lets CUDA work seamlessly throughout your entire AMD chip ecosystem with extra chip structure help to comply with quickly. Spectral has 19 full-time workers and continues to develop because it expands into AI, scientific analysis, and cloud computing sectors. The firm has plans to develop its engineering group to meet demand.

To study extra about Spectral and SCALE, please go to: https://spectralcompute.com/

The publish Spectral Compute Secures $6M to Eliminate Vendor Lock-In in HPC first appeared on AI-Tech Park.