spectral-compute:~$

The fastest compiler for CUDA code. On any GPU.

Spectral is the creator of SCALE, a toolkit that (re)compiles CUDA code for a broad range of GPUs to increase performance.

Explore SCALE

###### ############ ################## ########################## ################################ ###################################### ############################################ ####################### ####################### ####################### ####################### ######################## ######################## ######################## ######################## ######################## ######################## ####################### ## ####################### ####################### ########## ####################### ####################### ################ ####################### ####################### ###################### ####################### #################### ############################ #################### ################# ############################## ################# ############# ############################### ############# ########### ############################### # ########### ########### ############################### ######## ########### ########### ############################## ############## ########### ########### ############################ ################## ########### ########### ######################### ##################### ########### ########### ###################### ######################## ########### ########### ############# ########################### ########### ########### ########## ############################## ########### ########### ######### ################################## ########### ########### ######### ##################################### ########### ########### ######### ####################################### ########### ########### ######### ##################### ########## ########### ########### ########## ##################### ######## ########### ########### ############## ###################### ####### ########### ########### ######################################## ####### ########### ########### ###################################### ####### ########### ########### ################################## ######## ########### ########### ############################### ########## ########### ########### ############################ ################## ########### ########### ######################### ###################### ########### ########### ##################### ######################### ########### ########### ################## ############################ ########### ########### ############## ############################## ########### ########### ######## ############################## ########### ########### ## ############################## ########### ############# ############################## ############# ################# ############################## ################# #################### ############################ #################### ####################### ###################### ####################### ####################### ################ ####################### ####################### ########## ####################### ####################### ## ####################### ######################## ######################## ######################## ######################## ######################## ######################## ####################### ####################### ####################### ####################### ############################################ ###################################### ################################ ########################## ################## ############ ######

// Performance

Powered by modern LLVM and Clang, SCALE's novel compiler optimizations outperform native vendor toolchains—running 5.94× faster than HIP on AMD GPUs and matching native performance on NVIDIA GPUs. Check out our benchmarks.

// Vendor optionality

One CUDA codebase. Multiple accelerator targets. Switch the compiler backend, recompile, ship — no rewrites, no second codebase, no porting effort. AMD and NVIDIA today; more accelerators coming soon.

// Developer experience

Custom language extensions cut CUDA boilerplate, and SCALE's clangd integration brings expected C++ tooling to CUDA real-time semantic analysis, "Go to Definition," and precise diagnostics—straight into any clangd-aware editor.

spectral-compute:~$

// Why We're Doing This

The accelerator market is bifurcating. AMD's MI-series, custom silicon from hyperscalers, and new entrants every quarter are reshaping what HPC and AI infrastructure looks like — but the software stack hasn't caught up. Teams either maintain parallel codebases for each vendor or pay the vendor lock-in tax in hardware bills, procurement risk, and lost negotiating leverage.

SCALE removes that trade-off. Write CUDA code once, compile it for whatever accelerator is fastest, cheapest, or most available this quarter, and don't compromise on performance to get there. SCALE is an augmentation of what HPC-CUDA developers can already do — not a replacement language, not a JIT transpilation tool, not a porting utility.

The same source. More targets. Better tooling. Superior performance.

spectral-compute:~$

// Book a Demo

See SCALE in action. Our team will walk you through compiling your CUDA codebase for AMD and NVIDIA and show you what real-world performance gains look like for your workloads.

Book a Demo

spectral-compute:~$

We are always looking for exceptional talent to join our global team. If our mission and culture resonate with you, we'd love to talk.

spectral-compute:~$

Registrant Contact:

Name: Spectral Compute Ltd

Address: 483 Green Lanes, London, England, N13 4BS

Phone: +441392581900

Email: [email protected]

spectral-compute:~$

// Our Origin Story

Founded in 2018 by four engineers with over 60 years of combined HPC optimization experience, we saw the same problem across every industry: frustrating vendor lock-in. We grew tired of the status quo and decided to build the solution. That solution is SCALE, our main product and now our sole focus, built to finally bring true portability to high-performance computing.

2016-17

Growing industry frustration over NVIDIA's dominance and vendor lock-in for accelerated computing.

2018

Spectral was born. The development of SCALE began after attempts to improve the usability of HIPIFY hit a dead end.

2018-24

Spectral funded the development of SCALE by offering software performance-optimization consulting.

2025

Spectral becomes a venture-backed startup, raising $6M from VC and Angel Investors to bring SCALE to the entire chip developer ecosystem.

spectral-compute:~$

// Our Core Directives

1. We Love CUDA-And We're Setting It Free

Let's be clear: CUDA is a phenomenal platform, but its vendor lock-in is a critical limitation. We solve this with a compiler that acts as a drop-in nvcc replacement, giving you true hardware choice and enhanced performance, on top of an improved developer experience.

2. Pragmatic, Customer-Driven Innovation

Our innovation is guided by the direct needs of our users and customers. We focus on the painstaking work that matters: providing seamless support for the full CUDA ecosystem, including device and all major math libraries to enable your next breakthrough.

3. Unrivaled Technical Excellence

We estimate no more than 500 people in the world possess the level of expertise required for this work. Our competitive edge is our world-class, global team - true talent knows no borders. We seek out the best engineers on the planet to deliver on our promises.

spectral-compute:~$

// The OS of Our Team

Our culture is designed to solve the hardest problems. It starts with our people.

A Home for World-Class Talent We know our people are our greatest asset. We've created an environment where the best minds in the field can do their best work, supported by a culture of trust, autonomy, and a profound commitment to work-life balance.
Radical Inclusivity We are committed to an environment where every individual is seen, heard, and valued. We believe our collective strength is built on the foundation of individual well-being, and we actively embrace the diverse ways of thinking that are essential for true innovation.
Empowered Autonomy We trust our team implicitly. We give our people the freedom and flexibility to solve new and challenging problems in the way that works best for them.
Humble Expertise Our culture is one of "small ego, big impact." While the team is made up of brilliant individuals, we operate with a humility that prioritizes the best solution, regardless of where the idea comes from.
Collaborative Spirit We know our best work happens when we work together. Our environment encourages asking questions, because we believe it's far more efficient to learn from a colleague than to get stuck. We celebrate each other's successes everywhere.

spectral-compute:~$

// Founders

Michael Sondergaard, Founder and CEO

His journey from university bartender to CEO taught him how to handle complex orders-a skill he now applies to his clients' GPU needs. With over twelve years in software engineering, he's a seasoned pro, but at home, he's just "Dad." A creature of habit-his restaurant order is as consistent as his code, he finds his creative spark decompressing alone at the orchestral stage of music festivals.

Chris Kitching, Founder and CTO

Typically found either breaking CI pipelines, or buried under stuffed elephants. The CTO spends much of his life deleting code, doing questionable things to the compiler, or wondering why he wrote his bio in the third person.

Nicholas Tomlinson, Founder and Software Engineer

Nicholas has been curious and interested in computers for as long as he can remember. He fiddled with his dad's computer so much that it was replaced and given to him at age 6, on the condition that his dad would not fix it if he broke it. This led him to become adept at both breaking and fixing it, and to develop a strong interest in programming from a young age. He still likes to understand computers (and everything else) from the electrons, quarks, and assorted bosons that make up its transistors, to the design of its processors, the inner workings of the compiler, the mathematics and design of AI algorithms, and everything in between.

Francois Souchay, Founder and Software Engineer

As an ex-roadie and truck driver, Francois went to engineering school to polish his technical chops. His engineering career has spanned pro audio, graphics, and VoIP, where he developed acclaimed audio products and pioneering voice conference technology with features later adopted by industry leaders like Dolby. He mostly retired afterward to chase varied interests (aided by stock options), but the prospect of working with incredibly sharp engineers was too alluring to pass up, hence funding the first seven year chapter of Spectral Compute.

spectral-compute:~$

// Our Investors & Partners

We are proud to be backed by some of the best minds in the industry.

   Costanoa Ventures
   Crucible Capital

   ...and 12 Angel Investors 👼.

spectral-compute:~$

[Back to Home] | [Privacy Policy] | [Cookie Policy]

The fastest compiler for CUDA code. On any GPU.

// Performance

// Vendor optionality

// Developer experience

// Why We're Doing This

// Book a Demo

// Our Origin Story

// Our Core Directives

1. We Love CUDA-And We're Setting It Free

2. Pragmatic, Customer-Driven Innovation

3. Unrivaled Technical Excellence

// The OS of Our Team

// Founders

Michael Sondergaard, Founder and CEO

Chris Kitching, Founder and CTO

Nicholas Tomlinson, Founder and Software Engineer

Francois Souchay, Founder and Software Engineer

// Our Investors & Partners

// We respect your privacy.