Software Engineer (C++ Systems) (Fremont)
Thunder Compute, Fremont, CA, United States
About the Company
Thunder Compute is the lowest cost, easiest GPU cloud for developers. There are two parts to our company:
We run a cloud platform where developers rent GPUs, just like Coreweave, Lambda, or AWS.
We build virtualization software that uses GPUs 3x more efficiently than these competitors.
About the Role
You'll work on the C++ core of our virtualization stack: performance optimization, systems debugging, and research into oversubscription, checkpointing, and related problems. You will spend your days bouncing between the weeds of complex, performance-critical systems in production, and implementing customer-facing product improvements. This is a far more complicated low-level system than most early-stage startups and the work is hard - but for someone like you this is a feature, not a bug.
Responsibilities
Performance optimization of the C++ virtualization library
Systems-level debugging in production environments
Research into oversubscription, checkpointing, and distributed GPU clusters
Supporting new architectures with deep understanding across the system
Qualifications
Top 0.1% C++ ability
Experience optimizing NIC/C++ performance (not GPU-specific)
Deep understanding of low-level networking, compilers, and similar systems
Experience in production-facing latency-sensitive environments
Required Skills
Extreme ownership: you will be responsible for production systems from day 1
Ability to trace performance issues across the stack — from GPU drivers through kernel networking to user space.
High speed-to-output, we're an early stage startup after all.
Preferred Skills
Bachelor’s degree from top Computer Science program, 3.7+ GPA
Experience with low-level systems in production: e.g., at engineering at a trading firm, systems research lab, or similar.
Pay range and compensation package
Full-time, in-person in San Francisco.
Pay is highly competitive with significant equity. Thunder Compute is profitable on variable costs and growing rapidly.
How our virtualization works
Thunder Compute's core advantage versus other clouds is our proprietary virtualization software.
To understand how this virtualization works, think of GPUs like cars. Today, GPU clouds are like car rental agencies; you rent a car for a few days when you need it, but the car sits parked for much of your vacation while you sleep, go to the beach, etc.
Thunder Compute allocates GPUs more like Uber; each server has access to a GPU when it is needed, and that GPU serves someone else when it isn't.
The end user never has to think about this; they always have a GPU, so from their perspective our servers are identical to other clouds.
What's remarkable is that we can pull this off without actually owning any hardware. To extend this analogy, our cloud offering is an arbitrage where we rent cars from rental agencies (other clouds) and use them as Ubers. The efficiency gain from our software is so large that we have higher margins than the clouds we build on, without the CapEx of building data centers.
Equal Opportunity Statement
We are committed to diversity and inclusivity.
Thunder Compute is the lowest cost, easiest GPU cloud for developers. There are two parts to our company:
We run a cloud platform where developers rent GPUs, just like Coreweave, Lambda, or AWS.
We build virtualization software that uses GPUs 3x more efficiently than these competitors.
About the Role
You'll work on the C++ core of our virtualization stack: performance optimization, systems debugging, and research into oversubscription, checkpointing, and related problems. You will spend your days bouncing between the weeds of complex, performance-critical systems in production, and implementing customer-facing product improvements. This is a far more complicated low-level system than most early-stage startups and the work is hard - but for someone like you this is a feature, not a bug.
Responsibilities
Performance optimization of the C++ virtualization library
Systems-level debugging in production environments
Research into oversubscription, checkpointing, and distributed GPU clusters
Supporting new architectures with deep understanding across the system
Qualifications
Top 0.1% C++ ability
Experience optimizing NIC/C++ performance (not GPU-specific)
Deep understanding of low-level networking, compilers, and similar systems
Experience in production-facing latency-sensitive environments
Required Skills
Extreme ownership: you will be responsible for production systems from day 1
Ability to trace performance issues across the stack — from GPU drivers through kernel networking to user space.
High speed-to-output, we're an early stage startup after all.
Preferred Skills
Bachelor’s degree from top Computer Science program, 3.7+ GPA
Experience with low-level systems in production: e.g., at engineering at a trading firm, systems research lab, or similar.
Pay range and compensation package
Full-time, in-person in San Francisco.
Pay is highly competitive with significant equity. Thunder Compute is profitable on variable costs and growing rapidly.
How our virtualization works
Thunder Compute's core advantage versus other clouds is our proprietary virtualization software.
To understand how this virtualization works, think of GPUs like cars. Today, GPU clouds are like car rental agencies; you rent a car for a few days when you need it, but the car sits parked for much of your vacation while you sleep, go to the beach, etc.
Thunder Compute allocates GPUs more like Uber; each server has access to a GPU when it is needed, and that GPU serves someone else when it isn't.
The end user never has to think about this; they always have a GPU, so from their perspective our servers are identical to other clouds.
What's remarkable is that we can pull this off without actually owning any hardware. To extend this analogy, our cloud offering is an arbitrage where we rent cars from rental agencies (other clouds) and use them as Ubers. The efficiency gain from our software is so large that we have higher margins than the clouds we build on, without the CapEx of building data centers.
Equal Opportunity Statement
We are committed to diversity and inclusivity.