Writes and tunes the low-level GPU code, called kernels, that makes AI training, inference and scientific computing run fast. Levels.fyi (NVIDIA company pages) puts typical pay at $310,000 a year (2026). BLS does not publish separate figures for software developers; it counts them within software developer (SOC 15-1252), whose median pay is $135,980 a year (BLS OEWS, May 2025). Typical entry-level education: Bachelor's degree.
- SOC 15-1252
- Computer and Mathematical
- Last updated October 9, 2026
At a glance
GPU Kernel Engineer in four numbers
- Median pay
- $135,980a year, or $65.38/hrBLS OEWS, May 2025Lowest 10% $82KTop 10% $215K
US median, all jobs: $50,980
- Projected growth
- +10.2%Much faster than averageBLS EP, 2025–2035This career+10.2%All jobs+3.5%
Projected employment change, 2025–2035 Item Value This career +10.2% All jobs +3.5% - Openings a year
- 95,3001,717,800 jobs todayBLS EP, 2025–2035
- Jobs in 20251.7M
- Projected 20351.9M
Employment, 2025–2035 Item Value Jobs in 2025 1.7M Projected 2035 1.9M - Typical education
- Bachelor'sneeded to enterBLS EP, 2025–2035; mix: O*NET
- High school3.1%
- Certificate1.7%
- Associate's5.1%
- Bachelor's84.8%
- Post-bacc certificate0.8%
- Master's4.5%
Highest education held by software developers Segment Value High school 3.1% Certificate 1.7% Associate's 5.1% Bachelor's 84.8% Post-bacc certificate 0.8% Master's 4.5%
Also advertised as: CUDA Kernel Engineer, GPU Performance Engineer, CUDA Engineer, Kernel Optimization Engineer, Deep Learning Kernel Engineer.
Pay
How much do Software Developers make?
Industry-reported pay
$310,000
Median total compensation (base and stock), all US software engineers at NVIDIA, the largest employer of CUDA kernel engineers; high is the IC5 level average. Self-reported.
- Top earners
- $470,306
Source: Levels.fyi (NVIDIA company pages), 2026 (estimate from several industry sources). This is not a BLS figure: BLS counts software developers within software developer, whose pay is shown below. Levels.fyi: NVIDIA Software Engineer Salaries · Levels.fyi: NVIDIA Software Engineer IC5
Every gpu kernel engineer in the US, by percentile of annual wage
Median
$135,980
Middle half earn $105,210 to $171,980
- 10th
- $82,460
- 25th
- $105,210
- 75th
- $171,980
- 90th
- $214,670
All US occupations: $50,980
| Percentile | Value |
|---|---|
| 10th | $82,460 |
| 25th | $105,210 |
| Median | $135,980 |
| 75th | $171,980 |
| 90th | $214,670 |
| All US occupations | $50,980 |
Source: BLS OEWS, May 2025
- Median annual wage
- $135,980
- Median hourly wage
- $65.38/hr
- Mean annual wage
- $148,100
- People employed
- 1,687,890
Versus all US occupations
At $135,980, Software Developers earn 167% above the $50,980 median across all US occupations, a difference of $85,000 a year.
Versus computer and mathematical occupations
Inside the Computer and Mathematical group, where the middle occupation pays $116,580, this is 17% above the group median.
Pay range
The top 10% earn $214,670 or more while the bottom 10% earn $82,460, a gap of 2.6×. That is a typical spread for a US occupation.
Rank in the bank
This median pays more than 90% of the 1,741 occupations in the NueCareer career bank.
Highest-paying state
California pays the most at $174,410, and still leads at $157,523 once local prices are taken into account.
Pay by state
Which states pay Software Developers the most?
State medians come from the same OEWS release as the national figures. The cost-of-living column divides each state median by that state's BEA Regional Price Parity, restating pay in national-average dollars so a high-cost state and a low-cost one can be compared directly.
| State | Median | COL-adjusted | Employed |
|---|---|---|---|
| 1California | $174,410 | $157,523 | 284,390 |
| 2Washington | $166,540 | $155,626 | 107,030 |
| 3New York | $166,180 | $153,983 | 113,510 |
| 4Massachusetts | $165,210 | $156,217 | 48,190 |
| 5Oregon | $142,720 | $138,079 | 21,830 |
| 6New Hampshire | $139,720 | $134,133 | 10,840 |
| 7Maryland | $138,680 | $132,128 | 31,350 |
| 8Colorado | $138,390 | $134,291 | 43,320 |
| 9District of Columbia | $136,880 | $124,548 | 6,120 |
| 10Virginia | $136,460 | $134,970 | 88,280 |
Job outlook
Is GPU Kernel Engineer a growing career?
Employment change over 2025–2035, against every US occupation
Much faster than average. +10.2% projected growth vs 3.5% for all occupations.
| Item | Value |
|---|---|
| Software Developers | +10.2% |
| All US occupations | +3.5% |
Source: BLS EP, 2025–2035
Much faster than average
BLS projects employment of Software Developers to change 10.2% over 2025–2035, roughly 2.9× the 3.5% projected for all occupations. That is 174,700 more jobs.
Annual openings
About 95,300 openings are projected each year, 5.5% of the 1,717,800 jobs that existed in 2025. Most come from workers leaving the occupation, not from growth alone.
Getting in
How to become a GPU Kernel Engineer
Most GPU kernel engineers hold a bachelor's or master's degree in computer science, computer engineering or electrical engineering, and many have a doctorate in high-performance computing or computer architecture. Strong C and C++, an understanding of parallel programming and computer architecture, and comfort with linear algebra are essential.
People break in by writing CUDA code in graduate research, working on scientific computing or game graphics, or contributing optimized kernels to open-source projects such as PyTorch, Triton or CUTLASS. NVIDIA's developer documentation and courses are standard starting points. Interviews often include optimizing a kernel and explaining performance results.
- 1
Typical entry-level education
Bachelor's degree
- 2
Work experience in a related occupation
None
- 3
Typical on-the-job training
None
- 4
Preparation needed (O*NET job zone)
Medium preparation
BLS Employment Projections, 2025–2035; job zone from O*NET 31.0 Database.
What education people in the job have
O*NET incumbent survey: the highest level of education reported by people already working as software developers.
- Bachelor's Degree84.8%
- Associate's Degree or other 2-year degree5.1%
- Master's Degree4.5%
- High School Diploma or the equivalent (such as a GED)3.1%
- Post-Secondary Certificate1.7%
| Item | Value |
|---|---|
| Bachelor's Degree | 84.8% |
| Associate's Degree or other 2-year degree | 5.1% |
| Master's Degree | 4.5% |
| High School Diploma or the equivalent (such as a GED) | 3.1% |
| Post-Secondary Certificate | 1.7% |
Source: O*NET 31.0 Database
The work
What does a GPU Kernel Engineer do?
GPU kernel engineers write the small, heavily optimized programs that run on graphics processors and do the actual math behind AI models: matrix multiplications, attention layers, convolutions and data movement. A model framework like PyTorch hands work to these kernels, so their speed decides how much it costs to train and run a model.
The work is deeply technical. Kernel engineers study a chip's memory hierarchy and instruction set, write code in CUDA, Triton or assembly-like languages, and use profilers to find where every cycle goes. They often co-design with hardware teams, adapt kernels to each new GPU generation, and contribute to libraries that thousands of developers rely on.
Day-to-day tasks
- Write optimized GPU kernels for matrix math, attention and other operations
- Profile kernels to find memory and compute bottlenecks
- Port and retune kernels for new GPU architectures
- Contribute kernels to deep learning frameworks and libraries
- Work with hardware architects on features for future chips
- Benchmark performance against competing implementations
- Write tests that catch numerical errors in optimized code
Typical tasks, researched by NueCareer from the sources below.
Working conditions and hours
GPU kernel engineers work at chip companies, AI labs, cloud providers and AI infrastructure startups, as well as national laboratories. The work is desk-based, often hybrid, and involves long stretches of focused programming and profiling. Deadlines tied to new hardware launches or model releases can be intense.
Where the work happens
Indoors, climate-controlled
NeverEvery dayOutdoors in all weather
NeverEvery dayAlmost never works outdoors.
Physical demands
Sitting
NeverAlmost all the timeSeated for more than half the working day.
Standing
NeverAlmost all the timeBending or twisting
NeverAlmost all the time
Working with people
Contact with other people
NoneConstantFace-to-face discussions
NeverEvery dayDealing with the public
Not importantExtremely importantLittle or no dealing with customers or the public.
Pressure and stakes
Strict deadlines
NeverEvery dayConflict situations
NeverEvery dayConflict situations are rare.
Cost of a mistake
Not seriousExtremely serious
Autonomy
Freedom to decide
NoneA lotA lot of freedom to make decisions without checking first.
Conditions: O*NET Work Context survey. Each item has its own scale, so each dot is read against the two answers that end its own line.
Core skills
O*NET importance, 1 to 5
- Critical Thinking3.9
- Reading Comprehension3.5
- Active Learning3.5
- Active Listening3.4
- Writing3.3
- Speaking3.1
- Monitoring3.0
- Mathematics2.8
| Item | Value |
|---|---|
| Critical Thinking | 3.9 |
| Reading Comprehension | 3.5 |
| Active Learning | 3.5 |
| Active Listening | 3.4 |
| Writing | 3.3 |
| Speaking | 3.1 |
| Monitoring | 3.0 |
| Mathematics | 2.8 |
Source: O*NET 31.0 Database
Transferable skills
O*NET importance, 1 to 5
- Programming4.0
- Judgment and Decision Making3.6
- Systems Analysis3.5
- Complex Problem Solving3.4
- Technology Design3.4
- Systems Evaluation3.4
- Operations Analysis3.3
- Time Management3.1
| Item | Value |
|---|---|
| Programming | 4.0 |
| Judgment and Decision Making | 3.6 |
| Systems Analysis | 3.5 |
| Complex Problem Solving | 3.4 |
| Technology Design | 3.4 |
| Systems Evaluation | 3.4 |
| Operations Analysis | 3.3 |
| Time Management | 3.1 |
Source: O*NET 31.0 Database
Tools and software
- Analytical or scientific software · in demand
- Application server software · in demand
- Business intelligence and data analysis software · in demand
- Cloud-based data access and sharing software · in demand
- Cloud-based management software · in demand
- Configuration management software · in demand
- Content workflow software · in demand
- Customer relationship management CRM software · in demand
- Data base management system software · in demand
- Data base reporting software · in demand
- Data base user interface and query software · in demand
- Data mining software · in demand
“In demand” marks an O*NET Employer-Based Hot Technology, software named in real job postings for this occupation.
Personality fit
What personality type suits a GPU Kernel Engineer?
NueCareer estimate (Holland code IRC) built on O*NET's Software Developer profile, scored 1 to 7
| Axis | GPU Kernel Engineer |
|---|---|
| Realistic | 4 |
| Investigative | 6 |
| Artistic | 2 |
| Social | 2 |
| Enterprising | 2 |
| Conventional | 6 |
Source: O*NET 31.0 Database
What the work is built around
Day to day this is work built around investigating, analysing and solving problems with ideas and data, combined with organising, following procedures and working precisely with data.
- Investigative6.0
- Conventional5.6
- Realistic4.2
- Artistic2.4
- Enterprising1.9
- Social1.8
Closest MBTI types
NueCareer fit scores: cosine similarity between each MBTI type's interest vector and this occupation's O*NET interest profile. Not a BLS or O*NET figure; the method is documented on our methodology page.
Questions
GPU Kernel Engineer: frequently asked questions
How much do Software Developers make?
Levels.fyi (NVIDIA company pages) reports $310,000 (median total compensation (base and stock), all US software engineers at NVIDIA, the largest employer of CUDA kernel engineers; high is the IC5 level average. Self-reported, 2026), with top earners around $470,306. BLS does not publish a separate wage for software developers. It counts them within software developer, whose median pay is $135,980 a year (BLS OEWS, May 2025). The middle half of that occupation earn between $105,210 and $171,980, and the top 10% earn $214,670 or more.
Do you need a degree to become a GPU Kernel Engineer?
Yes. BLS reports a bachelor's degree as the typical entry-level education for Software Developers, and O*NET rates the preparation needed as medium preparation. In O*NET's survey of people already doing the job, 84.8% hold bachelor's degree.
Is GPU Kernel Engineer a growing career?
Yes, strongly. Employment is projected to grow 10.2% over 2025–2035, more than double the 3.5% projected for all occupations (BLS EP, 2025–2035). BLS still projects about 95,300 openings a year, mostly from workers retiring or moving to other work.
What does a GPU Kernel Engineer do?
GPU kernel engineers write the small, heavily optimized programs that run on graphics processors and do the actual math behind AI models: matrix multiplications, attention layers, convolutions and data movement. A typical task: write optimized GPU kernels for matrix math, attention and other operations. Employers most often ask for Analytical or scientific software, Application server software, Business intelligence and data analysis software.
Which states pay Software Developers the most?
The highest state medians for Software Developers are California ($174,410), Washington ($166,540), New York ($166,180) (BLS OEWS, May 2025).
What personality type suits a GPU Kernel Engineer?
O*NET scores Software Developers highest on the Investigative, Conventional and Realistic interest areas, so the work mixes investigating, analysing and solving problems with ideas and data with organising, following procedures and working precisely with data. On NueCareer's MBTI-to-interest mapping the closest types are ISTJ, INTJ. Personality is a fit signal, not a gate. The free NueCareer quiz scores your own profile against all 1,731 careers in the bank.






