NVIDIA GPU Architecture Engineer Jobs
NVIDIA's GPU architecture team decides what each new generation of graphics and compute silicon can do before a single line of RTL gets written. The work covers shader processors, tensor cores, ray tracing hardware, and the memory systems that feed them, across Hopper, Blackwell, and whatever comes next.
SemiDesignJobs tracks these roles mostly at NVIDIA's Santa Clara headquarters, with more in Austin and at international sites. You can filter by architectural domain, whether compute, graphics, memory, or interconnect, and by experience level.
The job leans on deep computer architecture knowledge and real performance modeling skill. Most architects spend their days inside microarchitectural simulators arguing about cycle counts and bandwidth, not drawing schematics. Knowing CUDA and how GPU compute workloads actually behave matters, because those workloads drive most of the design tradeoffs.
Day to day, GPU architects work on the pieces that separate one generation from the next: tensor core design for AI matrix math, shader execution improvements, cache hierarchy tuning, memory bandwidth, and the interconnect that ties hundreds of streaming multiprocessor blocks into one coherent parallel machine. If you have done SM or cache work before, the tradeoffs will feel familiar.
Blackwell shows how far this has gone. Introduced in 2024, it is a two-die design connected over NVLink-C2C, carries 192GB of HBM3E, and runs a fifth-generation transformer engine with FP4, FP6, and FP8 support. Architects on Blackwell and its successors are building the hardware that most large AI models now run on.
This is NVIDIA-specific work, but the skills carry across the field. If you are also looking at adjacent GPU and accelerator openings, browse NVIDIA ASIC design engineer jobs and general GPU architecture engineer roles. AI accelerator startups and hyperscaler in-house teams compete for the same people, which keeps compensation high.
Total comp for GPU architecture roles is among the highest in the field. Senior and staff architects commonly land in the $250K to $450K range once equity is counted, and principal packages run higher. Exact numbers depend on level, location, and stock timing; the semiconductor salary guide has broader benchmarks across roles.
If NVIDIA GPU work is the target, save a search on SemiDesignJobs and you will get an email when a matching architecture opening posts. Related AI silicon roles show up on the AI chip design engineer jobs page too.
FAQ
What GPU microarchitecture domains do NVIDIA architecture engineers work on?
SM design, tensor core arithmetic, ray tracing units, L2 cache and memory crossbar, NVLink interconnect, and the video encode and decode engines. Each area has its own specialized sub-team, so architects usually go deep on one domain rather than spreading across all of them.
What background is typical for NVIDIA GPU architecture engineers?
Most hold an MS or PhD in computer architecture or computer engineering with a GPU or CPU microarchitecture focus. Published academic work on GPU architecture, or prior time on a commercial GPU or CPU architecture team, makes a candidate competitive.
How does the GPU architecture role differ from RTL design at NVIDIA?
Architects define behavior, performance targets, and block-level interfaces for new features. RTL designers then implement those specs in synthesizable SystemVerilog. Architects spend more time on modeling, analysis, and specification; RTL engineers focus on cycle-accurate implementation.