- Distinguished Engineer, NVIDIA12
- Member of the Vulkan backend code-owner team (ggml-org/ggml-vulkan), llama.cpp project4
Jeff Bolz is a Distinguished Engineer at NVIDIA; NVIDIA's developer blog describes Bolz as a contributor to Vulkan who designs API abstractions and shading language programming models. The Vulkan registry lists Bolz as the contact for the VK_NV_cooperative_matrix2 extension, which builds on cooperative matrix types to accelerate work beyond simple matrix-multiply kernels. In llama.cpp, the C/C++ runtime for running language models on local hardware, CODEOWNERS lists Bolz in the team that owns the Vulkan backend. Bolz's 2026 commits there optimize the backend's matrix multiplication, FlashAttention, and operator fusion, and add support for more operators and quantized formats.12345
At a glance
- Current roles
- Distinguished Engineer +1 more
- Companies
- NVIDIA, llama.cpp project
- Sector
- Company
- Work listed
- 3
- Sources cited
- 13
- Profile reviewed
Vulkan and shading languages
NVIDIA's technical blog describes Jeff Bolz as a Distinguished Engineer who designs Vulkan API abstractions and shading language programming models for new hardware. The Vulkan registry lists Bolz as the contact for VK_EXT_descriptor_indexing, last modified in 2017, and for VK_KHR_buffer_device_address, last modified in 2019, which lets applications query 64-bit buffer device addresses.167
Bolz is the contact and first-listed contributor for VK_NV_cooperative_matrix (last modified in 2019), which adds cooperative matrix types to SPIR-V, and wrote NVIDIA's April 2019 blog post on machine learning acceleration in Vulkan with cooperative matrices. The cross-vendor VK_KHR_cooperative_matrix extension (last modified in 2023) lists Bolz among its contributors.8910
Bolz is also the contact for two NVIDIA extensions last modified in 2024. VK_NV_cooperative_vector adds cooperative vector types, and VK_NV_cooperative_matrix2 builds on cooperative matrices with type conversions, reductions, per-element operations and tensor addressing.113
Open-source work
The llama.cpp commit history attributes Vulkan backend commits to Bolz. A pull request Bolz opened in November 2024 and that was merged in December 2024 added VK_NV_cooperative_matrix2 support for matrix multiplication and FlashAttention2, so these operations can use tensor cores through NVIDIA drivers.512
CODEOWNERS lists Bolz, with one other maintainer, in the ggml-org/ggml-vulkan team that owns the backend's source directory. Commits from October 2026 extend sparse FlashAttention support to the coopmat2 path, handle mul_mat_id duplicates in the prepass, and optimize pool_2d for large pooling windows.45
Work and contributions
llama.cpp Vulkan backend Open-source code
CODEOWNERS lists Bolz in the ggml-org/ggml-vulkan team that owns the Vulkan backend. Commits authored in 2026 cover matrix multiplication and FlashAttention (including cooperative-matrix paths), topk_moe and rms_norm fusion, new operators such as CONV_3D, and quantized formats including Q2_0, MXFP4 and NVFP4. The repository is MIT-licensed.4513
VK_NV_cooperative_matrix Published research
Contact and first-listed contributor for the Vulkan extension that added cooperative matrix types to SPIR-V, and author of NVIDIA's 2019 blog post introducing it for machine learning.89
VK_NV_cooperative_vector Published research
Contact and listed contributor for the Vulkan extension that adds cooperative vector types to SPIR-V.11
Timeline
- Revision date of the VK_EXT_descriptor_indexing Vulkan extension, which lists Bolz as contact.6
- Revision date of the VK_NV_cooperative_matrix Vulkan extension, which lists Bolz as contact and first contributor.8
- Published "Machine Learning Acceleration in Vulkan with Cooperative Matrices" on NVIDIA's technical blog.9
- Revision date of the VK_KHR_buffer_device_address Vulkan extension, which lists Bolz as contact.7
- Revision date of the cross-vendor VK_KHR_cooperative_matrix extension, which lists Bolz as a contributor.10
- Revision date of the VK_NV_cooperative_vector Vulkan extension, which lists Bolz as contact.11
- Revision date of the VK_NV_cooperative_matrix2 Vulkan extension, which lists Bolz as contact.3
- Pull request adding VK_NV_cooperative_matrix2 support for mul_mat and FlashAttention2 merged into llama.cpp.12
- Extended sparse FlashAttention support in llama.cpp's Vulkan backend to the coopmat2 path.5
Where they work
- NVIDIA 43 people in the wiki
- llama.cpp project 12 people in the wiki
Related people
- Song HanAlso at NVIDIA
- Bill DallyAlso at NVIDIA
- Ian BuckAlso at NVIDIA
- Tianqi ChenAlso at NVIDIA
- Piotr BialeckiAlso at NVIDIA
- Sid ShethChips and hardware · Inference and serving
- Andrew FeldmanChips and hardware · Inference and serving
- Jonathan RossChips and hardware · Inference and serving
- Dylan PatelChips and hardware · Inference and serving
- Sudeep BhojaChips and hardware · Inference and serving
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
13 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- Jeff Bolz — author page, NVIDIA Technical Blog
- jeffbolznv (Jeff Bolz) on GitHub
- VK_NV_cooperative_matrix2 — Vulkan reference page
- ggml-org/llama.cpp CODEOWNERS
- ggml-org/llama.cpp commits authored by jeffbolznv
- VK_EXT_descriptor_indexing — Vulkan reference page
- VK_KHR_buffer_device_address — Vulkan reference page
- VK_NV_cooperative_matrix — Vulkan reference page
- Machine Learning Acceleration in Vulkan with Cooperative Matrices — NVIDIA Technical Blog
- VK_KHR_cooperative_matrix — Vulkan reference page
- VK_NV_cooperative_vector — Vulkan reference page
- vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and FlashAttention2 (llama.cpp PR #10206)
- ggml-org/llama.cpp LICENSE (MIT)