b9788

ggml-org/llama.cppb9788Jun 25, 2026by github-actions[bot]
AI Summary

Adds SYCL tensor parallelism support for dual-GPU setups, implementing a BF16-compressed all-reduce path for large tensors and achieving significant performance improvements on Intel Arc Pro B70.
Key Highlights

SYCL tensor parallelism (--split-mode tensor) for dual-GPU
BF16-compressed all-reduce path for large tensors (nelem >= 32768)
Performance improvements: +20.2% to +78.6% on various models
Implements comm_init/comm_free/comm_allreduce_tensor trio
New Features

SYCL tensor parallelism support
BF16 compression for all-reduce operations
Persistent buffer management for device communication
Full Release Notes

<details open>

sycl : support --split-mode tensor (#24152)

* Sycl tp stage1 (#1)

* SYCL: tensor parallelism (--split-mode tensor) for dual-GPU

Adds the comm_init/comm_free/comm_allreduce_tensor trio that the
meta-backend queries via get_proc_address to enable backend-specific
all-reduce, mirroring the pattern used by ggml-cuda.cu.

For N=2 (the common dual-GPU case) implements a degenerate ring
all-reduce with two size-branched paths:

  * Small (nelem < 32768): FP32 direct memcpy + per-device ADD kernel
    chained via depends_on(memcpy_event). 4 SYCL submissions/call.

  * Large (nelem >= 32768): BF16-compressed. Each device compresses
    FP32 -> BF16 in a local outbox, cross-device memcpys to the peer's
    inbox (HALF the PCIe bytes), then decompresses + adds into the
    local FP32 partial. 6 SYCL submissions/call but PCIe bytes halved
    -- wins for any tensor where PCIe dominates kernel time.

Threshold and BF16 path pattern mirror the CUDA NCCL allreduce.

Storage: ONE persistent uint8_t buffer per device, 4 * nelem bytes
(matches both path layouts: FP32 nelem floats; BF16 outbox+inbox =
2 * nelem uint16_t each). Single alloc+free per device keeps the
SYCL pool's strict-LIFO invariant trivial.

Initial impl handles N=2 FP32 contiguous tensors. Other cases return
false, causing the meta-backend to use its generic butterfly fallback.

Per-call sync is intentionally omitted. SYCL in-order queue semantics
ensure that the meta-backend's next compute on the same per-device
queue waits for our final ADD, and the next allreduce's first op on
the same persistent buffer waits via the same queue. Only comm_free
does an explicit final wait.

OneCCL is NOT used: OneCCL 2021.17 hardcodes single-device-per-process
in communicator_impl.hpp:47 (condition devices.size() == 1), which is
incompatible with llama.cpp's single-process multi-GPU model.

Measured on dual Intel Arc Pro B70 (NEO 26.05.x, oneAPI 2025.3 +
DPC++ nightly):

  Llama-3.3-70B Q4_K_M, -sm tensor -fa 1 -ctk f16 -ctv f16:
    pp512 = 377.08 t/s  (vs 313.65 layer mode = +20.2%)
    tg128 = 17.40 t/s   (vs   9.74 layer mode = +78.6%)

  Qwen3-Coder-Next-80B-A3B Q3_K_M (MoE):
    pp512 = 216.56 t/s  (vs 156.58 meta-backend butterfly = +38.3%)
    tg128 = 17.60 t/s   (vs  14.31 meta-backend butterfly = +23.0%)

  Qwen3-4B Q4_K_M:
    pp64  = 984.51 t/s, tg16 = 49.29 t/s

Llama-3.3-70B in SYCL TP now comfortably beats production layer mode
on both prefill and decode. Coder-Next-80B-A3B (MoE) also wins on
both — the BF16 path is what unlocks the many-medium-allreduces
prefill pattern.

Build/CMake: no changes. No new dependencies. ~210 lines added across
ggml-sycl.h and ggml-sycl.cpp.

* Fix comments

* documentation update to address PR feedback

* Bring over my device-to-device memcpy chagnes

* move the dev2dev_memcpy calls to the upstream 7-parameter variety

* Fix a typo and remove a trailing whitespace

</details>

**macOS/iOS:**
- [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-macos-arm64.tar.gz)
- macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780)
- [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-macos-x64.tar.gz)
- [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-xcframework.zip)

**Linux:**
- [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-x64.tar.gz)
- [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-arm64.tar.gz)
- [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-s390x.tar.gz)
- [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-vulkan-x64.tar.gz)
- [Ubuntu arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-vulkan-arm64.tar.gz)
- [Ubuntu x64 (ROCm 7.2)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-rocm-7.2-x64.tar.gz)
- [Ubuntu x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-openvino-2026.2-x64.tar.gz)
- [Ubuntu x64 (SYCL FP32)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-sycl-fp32-x64.tar.gz)
- [Ubuntu x64 (SYCL FP16)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-ubuntu-sycl-fp16-x64.tar.gz)

**Android:**
- [Android arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-android-arm64.tar.gz)

**Windows:**
- [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-cpu-x64.zip)
- [Windows arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-cpu-arm64.zip)
- [Windows arm64 (OpenCL Adreno)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-opencl-adreno-arm64.zip)
- [Windows x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-cuda-12.4-x64.zip) - [CUDA 12.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b9788/cudart-llama-bin-win-cuda-12.4-x64.zip)
- [Windows x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-cuda-13.3-x64.zip) - [CUDA 13.3 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b9788/cudart-llama-bin-win-cuda-13.3-x64.zip)
- [Windows x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-vulkan-x64.zip)
- [Windows x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-openvino-2026.2-x64.zip)
- [Windows x64 (SYCL)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-sycl-x64.zip)
- [Windows x64 (HIP)](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-bin-win-hip-radeon-x64.zip)

**openEuler:**
- [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23705)
- openEuler x86 (310p)
- openEuler x86 (910b, ACL Graph)
- openEuler aarch64 (310p)
- openEuler aarch64 (910b, ACL Graph)

**UI:**
- [UI](https://github.com/ggml-org/llama.cpp/releases/download/b9788/llama-b9788-ui.tar.gz)