ROCm / ROCm/FastFlowLM

poor TPS

Open
#581 11 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
1.9k
Forks
152
Avg merge
4h 14m
Merged PRs (30d)
11

Description

Environment:

xrt-smi examine
System Configuration
OS Name : Linux
Release : 7.0.11-200.fc44.x86_64
Machine : x86_64
CPU Cores : 24
Memory : 32 GB
Distribution : Fedora Linux 44 (Workstation Edition)
GLIBC : 2.43
Model : SER9
BIOS Vendor : American Megatrends International, LLC.
BIOS Version : STX.3xx.SER9.V218.P8C0M0C15.74.Link.SCPC
Processor : AMD Ryzen AI 9 HX 370 w/ Radeon 890M

XRT
Version : 2.25.0
Branch : HEAD
Hash : c63dac03a1dd0d4782befd127f65362ea12478c4
Hash Date : Thu, 28 May 2026 08:54:04 -0700
amdxdna Version : 2.25.0_20260607, 302b82f9120d6706c4e235c4283eb0fe7452b966
virtio-pci Version : 7.0.11-200.fc44.x86_64
NPU Firmware Version : 1.1.2.64

Device(s) Present

BDF Name Architecture Topology
[0000:66:00.1] NPU Strix aie2p 6x8

flm validate
[Linux] Kernel: 7.0.11-200.fc44.x86_64
[Linux] NPU: /dev/accel/accel0 with 8 columns
[Linux] NPU FW Version: 1.1.2.64
[Linux] amdxdna version: 0.15
[Linux] Memlock Limit: infinity

I'm launching benchmark
flm bench qwen3.5:2b

my results:

Model HW 1k 2k 4k 8k 16k 32k
Qwen3.5-2B NPU (FLM) 8.63 8.38 8.21 7.83 7.20 6.20

your results:

Model HW 1k 2k 4k 8k 16k 32k
Qwen3.5-2B NPU (FLM) 26.8 26.2 25.4 23.7 21.3 17.0

Why so poor TPS (Tokens Per Second) for me.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing flm bench qwen3.5:2b under the reported Fedora, XRT, amdxdna, firmware, and NPU configuration, then compare the measured TPS with the listed results. No source file or test is named; done means identifying the configuration or performance cause and documenting a verified explanation or fix.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, linux
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.