Axelera Edge and Server

Big day at Axelera today. Two things landed this morning and, while the hardware is the headline, I reckon the software is every bit as interesting, especially to everyone here in the community.AxScriptThe short version is, you’ll be able to compile your own models. Not your weights on one of our models. Your actual model, your operators, your data movement, your scheduling, up and running on Axelera silicon.If you’ve ever hit the end of a supported model list and thought "well, that's that", this is the fix. AxeleraScript, or AxScript, is a Python-based DSL that programs the AI cores directly. Two ways to use it:Write it by hand. There's documentation and an agentic harness that reads your source model and helps produce the AxScript for it, with you keeping control of what comes out. Evaluation and quantization tooling ships alongside, so you can check accuracy at each step rather than at the end. Or don't. AxGraph, the graph compiler, still takes ONNX and PyTorch and handles it autom

Mild disclosure up front: there's a sale on, and this is me telling you about it. 😄I don’t normally get too commercial in here, but the memory ceiling is one of the questions that comes up quite a lot, and the Axelera Edge 130p in its 16GB configuration is a pretty simple answer to that.Two things it actually buys you, as opposed to just "more is better":Bigger models load. The small LLMs are fine on 4GB. Llama 1B and 3B, no drama. Once you go past that it’s not much slow, as just doesn't fit, and there's not really any way round that. Model switching stops being a problem. If you're juggling four or more models and swapping between them, they all need to stay resident. 4GB runs out once a couple of those models are on the larger side. This is the one I'd underline, because the multi-model and multi-stream threads in here are pretty common.I think it’s actually a strong approach, prototyping on a nice, big 16GB card so you're not fighting memory limits while the pipeline is still unde

I am using armbian 6.1.115 after  installing everything lspci -vv0001:11:00.0 Processing accelerators: Axelera AI Metis AIPU (rev 02)    Subsystem: Axelera AI Metis AIPU (rev 02)    Control: I/O- Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-    Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-    Interrupt: pin A routed to IRQ 167    Capabilities: <access denied>    Kernel modules: metis  axdevice give me WARNING: 4PCI device count mismatch: lspci=1, triton=0I have also test it on https://joshua-riek.github.io/ubuntu-rockchip-download/boards/rock-5b-plus.html and get exactely the same message any ideas ?

Environment SoC/Board: Rockchip RK3588, Radxa ROCK 5B Plus    . It has 2 M.2 slots! OS/Kernel: Debian bookworm, 6.1.43-15-rk2312 (aarch64) RAM: 16GB (also tested with mem=4G kernel parameter; total reported ~3.8Gi) Device: Axelera Metis PCIe accelerator, PCI ID 1f9d:1100 (shows as “Axelera AI Metis AIPU”) PCIe topology: Endpoint at 0001:11:00.0 behind upstream bridge 0001:10:00.0 (Gen3 x2) Drivers: axl/metis-dkms 1.2.3; “Kernel driver in use: axl” shown; /dev/metis-1:11:0 device node present when driver bound Container: Voyager SDK tested in Ubuntu 22.04 container (privileged); device access tests moved back to host after hangs observed inside container Device-tree overlay (added to expand ranges window to 48MB)Reason: Default 32-bit non-prefetchable MEM window under fe160000.pcie was ~14MB; Metis BAR2 asked for 32MB (plus headroom), so a 48MB window was used in different place that seems free. Overlay below is applied via extlinux overlays:text/dts-v1/;/plugin/; / { compat

Hi everyone — I'm Lorenco, a Rust/C++ compiler & real-time/kernel systems engineer (based in Italy), getting to know the Metis/Voyager stack.While reading axelera-driver I came across GitHub issue #4 — the use-after-free in axl_aipu_remove() on `modprobe -r metis`. It looks correctly root-caused: the MSI handlers dereference dma_va, but the devm-managed IRQs are only released after remove() returns, so a late MSI hits the freed DMA buffer (synchronize_irq() alone doesn't mask the vector). Quiescing DMA/MSI before the free would also unblock a clean S3 suspend path.Two questions:1. Is the GitHub repo the right place to send driver patches, or do you handle those through another channel?2. More broadly — I do low-level compiler/runtime/driver work and would be glad to contribute. Where's the best spot for someone outside the team to pick things up?Background, in case useful: I recently contributed a canonicalization fix to Tenstorrent's tt-mlir compiler (github.com/tenstorrent/tt-mli

​ I have a 4 AIPU Axelera card, and it is plugged into the Dell PowerEdge R760 server.Currently the card fails to enumerate all the Metis devices, and I see PCIE AER errors in the logs relatd to the Metis devices.Below is the the output of the axdevice command. Attached problem report as well.Have you seen this issue before ? Command line logs:(axelera-env) user@r760-sales-demo:~$ sudo lspci -d 1f9d: -vt-[0000:0c]---01.0-[0d-12]----00.0--+-02.0-[11]----00.0 Axelera AI Metis AIPU (rev 02) \-03.0-[12]----00.0 Axelera AI Metis AIPU (rev 02)(axelera-env) user@r760-sales-demo:~$ cat /proc/cmdlineBOOT_IMAGE=/boot/vmlinuz-6.8.0-111-generic root=UUID=5c7f0039-5e33-4e4b-85e3-9d0215b77b59 ro iommu=pt loglevel=8(axelera-env) user@r760-sales-demo:~$ axdeviceWARNING: 4PCI device count mismatch: lspci=2, triton=0ERROR: min() iterable argument is empty(axelera-env) user@r760-sales-demo:~$ axdevice -vINFO: Found AIPU driver: metis 208896 0WARNING: 4

Hello Axelera team and community,We are running a Metis PCIe card (16GB, rev 02) in a Proxmox VE 9.1.9 cluster with VFIO PCIe passthrough into a KVM guest. We understand this is listed as Beta in SDK 1.6, and we want to share detailed diagnostics to help improve it.Environment:Host: Proxmox VE 9.1.9, kernel 7.0.0-3-pve (Linux 6.14), Intel VT-d enabled (intel_iommu=on iommu=pt) Metis firmware: flver=1.6.0, bcver=7.4 DKMS driver: metis-dkms=1.4.16 SDK: axelera-rt==1.6.0, axelera-runtime==1.6.0 Guest: Ubuntu 24.04.3, kernel 6.8.0-111-generic IOMMU group 22: only the Metis card (isolated, ideal for passthrough)What works (bare metal on Proxmox host):axelera.runtime Python API works perfectly on the host:ResNet18: 1.29ms per inference RetinaFace: 5.32ms per inference, face detected with confidence=1.000 load_model_instance completes instantlyWhat fails (inside KVM guest via VFIO):conn.load_model_instance(model, num_sub_devices=1, aipu_cores=4) fails:  [libaxldev_linux.c:675] DMA_GET_XFER_SY

Card: AXE-BME20P1AJ04A02 (Metis 1-chip PCIe)Host: ZimaBoard 2 (Intel N150), ZimaOS (kernel 6.12.25)SDK: device-1.5.3-1, Voyager SDK v1.5 Setup: Since ZimaOS isn't officially supported, I compiled the metis driver (v1.4.4) from the DKMS package inside a Docker container matching the kernel headers, then loaded via insmod. Everything worked — axdevice showed the card at 16GiB, flver=1.2.0-rc2, bcver=1.0, clock=800MHz, all MVMs at 100%. What happened: Ran interactive_flash_update.sh inside a privileged Docker container with device passthrough. Stage 1 completed successfully — DDR1/DDR2/FWIM written to secondary slots, "Update completed" confirmed.After power cycle (full unplug for 2+ minutes), the card no longer responds:dmesg: Invalid hdrv magic 0 / Fail to get hdrv area Both PCIe BARs read all zeros via /dev/mem axdevice: board_type=unknown (not responding) All triton_multi_ctx commands timeout (flashstatus, flashver, flashactivate, cold-boot 3, fwload) interactive_flash_downgrade.sh: "

is there any command to check vram usage on axelera metis aipu? 

Hi, I’m evaluating Metis PCIe card. It is working. I can run inference and run demo programs. But the fan is running at full speed all the time.I checked several articles in this community site and realized that BC FW must be upgraded to enable smart fan control.I confirmed the BC FW version of my card with axdevice command and it reports bcver==0.0.How can I update or flash the correct BC FW?Thanks.

I'm currently setting up the Axelera Voyager SDK on Ubuntu 22.04.5 LTS with a verified PCIe Metis AIPU card (rev 02 properly detected via lspci, and metis module loaded). However, during installation with install.sh --all --media, several symbolic link warnings occurred (e.g., media: invalid symlink), and although the runtime reports as installed, the file libaxruntime.so is missing entirely from runtime/lib/. This results in the inference script failing with: Failed to create device manager: libaxruntime.so: cannot open shared object file. Additionally, the AxInference.cpp build fails due to missing axruntime/axruntime.hpp. I’ve already tried clearing ~/.cache/axelera/ and reinstalling the runtime, but the issue persists. Would appreciate any guidance to resolve the missing runtime files and complete the setup.Thanks Pravin Kumar R

can’t get the pcie card to show up with lspci on this lenovo thinkcenter 720q, fan powers up but doesnt show up

Hello,I am trying to open ssh server in my Metis SBC board (with 1x AIPU, 4 GB of RAM, 16GB of host RAM and active cooling, Rev1), but right now I’m in the sh $ console with the “antelao” user, and whenever I need to run a sudo command, I’m asked to enter the sudo password, which I don’t know. Can you help me with this or point me to the right tutorial/guide?Thanks a lot

See:  

Hello everyone on the Axelera team!I'm having fun with the new Metis PCIe card that arrived!Thanks!I'm trying out voyager-sdk version 1.5.1 that supports this card, but it seems to throw an OpenCL runtime error and the sample won’t running.It doesn’t seem to be working with GST.For now, using the ‘--disable-opencl’ argument made it work.Alternatively,‘hardware_caps=config.HardwareCaps(…    opencl=config.HardwareEnable.disabel ...)source code.This isn't a solution, right? 

Hello,I am testing some models and I am regularly facing this error message :WARNING : No OpenCL GPU devices foundWARNING : Please check the documentation for installation instructionsDetecting... : 6%|██████▍ | 37/670 [00:01<00:14, 44.36frames/s][ERROR][axeCommandQueueExecuteCommandListsAsync]: Wait kernel failed with return code -1400.terminate called after throwing an instance of 'std::runtime_error' what(): axr_run_model failed with Error at zeCommandQueueExecuteCommandLists(cmdqueue, n_cmdlists, cmdlists, nullptr): cmdqueue_run_cmdlists: 319: Exit with error code: 0x7FFFFFFE : ZE_RESULT_ERROR_UNKNOWNAborted (core dumped)I have used this command :./inference.py yolov8m-coco media/alone.mp4 --no-display --show-stats --show-latencyHere is the link through my testing video : https://pixabay.com/videos/man-adult-road-walking-walk-76621/To fix

Big update from the Metis world: the epic new 4-core PCIe card is now available to order over on the store.Huge performance shift that opens up seriously scalable edge AI.📊 Key specs: 856 TOPS (INT8) Choice of 16GB and 64GB DRAM ~12,800 FPS on ResNet-50 Runs dozens of streams or parallel models per board 30–58W typical power draw It’s not just the raw speed, it’s the ability to compress multi-camera inference, object detection, tracking, segmentation, and more into a single card that I find especially exciting. You can prototype in days, using the Voyager SDK’s YAML-based pipeline builder and model zoo (everything from YOLOv5 to small LLMs).The benchmarks are where it gets interesting. YOLOv5m hits 1539 FPS at 640x640! YOLOv7 manages 859 FPS, YOLOv8l gets 720 FPS. For context, that's enough to process dozens of camera streams simultaneously without breaking a sweat.If you’re testing edge-heavy ideas (multi-stream CV, robotics, smart cities), this is one to keep on your radar.

Radxa Orion O6 board does not boot with Metis PCIe card installed in its PCIe slot.Does not show even the BIOS screen.Tested some other cards (Intel 10GbE NICs) works all okay with the board.However, Metis PCIe card works nice with my PC and Orange Pi 5Plus board.Compatibility problem..?

Hi all,I am trying to compile a high-resolution model (1504×1504) on a Metis PCIe card, but the compilation fails due to a memory constraint:ERROR : RuntimeError: Could not find a tiling factor that fits the memory constraints l1_constraint=4011520 l2_constraint=7894528. After attempt=7 and h_size=1 and adj_factor=1, memory usage still is memory_usage={L1: {190: 141376, 193: 5017600, 191: 143360}} and per-pool memory usage {L1: 5302336}.The error indicates a required L1 size (≈5.3 MiB) exceeds the constraint (≈4.0 MiB).My questions are: Voyager SDK Solution: Is there a Voyager SDK flag or setting to adjust the tiling to overcome this ≈4 MiB L1 limit? Hardware Solution: Are there any Metis card variants (or multi-AIPU cards) that feature a larger per-core L1 cache? Any guidance on compiling this high-resolution model is appreciated. Thank you!

Putting this up for folk who may hit a similar issue.If you have a machine which has a single root PCIe device, where the output of lspci -tv looks something like this;alsutton@svr204:~$ lspci -tv-[0000:00]-+-00.0 Intel Corporation 8th Gen Core Processor Host Bridge/DRAM Registers +-02.0 Intel Corporation CoffeeLake-S GT2 [UHD Graphics 630] +-08.0 Intel Corporation Xeon E3-1200 v5/v6 / E3-1500 v5 / 6th/7th/8th Gen Core Processor Gaussian Mixture Model +-14.0 Intel Corporation Cannon Lake PCH USB 3.1 xHCI Host Controller +-14.2 Intel Corporation Cannon Lake PCH Shared SRAM +-16.0 Intel Corporation Cannon Lake PCH HECI Controller +-1b.0-[01]----00.0 Micron/Crucial Technology P2 [Nick P2] / P3 / P3 Plus NVMe PCIe SSD (DRAM-less) +-1f.0 Intel Corporation B360 Chipset LPC/eSPI Controller +-1f.3 Intel Corporation Cannon Lake PCH cAVS +-1f.4 Intel Corporation Cannon Lake PCH SMBus Controller

I’m putting an idea out to see if anyone has some data to back a theory.I’ve been talking with ​@Spanner about some problems I was having with a card, and, after looking over everything I’ve done, I’m wondering if a 75W slot is *required* rather than recommended, which would mean you need to use a PCIe x16 slot.One issue I had with the card was a small v shaped cut-out in the full height PCIe back-plate, which meant that the card wouldn’t sit fully into any PCIe x16 slot that I had (photo attached so you can see what I mean). This meant I had to use a 8x or 4x slot.Looking through the PCIe spec, only 16x slots are supplied with 75W, meaning that the 8x and 4x slots I tried were limited to 25W (which I confirmed), which I’m thinking could be the cause of the problems I saw.I have a 2nd hand machine turning up next week which I can test this theory on more easily (eBay still is useful for some things), but I’m wondering whether the PCIe card should be sized for a 16x slot in order to ens

Hi, I’m trying to deploy and inference with a yolov8 segmentation model. I tried to follow the tutorial, but I just get:INFO: deploying model yolov8sseg-coco for 4 cores. This may take a while… How long should this run? I left it on for 3.5 hours and nothing happened (except the loading bar). This seems to be a model included in your library: yolov8sseg-coco.Any idea if I can turn on more extensive logging to see what is happening? Thank you in advance!  

I’m also not seeing the card recognised in `lspci`. I’m adding a post to this thread to keep all the info in one place for others, but let me know if it should be a different thread.I’ve enabled `pci-check.service` and get the following when I check the status;alsutton@svr220:~$ sudo systemctl status pcie-check.service● pcie-check.service - Check for PCIe devices with vendor ID 1f9d and reboot once if not found     Loaded: loaded (/etc/systemd/system/pcie-check.service; enabled; vendor preset: enabled)     Active: active (exited) since Wed 2025-08-06 09:33:01 UTC; 11min ago    Process: 1709 ExecStart=/usr/local/bin/check_pcie_device.sh (code=exited, status=0/SUCCESS)   Main PID: 1709 (code=exited, status=0/SUCCESS)        CPU: 6msAug 06 09:33:01 svr220 systemd[1]: Starting Check for PCIe devices with vendor ID 1f9d and reboot once if not found...Aug 06 09:33:01 svr220 check_pcie_device.sh[1709]: Reboot already performed; skipping further checks.Aug 06 09:33:01 svr220 systemd[1]: Finish

Hey everyone, I’m having trouble getting my Axelera Metis PCIe AI Accelerator to be recognized by `lspci`(and my system in general). I tested the card on two different systems. On my AMD setup, I am using an AMD 5950X with an ASUS B550-F. I tried installing the card in the slot I normally use for my RTX 3080 as well as in another slot that meets the specifications. In both cases, `lspci` does not list the card even though the fan spins and I know it is getting power. Also Voyager SDK dosn’t recognize the card. I also tried it on an Intel system with an Intel i5-8500T on a Supermicro X11SCA-F motherboard. The same issue occurs. The card is powered (fan spin) but not recognized. I noticed that the boot time increases significantly when the accelerator is installed. This makes me think that UEFI might be attempting to detect something, even though no error messages are shown. As I already tried a lot and I couldn’t get it running, I wonder if you have any ideas what I can test.Could there

Hey!PCIe card did run out of the box with a first running demo in minutes with no issues!Question: Is there a way to control the fan speed to reduce the noise of the card?Thanks!