Skip to content

Repository files navigation


#XAIGPUARC


ENGLISH WELCOME / DEUTSCH WILLKOMMEN

#9.) How to START your XAIGPUARC ONEKLICK AI-MACHINE?!

#0.) FIRST INTEL_ONE_API_BASEKIT MUST BE INSTALLED ON YOUR PC/LAPTOP/SYSTEM USE ---PREXAIGPUARC--- for HELP!!! Download the official Paket from Intel itself and install it manuall if you got problems above your Paketmanager Install.

#0.) Second. Use the MKL/LEVELZERO Script to proof everything goes right for your System!

#1.) Kopie XAIGPUARC.sh in your Home/PCNAME/ Folder!

#2.) Between XAIGPUARC Full INSTALLATION Download a gguf AI fit in your

#a.) V/RAM to /models/HereAINAME also in your Home/PCNAME/models/HereAINAME Folder!

#3.) Change your Modell manual in the XAIGPUARC.sh!

#b.) Open Console and Type: chmod +x ./XAIGPUARC.sh Enter...

#4.) START with type in Console ./XAIGPUARC.sh Enter...

sudo pacman -Rns intel-compute-runtime intel-level-zero-gpu oneapi-level-zero

Pakete installieren: Installiere die Arch-nativen Treiber, um Binär-Konflikte mit dem oneAPI-Installer zu vermeiden:

sudo pacman -S intel-compute-runtime level-zero-loader intel-opencl-clang ocl-icd

ICD-Konflikte lösen: Deaktiviere alte/doppelte Profile, damit clinfo nicht über intel64.icd stolpert:

sudo mv /etc/OpenCL/vendors/intel64.icd /etc/OpenCL/vendors/intel64.icd.disabled

Environment setzen: Zwinge das System auf das korrekte Intel-Profil (Wichtig für A770-Erkennung):

export OCL_ICD_VENDORS=/etc/OpenCL/vendors/intel.icd

oneAPI-Vars (falls nötig): Initialisiere die Pfade nur für den Compiler, nicht für die Runtime:

bass source /opt/intel/oneapi/setvars.sh --force

Funktionstest: Prüfe die GPU-Erkennung (muss jetzt ohne "Binärdatei-Fehler" laufen):

sycl-ls && clinfo | grep -i "Arc A770"

sudo pacman -Rns intel-compute-runtime intel-level-zero-gpu oneapi-level-zero des zuerst, dann im pakemanager oneapi raus und toolkit version installieren dann das hier bass source /opt/intel/oneapi/setvars.sh --force sudo pacman -S intel-compute-runtime level-zero-loader intel-opencl-clang ocl-icd und dann das hier wenn der befehl stimmt sudo mv /etc/OpenCL/vendors/intel64.icd /etc/OpenCL/vendors/intel64.icd.disabled auch das habe ich gemacht export OCL_ICD_VENDORS=/etc/OpenCL/vendors/intel.icd sycl-ls && clinfo | grep -i "Arc A770"

HARDWARE: CPU/iGPU/dGPU from INTEL

#-XAIGPUARC Hardware used to Build and Test

#-6x Intel ARC 2xA770LE 16GB + 4x750LE 8GB

#-90-142 Watt Chip Power Draw alone each Card at different LLMs

#-Example: GPT-OSS-20B-F16 does it very nice at low Wattage

#-but needs longer than full working MathTutor F16 with 142 Watt

#-Use multible Models for a better Workflow

#-All the Hardware is Modded and not Stock Compareable!

#-PLS watch your good Cooling and Dust Free System

#-3x Single and Dual dGPUs on AMD Ryzen 2600 2700x / Intel i7 6700K on Z170 RAM 16GB till 128GB

#-2x Intel iGPU XE Alder Lake Gen (12700H + 12650H) + A730m 12 GB + 32GB DDR4/5 RAM

#-1x Intel Core Ultra 7 155H Meteor Lake 8 Core Xe-LPG 128EU ARC 16GiB

#-Quad Channel High Bandwith RAM Gear2 with 718GB/s

#-11,5 GiB VRAM shared from this RAM

#-On 155H i7 GPT-OSS-20B-F16.gguf runs well but slow at 30 Watt allinone with mods

#-BF16 Models not recommend for Alechmist

#-1+ till 6+ means 1+ is best double ++ is Better +++ is Insane Wonderfull AI Magic

#-CTX-NPG is the Context Size and N-Predigt Size

#-Lower them if your Modell not fit! You see this do not care about the GPU alone below

#F16 Mode Only #6GB+ GPU A730m/A380/A310

#kani-tts-400m-en-f16_q8_0.gguf

#0.53 GB FAST CTX-NPG 8K A770LE: 588.6 Pt/s 62.4 Gt/s 100w 2.4Ghz - CPU FIRESTARTER 1++

#baidu.ERNIE-4.5-0.3B-Base-PT.f16.gguf

#0.69 GB FAST CTX-NPG 8K A770LE: 469.7 Pt/s 52.5 Gt/s 97w 2.4Ghz + CPU Mid 3+

#MedScholar-1.5B-f16_q8_0.gguf

#2.1 GB FAST CTX-NPG 8k A770LE: 528.2 Pt/s 25.2 Gt/s 109w 2.4Ghz - CPU 2+

#Qwen2.5-VL-3B-Instruct-f16-q4_k.ggu

#2.1 GB FAST CTX-NPG 16k A730m: 511.2 Pt/s 12.5 Gt/s 65w 2.05Ghz - CPU 1++

#yasserrmd.DentaInstruct-1.2B.f16.gguf

#2.2 GB

#DeepCoder-1.5B-Preview-f16_q8_0.gguf

#2.2 GB FAST CTX-NPG 8k A770LE: 513.2 Pt/s 23.3 Gt/s 112w 2.3Ghz + CPU Mid TK 3+

#ibm-granite.granite-4.0-1b.f16.gguf

#3 GB SLOW CTX-NPG 8k A770LE: 569.4 Pt/s 18.2 Gt/s 120w 2.3Ghz - CPU 5+

#Lucy-1.7B-F16.gguf

#3.2 GB FAST CTX-NPG 16k A770LE: 572.7 Pt/s 23.2 Gt/s 108w 2.4Ghz - CPU TK 1++

#FAST CTX-NPG 16k A730m: 382.2 Pt/s 14.7 Gt/s 65w 2.05Ghz - CPU TK 1++

#FAST CTX-NPG 16k iGPU8XE: 2384.5 Pt/s 9.0 Gt/s 30w 2.25Ghz - CPU TK 1++

#granite-4.0-micro-f16_q8_0.gguf

#4.6 GB 5+

#gemma-2-2b-it.F16.gguf

#4.9 GB FAST CTX-NPG 16k A730m: 305.8 Pt/s 14.4 Gt/s 65w 2.2Ghz - CPU 2+

#8GB+ GPU A750LE

#Fathom-Search-4B-f16_q8_0.gguf

#5.5 GB FAST CTX-NPG 8k A770LE: 569.4 Pt/s 18.2 Gt/s 118w 2.4Ghz - CPU TK 2+

#Qwen2.5-7B-Instruct-f16-q4_k.gguf

#5.7 GB FAST CTX-NPG 8k A770LE: 511.5 Pt/s 19.7 Gt/s 142w 2.4Ghz - CPU 1+

#Qwen2.5-VL-3B-Instruct-f16.gguf

#5.8 GB FAST 1+

#SmolLM3-3B-f16.gguf

#5.8 GB FAST CTX-NPG 16k A770LE: 2137.2 Pt/s 19.1 Gt/s 119w 2.4Ghz - CPU TK 1++

#FAST CTX-NPG 16k iGPU8XE: 4640.1 Pt/s 6.9 Gt/s 30w 2.25Ghz - CPU TK 1++

#FAST CTX-NPG 16k A730m: 965 Pt/s 13.3 Gt/s 65w 2.05Ghz - CPU TK 1++

#MiniCPM-V-4-f16.gguf

#6.7 GB FAST CTX-NPG 16k A770LE: 619.8 Pt/s 19.8 Gt/s 142w 2.4Ghz - CPU

#FAST CTX-NPG 16k A730m: 382.2 Pt/s 14.7 Gt/s 65w 2.05Ghz - CPU 1++

#FAST CTX-NPG 16k iGPU8XE: 880.5 Pt/s 7.2 Gt/s 30w 2.25Ghz - CPU 1++

#10-12GB+ iGPU Xe-LPG/A730m/A580/B570/B580/PROA60/B50

#Qwen3-4B-f16.gguf

#7.5 GB FAST CTX-NPG 8k A770LE: 613.4 Pt/s 14.5 Gt/s 120w 2.4Ghz - CPU TK 1+

#Nemotron-Mini-4B-Instruct-f16.gguf

#7.8 GB FAST CTX-NPG 8k A770LE: 717.8 Pt/s 17.8 Gt/s 118w 2.4Ghz - CPU 2+

#SLOW CTX-NPG 16k iGPU8XE: 534.8 Pt/s 6.5 Gt/s 30w 2.25Ghz - CPU 1++

#Minitron-4B-Base.FP16.gguf

#7.8 GB FAST CTX-NPG 4k A770LE: 764.3 Pt/s 16.3 Gt/s 131w 2.4Ghz + CPU Mid 4+

#t5-v1_1-xxl-encoder-f16.gguf

#8.9 GB FAST CTX-NPG 8k A770LE: 361,8 Pt/s 6 Gt/s 101w 2.4Ghz - CPU 2++

#16GB+ GPU A770LE + iGPU Meteor Lake

#MiMo-Embodied-7B-f16_q8_0.gguf

#10.7 GB

#MiniCPM4.1-8B-f16_q8_0.gguf

#11 GB FAST CTX-NPG 8k A770LE: 842.9 Pt/s 11.0 Gt/s 142w 2.4Ghz + CPU 1+

#KernelLLM-f16_q8_0.gguf

#11.1 GB FAST CTX-NPG 8k A770LE: 688.5 Pt/s 11.2 Gt/s 137w 2.4Ghz - CPU 1+

#SLOW CTX-NPG 16k iGPU8XE: 29.6 Pt/s 3.0 Gt/s 30w 2.25Ghz - CPU 1+

#Jan-v2-VL-high-f16_q8_0.gguf

#11.4 GB FAST CTX-NPG 8k A770LE: 639.6 Pt/s 10.2 Gt/s 135w 2.4Ghz - CPU TK 2+

#Orchestrator-8B-f16_q8_0.gguf

#11.4 GB FAST CTX-NPG 16k A770LE: 643.3 Pt/s 10.1 Gt/s 134w 2.4Ghz - CPU TK 2+

#MiroThinker-v1.0-8B-f16_q8_0.gguf

#11.4 GB 2+

#Seed-Coder-8B-Reasoning-f16_q8_0.gguf

#11.5 GB 2+

#Ministral-3-8B-Reasoning-2512-f16_q8_0.gguf

#11.7 GB 2+

#ggml-model-f16.gguf

#12.6 GB FAST CTX-NPG 4k A770LE: 1012.7 Pt/s 13.5 Gt/s 142w 2.4Ghz - CPU 4+

#gpt-oss-20b-F16.gguf

#12.8 GB SLOW CTX-NPG 16k A770LE: 34.5 Pt/s 7.7 Gt/s 90W 2.3Ghz + FULL CPU 1+

#SLOW CTX-NPG 16k iGPU8XE: 9.6 Pt/s 4.6 Gt/s 30w 2.25Ghz - CPU 1++

#Navid-AI.Yehia-7B-preview.f16.gguf

#13 GB FAST CTX-NPG 4k A770LE: 1273.4 Pt/s 13.4 Gt/s 142w 2.4Ghz - CPU 1++

#Mistral-7B-Instruct-v0.3.fp16.gguf

#13.5 GB 2+

#Mamba-Codestral-7B-v0.1-F16.gguf

#13.6 GB SLOW CTX-NPG 16k A770LE: 110.1 Pt/s 3.2 Gt/s 97w 2.4Ghz + CPU FULL 2+

#MathTutor-7B-H_v0.0.1.f16.gguf

#14.2 GB FAST CTX-NPG 16k A770LE: 525.0 Pt/s 13.8 Gt/s 142w 2.4Ghz - CPU 1++

#END F16 MODEL LIST

#START Q8-Q4-IQ4-2 MODEL LIST NOT F16!

#6GB+ GPU A730m/A380/A310

#phi-2.Q4_K_M.gguf

#1.7 GB FAST CTX-NPG 8k A770LE: 888.6 Pt/s 25.4 Gt/s 128w 2.4Ghz - CPU 1++

#openhermes-2.5-mistral-7b.Q4_K_M.gguf

#4.1 GB FAST 2+

#mistral-7b-instruct-v0.2.Q4_K_M.gguf

#4.1 GB SLOW 2+

#8GB+ GPU A750LE

#OpenMath-Mistral-7B-v0.1-hf_Q6_K.gguf

#5.5 GB FAST CTX-NPG 8k A770LE: 1233.9 Pt/s 14.4 Gt/s 145w 2.4Ghz - CPU 1+

#NVIDIA-Nemotron-Nano-12B-v2-IQ4_NL.gguf

#6.6 GB SLOW CTX-NPG 16k iGPU8XE: 28.6 Pt/s 2.3 Gt/s 30w 2.15Ghz - CPU 2++

#wizardcoder-python-7b-v1.0.Q8_0.gguf

#6.7 GB SLOW 2+

#sauerkrautlm-7b-v1.Q8_0.gguf

#6.7 GB FAST CTX-NPG 8k A770LE: 1364.6 Pt/s 12.1 Gt/s 142w 2.4Ghz - CPU 2+

#10-12GB+ iGPU Xe-LPG/A730m/A580/B570/B580/PRO60/B50

#Qwen3-16B-A3B-IQ4_NL.gguf

#8.5 GB FAST 2+

#Qwen3-30B-A3B-UD-IQ2_XXS.gguf

#9.7 GB SLOW CTX-NPG 16k iGPU8XE: 16.3 Pt/s 4.0 Gt/s 30w 2.25Ghz - CPU TK 2+

#solar-10.7b-instruct-v1.0-uncensored.Q8_0.gguf

#10.6 GB FAST CTX-NPG 16k A770LE: 985.6 Pt/s 7.5 Gt/s 135w 2.4Ghz - CPU 1++

#gpt-oss-20b-claude-4-distill.MXFP4_MOE.gguf

#11.3 GB SLOW CTX-NPG 8k A770LE: 35.4 Pt/s 8.7 Gt/s 92W 2.2Ghz + FULL CPU 3+

#SLOW CTX-NPG 16k A730m: 14.9 Pt/s 4.5 Gt/s 40w 1.3Ghz + FULL CPU 3+

#16GB+ A770LE #velara-11b-v2.Q8_0.gguf #11.3 GB FAST 2+++ #Deepseek-Coder-V2-Lite-13B-Instruct-sft-s1K.i1-Q6_K.gguf

#13.1 GB FAST CTX-NPG 8k A770LE: 22.7 Pt/s 7.9 Gt/s 98W 2.4Ghz - CPU 3+

About

This is a AI Installation Tool specialised for Linux Garuda Arch / ARC Intel dGPUs and older XE Iris iGPUS to run over SYCL with Llama.cpp.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages