informatique:egpu
Différences
Ci-dessous, les différences entre deux révisions de la page.
| Les deux révisions précédentesRévision précédenteProchaine révision | Révision précédente | ||
| informatique:egpu [14/01/2026 08:38] – [nvidia-headless-575-open] cyrille | informatique:egpu [08/02/2026 10:31] (Version actuelle) – [PNY OC 16GB Geforce RTX 5060 Ti] cyrille | ||
|---|---|---|---|
| Ligne 14: | Ligne 14: | ||
| * acheté | * acheté | ||
| * ✅ RTX 3060 ok | * ✅ RTX 3060 ok | ||
| - | * ✗ RTX 5060 à peu près ok ([[informatique: | + | * ❌ RTX 5060 à peu près ok ([[informatique: |
| - | * [[https:// | + | * Ça plante après plusieurs inférences (llama.cpp, cuda 12.9) |
| - | * Le vendeur dit qu' | + | |
| * [[http:// | * [[http:// | ||
| * acheté | * acheté | ||
| Ligne 22: | Ligne 21: | ||
| * ✅ RTX 3060 ok | * ✅ RTX 3060 ok | ||
| * ❌ RTX 5060 failed | * ❌ RTX 5060 failed | ||
| - | * [[https:// | + | |
| + | * Le vendeur dit qu' | ||
| + | * Mais des acheteurs signalent des déconnexions https:// | ||
| + | * ADT UT4G | ||
| + | * USB4/ | ||
| + | * https:// | ||
| + | * $128 https:// | ||
| + | | ||
| + | * AOOSTAR | ||
| + | * AG02 Oculink/ | ||
| + | * $219 https:// | ||
| + | * AOOSTAR EG02 TB5+Oculink | ||
| + | * $219 https:// | ||
| + | * acheté | ||
| + | * ✅ RTX 3060 ok | ||
| + | * ❌ RTX 5060 failed | ||
| + | * Minisforum DEG2 OCulink Thunderbolt 5 eGPU Dock | ||
| + | * Thunderbolt 5 Port | Up to 80Gps, OCuLink (PCIe 4.0 ×4) | Up to 64Gps, Built-in M.2 2280 SSD, Compatible with ATX / SFX PSU | ||
| + | * $259 https:// | ||
| + | * EXP-GDC TH5P4 | ||
| + | * ??? | ||
| Au final on ne fait tourner que de petits models avec de petit context ... | Au final on ne fait tourner que de petits models avec de petit context ... | ||
| Ligne 147: | Ligne 166: | ||
| parm: | parm: | ||
| parm: | parm: | ||
| + | |||
| + | $ systool -m nvidia_uvm -v | ||
| + | |||
| + | Module = " | ||
| + | Attributes: | ||
| + | coresize | ||
| + | initsize | ||
| + | initstate | ||
| + | refcnt | ||
| + | srcversion | ||
| + | taint = " | ||
| + | uevent | ||
| + | version | ||
| + | Parameters: | ||
| + | uvm_ats_mode | ||
| + | uvm_block_cpu_to_cpu_copy_with_ce= " | ||
| + | uvm_channel_gpfifo_loc= " | ||
| + | uvm_channel_gpput_loc= " | ||
| + | uvm_channel_num_gpfifo_entries= " | ||
| + | uvm_channel_pushbuffer_loc= " | ||
| + | uvm_conf_computing_channel_iv_rotation_limit= " | ||
| + | uvm_cpu_chunk_allocation_sizes= " | ||
| + | uvm_debug_enable_push_acquire_info= " | ||
| + | uvm_debug_enable_push_desc= " | ||
| + | uvm_debug_prints | ||
| + | uvm_disable_hmm | ||
| + | uvm_downgrade_force_membar_sys= " | ||
| + | uvm_enable_builtin_tests= " | ||
| + | uvm_enable_debug_procfs= " | ||
| + | uvm_enable_va_space_mm= " | ||
| + | uvm_exp_gpu_cache_peermem= " | ||
| + | uvm_exp_gpu_cache_sysmem= " | ||
| + | uvm_fault_force_sysmem= " | ||
| + | uvm_force_prefetch_fault_support= " | ||
| + | uvm_global_oversubscription= " | ||
| + | uvm_leak_checker | ||
| + | uvm_page_table_location= " | ||
| + | uvm_peer_copy | ||
| + | uvm_perf_access_counter_batch_count= " | ||
| + | uvm_perf_access_counter_migration_enable= " | ||
| + | uvm_perf_access_counter_threshold= " | ||
| + | uvm_perf_fault_batch_count= " | ||
| + | uvm_perf_fault_coalesce= " | ||
| + | uvm_perf_fault_max_batches_per_service= " | ||
| + | uvm_perf_fault_max_throttle_per_service= " | ||
| + | uvm_perf_fault_replay_policy= " | ||
| + | uvm_perf_fault_replay_update_put_ratio= " | ||
| + | uvm_perf_map_remote_on_eviction= " | ||
| + | uvm_perf_map_remote_on_native_atomics_fault= " | ||
| + | uvm_perf_migrate_cpu_preunmap_block_order= " | ||
| + | uvm_perf_migrate_cpu_preunmap_enable= " | ||
| + | uvm_perf_pma_batch_nonpinned_order= " | ||
| + | uvm_perf_prefetch_enable= " | ||
| + | uvm_perf_prefetch_min_faults= " | ||
| + | uvm_perf_prefetch_threshold= " | ||
| + | uvm_perf_reenable_prefetch_faults_lapse_msec= " | ||
| + | uvm_perf_thrashing_enable= " | ||
| + | uvm_perf_thrashing_epoch= " | ||
| + | uvm_perf_thrashing_lapse_usec= " | ||
| + | uvm_perf_thrashing_max_resets= " | ||
| + | uvm_perf_thrashing_nap= " | ||
| + | uvm_perf_thrashing_pin= " | ||
| + | uvm_perf_thrashing_pin_threshold= " | ||
| + | uvm_perf_thrashing_threshold= " | ||
| + | uvm_release_asserts = " | ||
| + | uvm_release_asserts_dump_stack= " | ||
| + | uvm_release_asserts_set_global_error= " | ||
| </ | </ | ||
| + | |||
| + | Le plantage de la RTX 5060 Ti arrive plus tard si '' | ||
| ==== Séries RTX ==== | ==== Séries RTX ==== | ||
| Ligne 186: | Ligne 274: | ||
| J'ai acheté un câble Thunderbolt certifié (50€) pour remplacer celui fourni avec l' | J'ai acheté un câble Thunderbolt certifié (50€) pour remplacer celui fourni avec l' | ||
| + | |||
| + | Toujours plantage avec driver nvidia 590, cuda 13.1 et conf modprobe. | ||
| + | |||
| === nvidia-kkms-565 === | === nvidia-kkms-565 === | ||
| Ligne 276: | Ligne 367: | ||
| === Avec le bridge Wikingoo WKGL17-C50 === | === Avec le bridge Wikingoo WKGL17-C50 === | ||
| - | Avec certains modèles ya " | + | Avec certains modèles ya " |
| < | < | ||
informatique/egpu.1768376287.txt.gz · Dernière modification : de cyrille