RTX 5070 too slow for 27B? Reddit weighs Intel Arc B60/B65 budget builds
hidden2u · reddit · 2026-08-28
A Reddit user gets only 3 t/s running Qwen3-30B-A3B on an RTX 5070 and has to fall back to a 35B model with offloading. On a sub-$1000 budget, they're eyeing the $650 Intel Arc B60 or $900 B65 Pro, reusing spare AM4 parts.
They know Arc's software stack lags CUDA and even ROCm, but YouTube tests show Intel's custom vLLM fork hitting 20 t/s on the larger $1200 B70. Better options like Strix, DGX and the 5090 all exceed the budget.
More from Infra
- Qualcomm's data center opportunity becomes tangible with custom ARM CPUs — BenBajarin · 2026-08-28
- Merge API Gateway Sees 11x Month-Over-Month Spike in Token Volume — shensi · 2026-08-28
- Local Inference Tool ds4 Adds Support for GLM 5.3 Flash — lakySK · 2026-08-28
- Running Qwen3.8-Flash on RTX 3090: Quantization & Performance — crusaderky · 2026-08-28
- Gemini 1.5 Flash inference speeds may exceed 300 tok/sec — Sentdex · 2026-08-28
- Paper analyzes NVSHMEM: System-level insights into GPU communication — thoefler · 2026-08-28