AI Briefing
KO

100tps with 5070 Ti + RX 9070

·2026.04.18 07:22

Key point

Combining an RX 9070 and RTX 5070 Ti, the author ran Qwen 35B Q4 at over 100tps.

Details

On an i5-12400, B660, 32GB DDR4 system with both an RX 9070 (non-XT) and an RTX 5070 Ti installed together, running Qwen3.6-35B-A3B-UD-Q4_K_M via llama.cpp Vulkan achieved over 100 tps at 64K context.

The settings used were --split-mode layer, --main-gpu 0, --cache-type-k q4_0, --cache-type-v q4_0, --ctx-size 65536, effectively pooling the VRAM of both GPUs to attempt running a large model.

Additionally, with the same setup:

  • 5070 Ti: handled Qwen 9B and Stable Diffusion without issue
  • 9070: about 50 tps for text generation
  • SillyTavern: split operation with the 9070 for text and the 5070 Ti for image generation

The author noted that the two cards together cost around $1300, and evaluated this combination as a potential cheap alternative for ~32GB-class VRAM.

benchmarkqwenlocal-llmvulkanllama-cppamd-gpunvidia-gpuvram-pooling
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditMica v0.1 4B Released: Open-Source Decision Model for Agent Loops Trained for Under $30TLDR AIQwen Intelligence Launches Three Mobile AI Agents and Open Benchmark SuiteRedditDynamic Quantiser Released, Improving Accuracy Over Standard Quantization in VRAM-Constrained EnvironmentsHacker NewsApple Releases LensVLM-9B for Processing Compressed Text ImagesAnthropic NewsAnthropic Releases Claude Opus 5.5 with 40% Cost Reduction and Record Alignment ScoresTLDR AIQwen Releases 'RecreationWorld' for Training AI Agents to Rebuild Apps