Tag: #gpu-benchmark8 results
Intel Arc Pro B70 GPU achieves 80% higher AI inference performance than B60 in MLPerf v6.0 benchmarks. 4x B70 setup (128GB VRAM) excels on 120B LLMs and Llama models with Xeon 6. Software upgrades boost existing B60 by 18%.
Achieved 1.1M total tok/s serving Qwen 3.5 27B (FP8) on 96 B200 GPUs using vLLM v0.18.0. DP=8 outperforms TP=8 by 4x; MTP-1 boosts utilization, MTP-5 crashes. 97% scaling efficiency, but Inference Gateway adds 35% overhead.
Benchmarks show Intel Arc Pro B70 32GB achieving 12 tps on single Qwen3.5-27B@Q4 queries via llama.cpp/vllm, scaling to 135 tps at 32 concurrency – 20% behind RTX PRO 4500 but with 50% higher power draw. Requires Ubuntu 26.04 beta and beta vllm fork. Detailed Docker command provided for setup.
Reddit user benchmarks NVIDIA RTX 6000 96GB against single AMD W7800 48GB for LLM inference like GPT120b. Tokens/sec ratio (0.49) closely matches VRAM bandwidth ratio (0.48), proving speed dominance. Multi-GPU setups average performance, advising bandwidth priority over capacity alone.
Reported testing indicates the AMD Radeon RX 9050 performs about 20% below the RTX 5050 at 1080p. The GPU launched in late July for Latin America, Asia-Pacific, and Japan, but AMD reportedly did not provide review samples, resulting in limited global coverage.
Tested aging Titan X Pascal GPU for local AI tasks with llama.cpp and OpenCode. Achieved 500 tokens/sec prompt processing and 25 tokens/sec generation, rivaling newer AMD 9070 XT at half the gen speed. Proves old hardware remains viable for inference.
A new retest finds that the GTX 1080 Ti remains capable for 1080p gaming in many modern titles. Its 11GB VRAM helps it stay relevant, but its power efficiency and overall performance show the effects of age.
3DMark 2001 celebrates 25 years since its March 13, 2001 release. It popularized 3D graphics benchmarking for consumers. Article speculates on scores achievable with RTX 5090 GPU.