Machine Updated 21 October 2026

RTX 5070 Ti desktop

A desktop with an NVIDIA GeForce RTX 5070 Ti: 16GB of graphics memory, a Ryzen 9 5950X and 32GB of memory.

A GeForce RTX 5070 family graphics card — NVIDIA's RTX 5070 family; the desktop has an RTX 5070 Ti. Image: NVIDIA

Measured on mine measured by me

Set-up	Model	Writing code	Writing prose	Reading (prefill)	Tested
RTX 5070 Ti desktopllama.cpp on CUDA	Gemma 4 12BGoogle's own 4-bit QAT file (Q4_0), GGUF	92.5 tok/s	94.4 tok/s	2,910 tok/s	20 Oct 2026
RTX 5070 Ti desktopllama.cpp on CUDA	Gemma 4 26B-A4BUD-IQ4_XS, about 4 bits, GGUF	59.4 tok/s	59.8 tok/s	2,199 tok/s	21 Oct 2026
RTX 5070 Ti desktopllama.cpp on CUDA	gpt-oss-20bMXFP4, GGUF, thinking at its lowest level	213.2 tok/s	214.4 tok/s	7,884 tok/s	21 Oct 2026
RTX 5070 Ti desktopllama.cpp on CUDA	Qwen 3.5 9BQ4_K_M, GGUF	126.6 tok/s	127.0 tok/s	5,275 tok/s	20 Oct 2026
RTX 5070 Ti desktopPrism ML's llama.cpp on CUDA	Ternary Bonsai 2 27BPTQ1_0, about 1.75 bits a weight, GGUF	98.8 tok/s	99.8 tok/s	1,730 tok/s	20 Oct 2026

One user, the model's thinking switched off, the same prompts and the same test each time. All of my results, and how I test, are on the hardware leaderboard.

The hardware, exactly

Graphics card: NVIDIA GeForce RTX 5070 Ti with 16GB of its own memory.
Processor: AMD Ryzen 9 5950X, 16 cores and 32 threads.
Memory: 32GB of DDR4 as two 16GB modules running at 4,000MT/s (G.Skill F4-4000C16-16GTZRA).
Motherboard: Gigabyte B550 AORUS PRO AC.
System: Windows 11 Pro.