Builds / RTX 5090 32GB

RTX 5090 on Windows 11: Ornith 1.5 35B-A3B at 273 tok/s, Gemma 4 26B-A4B at 232, Qwen3.8 27B 73.6 (137 with MTP)

Name: RTX 5090 on Windows 11: Ornith 1.5 35B-A3B at 273 tok/s, Gemma 4 26B-A4B at 232, Qwen3.8 27B 73.6 (137 with MTP)
Creator: RigBase
License: https://creativecommons.org/licenses/by/4.0/

Source: kgptalkie.com ↗More source: kgptalkie.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source

GPU(s): RTX 5090 32GB
Amazon · eBay (used) · Newegg
VRAM / unified memory (GB): 32
CPU: not stated
Motherboard: not stated
RAM: not stated
PSU: not stated
Other: llama-server/llama.cpp build b10448 (Windows 11 per the Ornith article), full GPU offload, flash attention on. The speed-settings article reports the card at 599 W against its 600 W cap under load
Reported total cost: not stated
Power: not stated

Measured speed (generation tokens/sec)

Model	Quant	Engine	tok/s	Prompt processing (tok/s)	Details
Ornith 1.5 35B-A3B	Q4_K_M, f16 KV	llama-server b10448, no MTP	273.2	7,547 @ 8k-token prompt est. TTFT @8k ≈ 1.1 s (calculated from measured prefill)	context: 8,192 ctx, 512 generated tokens (ignore_eos)📏 Decode table in the Ornith 1.5 benchmark article
Gemma 4 26B-A4B	Q4_K_M (text-only build), f16 KV	llama-server b10448	232.4	9,719 @ 8k-token prompt est. TTFT @8k ≈ 0.84 s (calculated from measured prefill)	context: 8,192 ctx, 512 generated tokens (ignore_eos)📏 Decode table in the Ornith 1.5 benchmark article
Qwen3.8 27B	Q4_K_M, f16 KV	llama.cpp b10448, MTP off	73.6	3,036 (prompt length not stated)	📏 Median of 3 runs, thinking on, f16 KV, MTP off (speed-settings article). The Ornith article measured 77.2 tok/s for the same model and card in a separate session
Qwen3.8 27B	Q4_K_M, q4_0 KV	llama.cpp b10448, MTP n=2 (--spec-type draft-mtp)	136.7	not stated	⚡ spec-decode (MTP)context: 32,768 ctx📏 Median of 3 runs at 32,768 ctx (the article's recommended config); with f16 KV the n=2/3/4 sweep gave 125.5 / 133.6 / 119.5 tok/s

Prompt processing = prefill speed (how fast the prompt is read) as reported by the source, with the prompt length it was measured at; blank = the source does not report it. est. TTFT @8k is calculated from the measured prefill speed (8,192 ÷ prefill tok/s), shown only when prefill was measured on a prompt of 512+ tokens. It is not a measurement and ignores load time and caching. measured TTFT is the source's own time-to-first-token. Why prefill matters for coding assistants.

KGP Talkie (Laxmi Kant Tiwari): Qwen3.8 27B speed-settings article (16 Aug 2026, updated 29 Aug) and Ornith 1.5 benchmark (25 Aug 2026, updated 29 Aug), same card and llama.cpp build b10448. CPU, board, RAM and PSU not stated.

Current GPU street price (estimate, not the builder's cost)

1× RTX 5090: ≈€7,140 each (≈$8,003), new, approx. price as of 2026-10-24, lowest listing on billiger.de (MSI GeForce RTX 5090 (lowest variant))

Estimated GPU cost for this build today: ≈€7,140
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.

Source: https://kgptalkie.com/tutorials/llm-benchmarking/qwen-3-8-27b-llama-cpp-speed-settings
https://kgptalkie.com/tutorials/llm-benchmarking/ornith-1-5-9b-vs-35b-a3b-benchmark

Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Quick answers: Can RTX 5090 run Gemma 4 26B-A4B? · Can RTX 5090 run Ornith 1.5 35B-A3B? · Can RTX 5090 run Qwen3.8 27B?