RTX 3090 Ti 24GB

NVIDIA · 24 GB GDDR6X · 1008 GB/s · 450 W · MSRP $1,999 · Search Amazon (affiliate)

The RTX 3090 Ti 24GB is a NVIDIA accelerator with 24 GB GDDR6X and about 1008 GB/s peak memory bandwidth. AI Hashrate estimates local LLM decode (batch 1) at 8K context: fit counts weights + KV cache + 1 GB runtime overhead — not weights alone. At Q4, about 33 curated models fit fully on this card; at FP16, about 17 fit. Top Q4 speeds: Qwen3-0.6B ≈ 801.8 tok/s (estimated, fits); gemma-3-1b-it ≈ 641.5 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 534.5 tok/s (estimated, fits); Qwen3-1.7B ≈ 320.7 tok/s (estimated, fits); Qwen3-30B-A3B ≈ 213.8 tok/s (estimated, fits). Relative ranking is more reliable than absolute tok/s. See methodology for the bandwidth formula and measured-anchor policy. Methodology.

Table: decode speed estimates (Q4 / FP16). Measured rows override estimates. MSRP is list price, not live retail. Context default 8K.

Model	Params (B)	Quant	tok/s	Fits?
Qwen3-0.6B	0.8	Q4	801.8 est.	Yes
gemma-3-1b-it	1.0	Q4	641.5 est.	Yes
Llama-3.2-1B-Instruct	1.2	Q4	534.5 est.	Yes
Qwen3-1.7B	2.0	Q4	320.7 est.	Yes
Qwen3-0.6B	0.8	FP16	220.5 est.	Yes
Qwen3-30B-A3B	30.5	Q4	213.8 est.	Yes
Qwen3-Coder-30B-A3B-Instruct	30.5	Q4	213.8 est.	Yes
GLM-4.7-Flash	31.2	Q4	213.8 est.	Yes
Qwen3-30B-A3B-Instruct-2507	30.5	Q4	213.8 est.	Yes
Llama-3.2-3B-Instruct	3.2	Q4	200.5 est.	Yes
gpt-oss-20b	21.5	Q4	178.2 est.	Yes
gemma-3-1b-it	1.0	FP16	176.4 est.	Yes
Qwen3-Coder-Next	79.7	Q4	170.4 est.	No
Phi-4-mini-instruct	3.8	Q4	168.8 est.	Yes
Phi-4-mini-reasoning	3.8	Q4	168.8 est.	Yes
gemma-3-4b-it	4.3	Q4	164.5 est.	Yes
Qwen3-Next-80B-A3B-Instruct	81.3	Q4	164.0 est.	No
Qwen3-4B-Instruct-2507	4.0	Q4	160.4 est.	Yes
Qwen3-4B-Thinking-2507	4.0	Q4	160.4 est.	Yes
Llama-3.2-1B-Instruct	1.2	FP16	147.0 est.	Yes
Mixtral-8x7B-Instruct-v0.1	46.7	Q4	95.5 est.	No
Mistral-7B-Instruct-v0.3	7.2	Q4	89.1 est.	Yes
Qwen3-1.7B	2.0	FP16	88.2 est.	Yes
Qwen2.5-7B-Instruct	7.6	Q4	84.4 est.	Yes
Qwen2.5-Coder-7B-Instruct	7.6	Q4	84.4 est.	Yes
DeepSeek-R1-Distill-Qwen-7B	7.6	Q4	84.4 est.	Yes
Llama-3.1-8B-Instruct	8.0	Q4	80.2 est.	Yes
Qwen3-8B	8.2	Q4	78.2 est.	Yes
DeepSeek-R1-0528-Qwen3-8B	8.2	Q4	78.2 est.	Yes
Llama-3.2-3B-Instruct	3.2	FP16	55.1 est.	Yes
gemma-3-12b-it	12.2	Q4	53.5 est.	Yes
Qwen3-32B	32.8	Q4	49.1 est.	No
Qwen2.5-32B-Instruct	32.8	Q4	49.1 est.	No
Qwen2.5-Coder-32B-Instruct	32.8	Q4	49.1 est.	No
DeepSeek-R1-Distill-Qwen-32B	32.8	Q4	49.1 est.	No
Phi-4-mini-instruct	3.8	FP16	46.4 est.	Yes
Phi-4-mini-reasoning	3.8	FP16	46.4 est.	Yes
gemma-3-4b-it	4.3	FP16	45.2 est.	Yes
gpt-oss-120b	120.4	Q4	44.5 est.	No
Qwen3-4B-Instruct-2507	4.0	FP16	44.1 est.	Yes
Qwen3-4B-Thinking-2507	4.0	FP16	44.1 est.	Yes
Phi-4	14.7	Q4	43.6 est.	Yes
Phi-4-reasoning	14.7	Q4	43.6 est.	Yes
Qwen3-14B	14.8	Q4	43.3 est.	Yes
Qwen2.5-14B-Instruct	14.8	Q4	43.3 est.	Yes
Qwen2.5-Coder-14B-Instruct	14.8	Q4	43.3 est.	Yes
DeepSeek-R1-Distill-Qwen-14B	14.8	Q4	43.3 est.	Yes
gpt-oss-20b	21.5	FP16	40.3 est.	No
gemma-3-12b-it	12.2	FP16	31.3 est.	No
Devstral-Small-2507	23.6	Q4	27.2 est.	Yes
Mistral-Small-3.2-24B-Instruct-2506	24.0	Q4	26.7 est.	Yes
Magistral-Small-2509	24.0	Q4	26.7 est.	Yes
Qwen3-30B-A3B	30.5	FP16	24.7 est.	No
Qwen3-Coder-30B-A3B-Instruct	30.5	FP16	24.7 est.	No
Qwen3-30B-A3B-Instruct-2507	30.5	FP16	24.7 est.	No
Mistral-7B-Instruct-v0.3	7.2	FP16	24.5 est.	Yes
gemma-3-27b-it	27.4	Q4	23.6 est.	Yes
GLM-4.7-Flash	31.2	FP16	23.6 est.	No
Qwen2.5-7B-Instruct	7.6	FP16	23.2 est.	Yes
Qwen2.5-Coder-7B-Instruct	7.6	FP16	23.2 est.	Yes
DeepSeek-R1-Distill-Qwen-7B	7.6	FP16	23.2 est.	Yes
Llama-3.1-8B-Instruct	8.0	FP16	22.0 est.	Yes
Qwen3-8B	8.2	FP16	21.5 est.	Yes
DeepSeek-R1-0528-Qwen3-8B	8.2	FP16	21.5 est.	Yes
GLM-4.5-Air	110.5	Q4	21.4 est.	No
Phi-4	14.7	FP16	17.8 est.	No
Phi-4-reasoning	14.7	FP16	17.8 est.	No
Qwen3-14B	14.8	FP16	17.4 est.	No
Qwen2.5-14B-Instruct	14.8	FP16	17.4 est.	No
Qwen2.5-Coder-14B-Instruct	14.8	FP16	17.4 est.	No
DeepSeek-R1-Distill-Qwen-14B	14.8	FP16	17.4 est.	No
DeepSeek-V4-Flash	158.1	Q4	9.9 est.	No
Llama-3.1-70B-Instruct	70.6	Q4	5.2 est.	No
Llama-3.3-70B-Instruct	70.6	Q4	5.2 est.	No
DeepSeek-R1-Distill-Llama-70B	70.6	Q4	5.1 est.	No
Qwen2.5-72B-Instruct	72.7	Q4	4.7 est.	No
Devstral-Small-2507	23.6	FP16	4.4 est.	No
Mistral-Small-3.2-24B-Instruct-2506	24.0	FP16	4.2 est.	No
Magistral-Small-2509	24.0	FP16	4.2 est.	No
gemma-3-27b-it	27.4	FP16	2.8 est.	No
Qwen3-32B	32.8	FP16	1.7 est.	No
Qwen2.5-32B-Instruct	32.8	FP16	1.7 est.	No
Qwen2.5-Coder-32B-Instruct	32.8	FP16	1.7 est.	No
DeepSeek-R1-Distill-Qwen-32B	32.8	FP16	1.7 est.	No

Fit checks

• Qwen3-0.6B — fits · ≈801.8 tok/s Q4 est.
• Qwen3-8B — fits · ≈78.2 tok/s Q4 est.
• Qwen2.5-7B-Instruct — fits · ≈84.4 tok/s Q4 est.
• Llama-3.2-1B-Instruct — fits · ≈534.5 tok/s Q4 est.
• Llama-3.1-8B-Instruct — fits · ≈80.2 tok/s Q4 est.
• gpt-oss-20b — fits · ≈178.2 tok/s Q4 est.
• Qwen3-32B — does not fit · ≈49.1 tok/s Q4 est.
• Qwen3-1.7B — fits · ≈320.7 tok/s Q4 est.
• Mistral-7B-Instruct-v0.3 — fits · ≈89.1 tok/s Q4 est.
• Qwen3-14B — fits · ≈43.3 tok/s Q4 est.
• Qwen3-4B-Instruct-2507 — fits · ≈160.4 tok/s Q4 est.

All 61 model fit checks — see the linked Fits? column in the table above