# Multi-GPU for local LLMs: dual GPU vs one card (2026) > Can two or more GPUs run bigger local LLMs? VRAM, list-price cost and est. tok/s for 2–4 identical cards vs one card or a Mac, for 70B, gpt-oss-120b and 32B at Q8. Source: https://whatcanirunai.vercel.app/multi-gpu-for-local-llm/ # Multi-GPU for local LLMs: two cards or one big one? When one graphics card doesn't have enough VRAM, you can split a model across two, three or four. This page shows what each setup holds, what it costs at list price, and how it compares with one bigger card or a unified-memory Mac or mini PC. All prices come from the 82 devices we track; speeds are estimates. GPUs tracked 40 Biggest single card 96 GB 4 of them 384 GB Short answer: extra cards add memory, not speed. To run Llama-3.3-70B-Instruct at Q4 with \~4k context, the cheapest multi-GPU option we track is **4× Radeon RX 7600 XT** ($1,316 for the cards, est. 4–6 tok/s), the cheapest pair is **2× Intel Arc Pro B70 32GB** ($1,898, est. 9–12 tok/s); the cheapest single box that fits it is the **Framework Desktop (Ryzen AI Max+ 395) 128GB** ($3,449, est. 3–5 tok/s). ## Cheapest way to each memory size Up to four identical cards at list price (GPU cost only: a multi-GPU PC also needs a bigger PSU, board and case). Unified memory counts \~75% as usable. Board power is the cards' total from the maker's spec page. __Cheapest single card, cheapest multi-GPU setup and cheapest unified-memory machine for each memory size__ | Memory needed | One card | Two cards | Three or four cards | Mac / mini PC | | ----------------------------------- | ----------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------- | | [24 GB](https://whatcanirunai.vercel.app/best-llm-for/24gb-vram/) | [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$949 · 608 GB/s | 2× [Intel Arc B580](https://whatcanirunai.vercel.app/hardware/arc-b580/)$498 for the cards · 24 GB · 456 GB/s per card · — board power | no cheaper than two | [Mac mini M6 32GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-32/)$1,799 · 24 GB usable · 170 GB/s | | [32 GB](https://whatcanirunai.vercel.app/best-llm-for/32gb-vram/) | [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$949 · 608 GB/s | 2× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$658 for the cards · 32 GB · 288 GB/s per card · 380 W board power | no cheaper than two | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)$3,199 · 48 GB usable · 307 GB/s | | [48 GB](https://whatcanirunai.vercel.app/best-llm-for/48gb-vram/) | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)$6,799 · 960 GB/s | 2× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$1,898 for the cards · 64 GB · 608 GB/s per card · 460 W board power | 3× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$987 for the cards · 48 GB · 288 GB/s per card · 570 W board power | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)$3,199 · 48 GB usable · 307 GB/s | | [64 GB](https://whatcanirunai.vercel.app/best-llm-for/64gb-vram/) | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$8,565 · 1792 GB/s | 2× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$1,898 for the cards · 64 GB · 608 GB/s per card · 460 W board power | 4× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$1,316 for the cards · 64 GB · 288 GB/s per card · 760 W board power | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)$3,449 · 96 GB usable · 256 GB/s | | [96 GB](https://whatcanirunai.vercel.app/best-llm-for/96gb-vram/) | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$8,565 · 1792 GB/s | 2× [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)$13,598 for the cards · 96 GB · 960 GB/s per card · 600 W board power | 3× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$2,847 for the cards · 96 GB · 608 GB/s per card · 690 W board power | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)$3,449 · 96 GB usable · 256 GB/s | | [128 GB](https://whatcanirunai.vercel.app/best-llm-for/128gb-vram/) | no single card | 2× [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$17,130 for the cards · 192 GB · 1792 GB/s per card · 1,200 W board power | 4× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$3,796 for the cards · 128 GB · 608 GB/s per card · 920 W board power | [Mac Studio M5 Ultra 256GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-256/)$9,499 · 192 GB usable · 1200 GB/s | ## Cheapest setup for big models The cheapest option of each kind that fits the model with room for \~4k context, with estimated decode speed ([how we estimate speed](https://whatcanirunai.vercel.app/method/#speed); multi-GPU uses one card's bandwidth; MoE models show an upper bound only). __Cheapest single GPU, multi-GPU setup and unified-memory machine that runs each model, with estimated tokens per second__ | Model and quant | One card | Two cards | Three or four cards | Mac / mini PC | | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | | [Qwen3.8-27B](https://whatcanirunai.vercel.app/models/qwen3.8-27b/)Q8\_0 · needs \~34.9 GB | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)$6,799 · 20–28 tok/s | 2× [Radeon RX 7900 XT](https://whatcanirunai.vercel.app/hardware/rx-7900-xt/)$1,798 for the cards · 16–23 tok/s · 630 W | 4× [Intel Arc B570](https://whatcanirunai.vercel.app/hardware/arc-b570/)$876 for the cards · 8–11 tok/s · 600 W | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)$3,199 · 5–8 tok/s | | [Qwen3-32B](https://whatcanirunai.vercel.app/models/qwen3-32b/)Q8\_0 · needs \~41.0 GB | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)$6,799 · 17–23 tok/s | 2× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$1,898 for the cards · 10–15 tok/s · 460 W | 3× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$987 for the cards · 5–7 tok/s · 570 W | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)$3,199 · 4–7 tok/s | | [Gemma 4 31B](https://whatcanirunai.vercel.app/models/gemma-4-31b-it/)Q8\_0 · needs \~39.2 GB | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)$6,799 · 17–25 tok/s | 2× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$1,898 for the cards · 11–16 tok/s · 460 W | 3× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$987 for the cards · 5–7 tok/s · 570 W | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)$3,199 · 4–7 tok/s | | [Llama-3.3-70B-Instruct](https://whatcanirunai.vercel.app/models/llama-3.3-70b-instruct/)Q4\_K\_M · needs \~50.2 GB | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$8,565 · 25–36 tok/s | 2× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$1,898 for the cards · 9–12 tok/s · 460 W | 4× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$1,316 for the cards · 4–6 tok/s · 760 W | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)$3,449 · 3–5 tok/s | | [Qwen2.5-72B-Instruct](https://whatcanirunai.vercel.app/models/qwen2.5-72b-instruct/)Q4\_K\_M · needs \~51.7 GB | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$8,565 · 24–35 tok/s | 2× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$1,898 for the cards · 8–12 tok/s · 460 W | 4× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)$1,316 for the cards · 4–6 tok/s · 760 W | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)$3,449 · 3–5 tok/s | | [gpt-oss-120b](https://whatcanirunai.vercel.app/models/gpt-oss-120b/)Q4\_K\_M · needs \~82.5 GB | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$8,565 · up to 493 tok/s | 2× [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)$13,598 for the cards · up to 264 tok/s · 600 W | 3× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$2,847 for the cards · up to 167 tok/s · 690 W | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)$3,449 · up to 66 tok/s | | [Llama-3.3-70B-Instruct](https://whatcanirunai.vercel.app/models/llama-3.3-70b-instruct/)Q8\_0 · needs \~87.2 GB | no single card | 2× [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$17,130 for the cards · 14–20 tok/s · 1,200 W | 4× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)$3,796 for the cards · 5–7 tok/s · 920 W | [Mac Studio M5 Ultra 256GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-256/)$9,499 · 7–13 tok/s | | [Qwen3.8-Flash-Next](https://whatcanirunai.vercel.app/models/qwen3.8-flash-next/)Q4\_K\_M · needs \~126.5 GB | no single card | 2× [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)$17,130 for the cards · 10–14 tok/s · 1,200 W | no cheaper than two | [Mac Studio M5 Ultra 256GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-256/)$9,499 · 5–9 tok/s | ## Every card at 1×, 2×, 3× and 4× Total VRAM and list-price cost for each card count, combined board power for two cards, and the biggest model we track that fits at Q4 with two and with four cards. __VRAM and cost of 1 to 4 identical cards, with the biggest model that fits__ | GPU | 1× | 2× | 3× | 4× | Board power (2×) | Biggest at Q4 (2×) | Biggest at Q4 (4×) | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------- | ------------- | ------------- | ------------- | ---------------- | ------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------ | | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)96 GB · 1792 GB/s · $8,565 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 96 GB$8,565 | 192 GB$17,130 | 288 GB$25,695 | 384 GB$34,260 | 1,200 W | [Intern-S1](https://whatcanirunai.vercel.app/models/intern-s1/)\~168.8 GB | [Qwen3-Coder-480B-A35B-Instruct](https://whatcanirunai.vercel.app/models/qwen3-coder-480b-a35b-instruct/)\~335.8 GB | | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)48 GB · 960 GB/s · $6,799 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 48 GB$6,799 | 96 GB$13,598 | 144 GB$20,397 | 192 GB$27,196 | 600 W | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB | [Intern-S1](https://whatcanirunai.vercel.app/models/intern-s1/)\~168.8 GB | | [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)32 GB · 608 GB/s · $949 list price (MSRP), [source](https://www.igorslab.de/en/intel-launches-the-arc-pro-b70-with-32-gb-of-vram-for-949-big-battlemage-is-coming-first-for-ai-and-workstations/), 2026-10-29 | 32 GB$949 | 64 GB$1,898 | 96 GB$2,847 | 128 GB$3,796 | 460 W | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | [Mixtral-8x22B-Instruct-v0.1](https://whatcanirunai.vercel.app/models/mixtral-8x22b-instruct-v0.1/)\~99.0 GB | | [GeForce RTX 5090](https://whatcanirunai.vercel.app/hardware/rtx-5090/)32 GB · 1792 GB/s · $1,999 list price (MSRP), [source](https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/), 2026-10-22 | 32 GB$1,999 | 64 GB$3,998 | 96 GB$5,997 | 128 GB$7,996 | 1,150 W | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | [Mixtral-8x22B-Instruct-v0.1](https://whatcanirunai.vercel.app/models/mixtral-8x22b-instruct-v0.1/)\~99.0 GB | | [Radeon RX 7900 XTX](https://whatcanirunai.vercel.app/hardware/rx-7900-xtx/)24 GB · 960 GB/s · $999 list price (MSRP), [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries), 2026-10-22 | 24 GB$999 | 48 GB$1,998 | 72 GB$2,997 | 96 GB$3,996 | 710 W | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB | | [GeForce RTX 3090](https://whatcanirunai.vercel.app/hardware/rtx-3090/)24 GB · 936 GB/s · $1,499 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 24 GB$1,499 | 48 GB$2,998 | 72 GB$4,497 | 96 GB$5,996 | 700 W | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB | | [GeForce RTX 4090](https://whatcanirunai.vercel.app/hardware/rtx-4090/)24 GB · 1008 GB/s · $1,599 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 24 GB$1,599 | 48 GB$3,198 | 72 GB$4,797 | 96 GB$6,396 | 900 W | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB | | [GeForce RTX 3090 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3090-ti/)24 GB · 1008 GB/s · $1,999 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 24 GB$1,999 | 48 GB$3,998 | 72 GB$5,997 | 96 GB$7,996 | 900 W | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB | | [Radeon RX 7900 XT](https://whatcanirunai.vercel.app/hardware/rx-7900-xt/)20 GB · 800 GB/s · $899 list price (MSRP), [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries), 2026-10-22 | 20 GB$899 | 40 GB$1,798 | 60 GB$2,697 | 80 GB$3,596 | 630 W | [Llama-3\_3-Nemotron-Super-49B-v1\_5](https://whatcanirunai.vercel.app/models/llama-3-3-nemotron-super-49b-v1-5/)\~35.8 GB | [Qwen3-Next-80B-A3B-Instruct](https://whatcanirunai.vercel.app/models/qwen3-next-80b-a3b-instruct/)\~57.7 GB | | [NVIDIA RTX 4000 Ada 20GB](https://whatcanirunai.vercel.app/hardware/rtx-4000-ada/)20 GB · 360 GB/s · $1,250 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 20 GB$1,250 | 40 GB$2,500 | 60 GB$3,750 | 80 GB$5,000 | 260 W | [Llama-3\_3-Nemotron-Super-49B-v1\_5](https://whatcanirunai.vercel.app/models/llama-3-3-nemotron-super-49b-v1-5/)\~35.8 GB | [Qwen3-Next-80B-A3B-Instruct](https://whatcanirunai.vercel.app/models/qwen3-next-80b-a3b-instruct/)\~57.7 GB | | [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)16 GB · 288 GB/s · $329 list price (MSRP), [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries), 2026-10-22 | 16 GB$329 | 32 GB$658 | 48 GB$987 | 64 GB$1,316 | 380 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [Radeon RX 9060 XT 16GB](https://whatcanirunai.vercel.app/hardware/rx-9060-xt-16gb/)16 GB · 320 GB/s · $349 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$349 | 32 GB$698 | 48 GB$1,047 | 64 GB$1,396 | 320 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [GeForce RTX 5060 Ti 16GB](https://whatcanirunai.vercel.app/hardware/rtx-5060-ti-16gb/)16 GB · 448 GB/s · $429 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$429 | 32 GB$858 | 48 GB$1,287 | 64 GB$1,716 | 360 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [GeForce RTX 4060 Ti 16GB](https://whatcanirunai.vercel.app/hardware/rtx-4060-ti-16gb/)16 GB · 288 GB/s · $499 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 16 GB$499 | 32 GB$998 | 48 GB$1,497 | 64 GB$1,996 | 330 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [Radeon RX 7800 XT](https://whatcanirunai.vercel.app/hardware/rx-7800-xt/)16 GB · 624 GB/s · $499 list price (MSRP), [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries), 2026-10-22 | 16 GB$499 | 32 GB$998 | 48 GB$1,497 | 64 GB$1,996 | 526 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [Radeon RX 7900 GRE](https://whatcanirunai.vercel.app/hardware/rx-7900-gre/)16 GB · 576 GB/s · $549 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$549 | 32 GB$1,098 | 48 GB$1,647 | 64 GB$2,196 | 520 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [Radeon RX 9070](https://whatcanirunai.vercel.app/hardware/rx-9070/)16 GB · 640 GB/s · $549 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$549 | 32 GB$1,098 | 48 GB$1,647 | 64 GB$2,196 | 440 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [Radeon RX 9070 XT](https://whatcanirunai.vercel.app/hardware/rx-9070-xt/)16 GB · 640 GB/s · $599 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$599 | 32 GB$1,198 | 48 GB$1,797 | 64 GB$2,396 | 608 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [NVIDIA RTX 2000 Ada 16GB](https://whatcanirunai.vercel.app/hardware/rtx-2000-ada/)16 GB · 224 GB/s · $649 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$649 | 32 GB$1,298 | 48 GB$1,947 | 64 GB$2,596 | 140 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [GeForce RTX 5070 Ti](https://whatcanirunai.vercel.app/hardware/rtx-5070-ti/)16 GB · 896 GB/s · $749 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 16 GB$749 | 32 GB$1,498 | 48 GB$2,247 | 64 GB$2,996 | 600 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [GeForce RTX 4070 Ti Super](https://whatcanirunai.vercel.app/hardware/rtx-4070-ti-super/)16 GB · 672 GB/s · $799 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 16 GB$799 | 32 GB$1,598 | 48 GB$2,397 | 64 GB$3,196 | 570 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [GeForce RTX 4080 Super](https://whatcanirunai.vercel.app/hardware/rtx-4080-super/)16 GB · 736 GB/s · $999 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 16 GB$999 | 32 GB$1,998 | 48 GB$2,997 | 64 GB$3,996 | 640 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [GeForce RTX 5080](https://whatcanirunai.vercel.app/hardware/rtx-5080/)16 GB · 960 GB/s · $999 list price (MSRP), [source](https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5080/), 2026-10-22 | 16 GB$999 | 32 GB$1,998 | 48 GB$2,997 | 64 GB$3,996 | 720 W | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | [Hunyuan-A13B-Instruct](https://whatcanirunai.vercel.app/models/hunyuan-a13b-instruct/)\~57.0 GB | | [Intel Arc B580](https://whatcanirunai.vercel.app/hardware/arc-b580/)12 GB · 456 GB/s · $249 list price (MSRP), [source](https://en.wikipedia.org/wiki/Intel%5FArc#Battlemage), 2026-10-22 | 12 GB$249 | 24 GB$498 | 36 GB$747 | 48 GB$996 | — | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [GeForce RTX 3060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3060-12gb/)12 GB · 360 GB/s · $329 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 12 GB$329 | 24 GB$658 | 36 GB$987 | 48 GB$1,316 | 340 W | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [Radeon RX 7700 XT](https://whatcanirunai.vercel.app/hardware/rx-7700-xt/)12 GB · 432 GB/s · $449 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 12 GB$449 | 24 GB$898 | 36 GB$1,347 | 48 GB$1,796 | 490 W | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [GeForce RTX 5070](https://whatcanirunai.vercel.app/hardware/rtx-5070/)12 GB · 672 GB/s · $549 list price (MSRP), [source](https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5070-family/), 2026-10-22 | 12 GB$549 | 24 GB$1,098 | 36 GB$1,647 | 48 GB$2,196 | 500 W | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [GeForce RTX 4070](https://whatcanirunai.vercel.app/hardware/rtx-4070/)12 GB · 504 GB/s · $599 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 12 GB$599 | 24 GB$1,198 | 36 GB$1,797 | 48 GB$2,396 | 400 W | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [GeForce RTX 4070 Super](https://whatcanirunai.vercel.app/hardware/rtx-4070-super/)12 GB · 504 GB/s · $599 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 12 GB$599 | 24 GB$1,198 | 36 GB$1,797 | 48 GB$2,396 | 440 W | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [GeForce RTX 3080 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3080-ti/)12 GB · 912 GB/s · $1,199 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 12 GB$1,199 | 24 GB$2,398 | 36 GB$3,597 | 48 GB$4,796 | — | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB | | [Intel Arc B570](https://whatcanirunai.vercel.app/hardware/arc-b570/)10 GB · 380 GB/s · $219 list price (MSRP), [source](https://en.wikipedia.org/wiki/Intel%5FArc), 2026-10-28 | 10 GB$219 | 20 GB$438 | 30 GB$657 | 40 GB$876 | 300 W | [Voxtral-Small-24B-2507](https://whatcanirunai.vercel.app/models/voxtral-small-24b-2507/)\~17.9 GB | [Llama-3\_3-Nemotron-Super-49B-v1\_5](https://whatcanirunai.vercel.app/models/llama-3-3-nemotron-super-49b-v1-5/)\~35.8 GB | | [GeForce RTX 3080 10GB](https://whatcanirunai.vercel.app/hardware/rtx-3080-10gb/)10 GB · 760 GB/s · $699 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 10 GB$699 | 20 GB$1,398 | 30 GB$2,097 | 40 GB$2,796 | 640 W | [Voxtral-Small-24B-2507](https://whatcanirunai.vercel.app/models/voxtral-small-24b-2507/)\~17.9 GB | [Llama-3\_3-Nemotron-Super-49B-v1\_5](https://whatcanirunai.vercel.app/models/llama-3-3-nemotron-super-49b-v1-5/)\~35.8 GB | | [GeForce RTX 3050 8GB](https://whatcanirunai.vercel.app/hardware/rtx-3050-8gb/)8 GB · 224 GB/s · $249 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 8 GB$249 | 16 GB$498 | 24 GB$747 | 32 GB$996 | 260 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [Radeon RX 7600](https://whatcanirunai.vercel.app/hardware/rx-7600/)8 GB · 288 GB/s · $269 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 8 GB$269 | 16 GB$538 | 24 GB$807 | 32 GB$1,076 | 330 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [GeForce RTX 4060](https://whatcanirunai.vercel.app/hardware/rtx-4060/)8 GB · 272 GB/s · $299 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries), 2026-10-22 | 8 GB$299 | 16 GB$598 | 24 GB$897 | 32 GB$1,196 | 230 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [GeForce RTX 5060](https://whatcanirunai.vercel.app/hardware/rtx-5060/)8 GB · 448 GB/s · $299 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 8 GB$299 | 16 GB$598 | 24 GB$897 | 32 GB$1,196 | 290 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [Radeon RX 9060 XT 8GB](https://whatcanirunai.vercel.app/hardware/rx-9060-xt-8gb/)8 GB · 320 GB/s · $299 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 8 GB$299 | 16 GB$598 | 24 GB$897 | 32 GB$1,196 | 300 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [GeForce RTX 5060 Ti 8GB](https://whatcanirunai.vercel.app/hardware/rtx-5060-ti-8gb/)8 GB · 448 GB/s · $379 list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | 8 GB$379 | 16 GB$758 | 24 GB$1,137 | 32 GB$1,516 | 360 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [GeForce RTX 3060 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3060-ti/)8 GB · 448 GB/s · $399 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 8 GB$399 | 16 GB$798 | 24 GB$1,197 | 32 GB$1,596 | 400 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | | [GeForce RTX 3070](https://whatcanirunai.vercel.app/hardware/rtx-3070/)8 GB · 448 GB/s · $499 list price (MSRP), [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries), 2026-10-22 | 8 GB$499 | 16 GB$998 | 24 GB$1,497 | 32 GB$1,996 | 440 W | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB | ## Before you buy a second card * **Speed:** a layer split runs at about one card's speed. A single card with enough VRAM is simpler and as fast. * **Mixing cards:** you can mix sizes, but the slower card sets the pace for its layers. This page only shows identical cards. * **Power and cooling:** two RTX 5090s are 1,150 W of board power before the CPU. Check the PSU math in the [planner](https://whatcanirunai.vercel.app/planner/). * **Ready-made lists:** [dual used RTX 3090](https://whatcanirunai.vercel.app/builds/dual-used-rtx-3090-70b/), [dual RX 7900 XTX](https://whatcanirunai.vercel.app/builds/dual-rx-7900-xtx-48gb-build/) and [dual used RTX 3060](https://whatcanirunai.vercel.app/builds/dual-used-rtx-3060-24gb/) recipes, or [Mac Studio vs dual RTX 3090](https://whatcanirunai.vercel.app/builds/mac-studio-vs-dual-rtx-3090/). ## Quick answers Can two GPUs share VRAM to run a bigger LLM? Yes. llama.cpp-based apps (LM Studio, Ollama) and vLLM can split a model's layers across cards, so two 24 GB cards hold models up to about 48 GB. The memory adds up; the speed does not (see below). Each card also keeps some memory for its own buffers, so leave a little headroom. Do two GPUs make a local LLM run faster? Usually not for one chat. With the common layer split, the cards take turns: each processes its share of layers, so decode speed is about what one card would do if it had all the memory. Our estimates use one card's bandwidth for multi-GPU setups. Tensor parallelism (vLLM, ExLlama) can speed things up, but needs a fast link between the cards and we don't model it. Can dual RTX 3090s run a 70B model? At Q4\_K\_M with \~4k context our formula puts Llama-3.3-70B-Instruct at about 50.2 GB, so two 24 GB cards (48 GB) are just short. People do run 70B on two 3090s with smaller quants (Q3/IQ4) or less context. The cheapest multi-GPU setup we track that fits it at Q4 with \~4k context is 4× Radeon RX 7600 XT ($1,316 for the cards); the cheapest pair is 2× Intel Arc Pro B70 32GB ($1,898). What is the cheapest way to get 48 GB of VRAM? From list prices in our data: 3× Radeon RX 7600 XT for $987 (GPU cost only, 288 GB/s per card). The cheapest pair is 2× Intel Arc Pro B70 32GB for $1,898 (608 GB/s per card, so much faster). The cheapest unified-memory machine with at least 48 GB usable is the Mac mini M5 Pro 64GB at $3,199, a complete computer. Used cards can be cheaper or dearer; prices swing a lot. How many GPUs do I need for gpt-oss-120b? It needs about 82.5 GB at Q4\. The cheapest multi-GPU fit we track is 3× Intel Arc Pro B70 32GB ($2,847 for the cards). A single Framework Desktop (Ryzen AI Max+ 395) 128GB ($3,449) also holds it. What else does a multi-GPU PC need? A board with enough PCIe slots and spacing (x8/x8 or better helps, but layer split works over x4), a PSU sized for every card's board power plus the CPU, the right power connectors, and a case with room and airflow. Three or four cards usually need a workstation board, risers or an open frame. The planner checks PSU and case fit for 1–4 cards. [Best GPU for local LLMs](https://whatcanirunai.vercel.app/best-gpu-for-local-llm/) · [Best Mac](https://whatcanirunai.vercel.app/best-mac-for-local-llm/) · [NVIDIA vs AMD vs Mac](https://whatcanirunai.vercel.app/nvidia-vs-amd-vs-mac/) · [Tokens per second](https://whatcanirunai.vercel.app/tokens-per-second/) · [How much VRAM for 70B](https://whatcanirunai.vercel.app/how-much-vram/70b/) · [Build recipes](https://whatcanirunai.vercel.app/builds/) --- Agent guide: https://whatcanirunai.vercel.app/llms.txt · JSON API: https://whatcanirunai.vercel.app/developers/