# Best used GPU for local LLMs: RTX 3090, P40, MI50, A6000 (2026) > Used GPUs for running LLMs locally: 40 cards incl. RTX 3090, Tesla P40, V100, Instinct MI50 and RTX A6000 by VRAM, bandwidth and est. tok/s, with the catches and a new-card price ceiling. Source: https://whatcanirunai.vercel.app/best-used-gpu-for-local-llm/ # Best used GPU for local LLMs Used and previous-generation cards are the cheapest way to get a lot of VRAM. Here are the 40 used-market GPUs we track (gaming, workstation and retired datacenter cards like the Tesla P40, V100 and Instinct MI50), ranked by VRAM, with bandwidth, estimated tokens per second and the catches that listings don't mention. Used GPUs tracked 40 VRAM range 8–48 GB Bandwidth range 224–1024 GB/s Short answer: VRAM decides which models fit, bandwidth decides speed. A used **RTX 3090** (24 GB, 936 GB/s) is the safe default. Datacenter cards (P40, V100, MI50) give 24–32 GB cheaply but need cooling and software work. **About prices:** we don't track used street prices, so none are shown. Each card lists its launch MSRP (what it cost new, not what it sells for now), and the [price ceiling](#ceiling) table shows the cheapest new card with the same VRAM at today's list price. ## Our short list 12 GB entry card ### [GeForce RTX 3060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3060-12gb/) VRAM 12 GB Bandwidth 360 GB/s Run at Q4 350 / 656 Qwen3-8B at Q4: est. 43–62 tok/s. Biggest at Q4: [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/). The usual cheap way into 12 GB on NVIDIA. Launch MSRP $329 (not a used price, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries)) Pay less than a new Intel Arc B580 ($249 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=GeForce%20RTX%203060%2012GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=GeForce%20RTX%203060%2012GB) 16 GB on AMD ### [Radeon RX 6800](https://whatcanirunai.vercel.app/hardware/rx-6800/) VRAM 16 GB Bandwidth 512 GB/s Run at Q4 388 / 656 Qwen3-14B at Q4: est. 34–49 tok/s. Biggest at Q4: [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/). RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. Launch MSRP $579 (not a used price, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits)) Pay less than a new Radeon RX 7600 XT ($329 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=Radeon%20RX%206800&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=Radeon%20RX%206800) The classic 24 GB pick ### [GeForce RTX 3090](https://whatcanirunai.vercel.app/hardware/rtx-3090/) VRAM 24 GB Bandwidth 936 GB/s Run at Q4 423 / 656 Qwen3.8-27B at Q4: est. 33–47 tok/s. Biggest at Q4: [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/). Memory chips sit on both sides of the board, so check memory temperature under load. 3-slot reference cooler. Launch MSRP $1,499 (not a used price, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries)) Pay less than a new Intel Arc Pro B70 32GB ($949 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=GeForce%20RTX%203090&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=GeForce%20RTX%203090) 24 GB on the cheap, with work ### [NVIDIA Tesla P40 24GB](https://whatcanirunai.vercel.app/hardware/tesla-p40/) VRAM 24 GB Bandwidth 346 GB/s Run at Q4 423 / 656 Qwen3.8-27B at Q4: est. 12–17 tok/s. Biggest at Q4: [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/). Passive cooler (needs server airflow or a fan shroud), no display output, CPU-style 8-pin (EPS) power input. Pascal has very slow FP16 math, so prompt processing is slow; llama.cpp runs it. No list price: sold through OEMs. Pay less than a new Intel Arc Pro B70 32GB ($949 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=NVIDIA%20Tesla%20P40%2024GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=NVIDIA%20Tesla%20P40%2024GB) 32 GB with lots of bandwidth, with work ### [AMD Radeon Instinct MI50 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi50-32gb/) VRAM 32 GB Bandwidth 1024 GB/s Run at Q4 504 / 656 Qwen3-32B at Q4: est. 31–44 tok/s. Biggest at Q4: [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/). Passive cooler (needs a fan shroud). gfx906 is not supported by current ROCm releases (last fully supported: ROCm 5.7); community ROCm builds and llama.cpp Vulkan are the usual routes. No list price: sold through OEMs. Pay less than a new Intel Arc Pro B70 32GB ($949 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=AMD%20Radeon%20Instinct%20MI50%2032GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=AMD%20Radeon%20Instinct%20MI50%2032GB) 32 GB on NVIDIA, with work ### [NVIDIA Tesla V100 32GB (PCIe)](https://whatcanirunai.vercel.app/hardware/tesla-v100-32gb/) VRAM 32 GB Bandwidth 900 GB/s Run at Q4 504 / 656 Qwen3-32B at Q4: est. 27–39 tok/s. Biggest at Q4: [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/). Passive cooler (needs server airflow or a fan shroud), no display output. Volta: no FlashAttention 2 support; llama.cpp runs it. No list price: sold through OEMs. Pay less than a new Intel Arc Pro B70 32GB ($949 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=NVIDIA%20Tesla%20V100%2032GB%20%28PCIe%29&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=NVIDIA%20Tesla%20V100%2032GB%20%28PCIe%29) 48 GB in one card ### [NVIDIA RTX A6000](https://whatcanirunai.vercel.app/hardware/rtx-a6000/) VRAM 48 GB Bandwidth 768 GB/s Run at Q4 511 / 656 Qwen3-32B at Q4: est. 23–33 tok/s. Biggest at Q4: [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/). 48 GB in a 2-slot blower card at 300 W, so two fit in a normal desktop. Launch MSRP $4,649 (not a used price, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits)) Pay less than a new NVIDIA RTX Pro 6000 Blackwell 96GB ($8,565 list) for this to make sense. [Amazon](https://www.amazon.com/s?k=NVIDIA%20RTX%20A6000&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=NVIDIA%20RTX%20A6000) We may earn a commission when you buy through these links, at no extra cost to you. ## Price ceiling: the new card with the same VRAM A used card is only worth it below the price of a new card that does the same job. This table shows the cheapest new, current GPU we track with at least that much VRAM, at list price (MSRP). New cards also come with a warranty and current driver support. __For each VRAM size, the used GPUs we track and the cheapest new GPU with at least that VRAM at list price__ | VRAM | Used options | Cheapest new card with that VRAM | New list price | | -------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 12–15 GB | [GeForce RTX 3080 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3080-ti/), [GeForce RTX 3080 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3080-12gb/), [GeForce RTX 4070 Ti](https://whatcanirunai.vercel.app/hardware/rtx-4070-ti/), [Radeon RX 6750 XT](https://whatcanirunai.vercel.app/hardware/rx-6750-xt/), [Radeon RX 6700 XT](https://whatcanirunai.vercel.app/hardware/rx-6700-xt/), [GeForce RTX 3060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3060-12gb/), [GeForce RTX 2060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-2060-12gb/) | [Intel Arc B580](https://whatcanirunai.vercel.app/hardware/arc-b580/)12 GB · 456 GB/s | $249list price (MSRP), [source](https://en.wikipedia.org/wiki/Intel%5FArc#Battlemage), 2026-10-22 | | 16–20 GB | [GeForce RTX 4080](https://whatcanirunai.vercel.app/hardware/rtx-4080/), [Radeon RX 6950 XT](https://whatcanirunai.vercel.app/hardware/rx-6950-xt/), [Intel Arc A770 16GB](https://whatcanirunai.vercel.app/hardware/arc-a770-16gb/), [Radeon RX 6800](https://whatcanirunai.vercel.app/hardware/rx-6800/), [Radeon RX 6800 XT](https://whatcanirunai.vercel.app/hardware/rx-6800-xt/), [Radeon RX 6900 XT](https://whatcanirunai.vercel.app/hardware/rx-6900-xt/), [NVIDIA RTX A4000](https://whatcanirunai.vercel.app/hardware/rtx-a4000/) | [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)16 GB · 288 GB/s | $329list price (MSRP), [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries), 2026-10-22 | | 24 GB | [GeForce RTX 3090 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3090-ti/), [GeForce RTX 4090](https://whatcanirunai.vercel.app/hardware/rtx-4090/), [Radeon RX 7900 XTX](https://whatcanirunai.vercel.app/hardware/rx-7900-xtx/), [GeForce RTX 3090](https://whatcanirunai.vercel.app/hardware/rtx-3090/), [NVIDIA RTX A5000](https://whatcanirunai.vercel.app/hardware/rtx-a5000/), [NVIDIA Titan RTX](https://whatcanirunai.vercel.app/hardware/titan-rtx/), [NVIDIA Tesla P40 24GB](https://whatcanirunai.vercel.app/hardware/tesla-p40/) | [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)32 GB · 608 GB/s | $949list 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 | [AMD Radeon Instinct MI50 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi50-32gb/), [AMD Radeon Instinct MI60 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi60/), [NVIDIA Tesla V100 32GB (PCIe)](https://whatcanirunai.vercel.app/hardware/tesla-v100-32gb/) | [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)32 GB · 608 GB/s | $949list 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 | | 48+ GB | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/), [NVIDIA RTX A6000](https://whatcanirunai.vercel.app/hardware/rtx-a6000/) | [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)96 GB · 1792 GB/s | $8,565list price (MSRP), [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits), 2026-10-28 | ## Every used GPU, by VRAM Sorted by VRAM, then bandwidth. Board power is the maker's reference spec. Launch MSRP is for reference only ([how we calculate](https://whatcanirunai.vercel.app/method/)). __Used-market GPUs ranked by VRAM with bandwidth, board power, launch MSRP and the biggest model each runs at Q4__ | GPU | VRAM | Bandwidth | Board power | Launch MSRP | Biggest model at Q4 | Models that fit | | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- | --------- | ----------- | ------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------- | --------------- | | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)Workstation | 48 GB | 960 GB/s | 300 W | $6,799launch MSRP 2022, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB at Q4 | 511 / 656 | | [NVIDIA RTX A6000](https://whatcanirunai.vercel.app/hardware/rtx-a6000/)Workstation · Ampere · 48 GB in a 2-slot blower card at 300 W, so two fit in a normal desktop. | 48 GB | 768 GB/s | 300 W | $4,649launch MSRP 2020, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB at Q4 | 511 / 656 | | [AMD Radeon Instinct MI50 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi50-32gb/)Datacenter · GCN 5.1 (gfx906) · Passive cooler (needs a fan shroud). gfx906 is not supported by current ROCm releases (last fully supported: ROCm 5.7); community ROCm builds and llama.cpp Vulkan are the usual routes. | 32 GB | 1024 GB/s | 300 W | No list price (sold through OEMs) | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB at Q4 | 504 / 656 | | [AMD Radeon Instinct MI60 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi60/)Datacenter · GCN 5.1 (gfx906) · Passive cooler (needs a fan shroud). gfx906 is not supported by current ROCm releases (last fully supported: ROCm 5.7); community ROCm builds and llama.cpp Vulkan are the usual routes. | 32 GB | 1024 GB/s | 300 W | No list price (sold through OEMs) | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB at Q4 | 504 / 656 | | [NVIDIA Tesla V100 32GB (PCIe)](https://whatcanirunai.vercel.app/hardware/tesla-v100-32gb/)Datacenter · Volta · Passive cooler (needs server airflow or a fan shroud), no display output. Volta: no FlashAttention 2 support; llama.cpp runs it. | 32 GB | 900 GB/s | 250 W | No list price (sold through OEMs) | [IQuest-Coder-V1-40B-Loop-Instruct](https://whatcanirunai.vercel.app/models/iquest-coder-v1-40b-loop-instruct/)\~28.7 GB at Q4 | 504 / 656 | | [GeForce RTX 3090 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3090-ti/)Gaming · Ampere · Highest-bandwidth 24 GB card of its generation; 450 W board power. | 24 GB | 1008 GB/s | 450 W | $1,999launch MSRP 2022, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [GeForce RTX 4090](https://whatcanirunai.vercel.app/hardware/rtx-4090/)Gaming · Ada | 24 GB | 1008 GB/s | 450 W | $1,599launch MSRP 2022, [source](https://en.wikipedia.org/wiki/GeForce%5F40%5Fseries) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [Radeon RX 7900 XTX](https://whatcanirunai.vercel.app/hardware/rx-7900-xtx/)Gaming · RDNA 3 | 24 GB | 960 GB/s | 355 W | $999launch MSRP 2022, [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [GeForce RTX 3090](https://whatcanirunai.vercel.app/hardware/rtx-3090/)Gaming · Ampere · Memory chips sit on both sides of the board, so check memory temperature under load. 3-slot reference cooler. | 24 GB | 936 GB/s | 350 W | $1,499launch MSRP 2020, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [NVIDIA RTX A5000](https://whatcanirunai.vercel.app/hardware/rtx-a5000/)Workstation · Ampere · 24 GB in a 2-slot blower card at 230 W. | 24 GB | 768 GB/s | 230 W | $2,250launch MSRP 2021, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [NVIDIA Titan RTX](https://whatcanirunai.vercel.app/hardware/titan-rtx/)Workstation · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 24 GB | 672 GB/s | 280 W | $2,499launch MSRP 2018, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [NVIDIA Tesla P40 24GB](https://whatcanirunai.vercel.app/hardware/tesla-p40/)Datacenter · Pascal · Passive cooler (needs server airflow or a fan shroud), no display output, CPU-style 8-pin (EPS) power input. Pascal has very slow FP16 math, so prompt processing is slow; llama.cpp runs it. | 24 GB | 346 GB/s | 250 W | No list price (sold through OEMs) | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [Radeon RX 7900 XT](https://whatcanirunai.vercel.app/hardware/rx-7900-xt/)Gaming · RDNA 3 | 20 GB | 800 GB/s | 315 W | $899launch MSRP 2022, [source](https://en.wikipedia.org/wiki/Radeon%5FRX%5F7000%5Fseries) | [Voxtral-Small-24B-2507](https://whatcanirunai.vercel.app/models/voxtral-small-24b-2507/)\~17.9 GB at Q4 | 404 / 656 | | [GeForce RTX 4080](https://whatcanirunai.vercel.app/hardware/rtx-4080/)Gaming · Ada | 16 GB | 717 GB/s | 320 W | $1,199launch MSRP 2022, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [Radeon RX 6950 XT](https://whatcanirunai.vercel.app/hardware/rx-6950-xt/)Gaming · RDNA 2 · RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. | 16 GB | 576 GB/s | 335 W | $1,099launch MSRP 2022, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [Intel Arc A770 16GB](https://whatcanirunai.vercel.app/hardware/arc-a770-16gb/)Gaming · Alchemist · 16 GB; check your runtime supports Intel GPUs (llama.cpp SYCL or Vulkan). | 16 GB | 560 GB/s | 225 W | $349launch MSRP 2022, [source](https://en.wikipedia.org/wiki/Intel%5FArc) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [Radeon RX 6800](https://whatcanirunai.vercel.app/hardware/rx-6800/)Gaming · RDNA 2 · RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. | 16 GB | 512 GB/s | 250 W | $579launch MSRP 2020, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [Radeon RX 6800 XT](https://whatcanirunai.vercel.app/hardware/rx-6800-xt/)Gaming · RDNA 2 · RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. | 16 GB | 512 GB/s | 300 W | $649launch MSRP 2020, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [Radeon RX 6900 XT](https://whatcanirunai.vercel.app/hardware/rx-6900-xt/)Gaming · RDNA 2 · RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. | 16 GB | 512 GB/s | 300 W | $999launch MSRP 2020, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [NVIDIA RTX A4000](https://whatcanirunai.vercel.app/hardware/rtx-a4000/)Workstation · Ampere · 16 GB single-slot card at 140 W. | 16 GB | 448 GB/s | 140 W | $1,000launch MSRP 2021, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [Ling-Coder-lite](https://whatcanirunai.vercel.app/models/ling-coder-lite/)\~12.7 GB at Q4 | 388 / 656 | | [GeForce RTX 3080 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3080-ti/)Gaming · Ampere | 12 GB | 912 GB/s | — | $1,199launch MSRP 2021, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [GeForce RTX 3080 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3080-12gb/)Gaming · Ampere | 12 GB | 912 GB/s | 350 W | $799launch MSRP 2022, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [GeForce RTX 4070 Ti](https://whatcanirunai.vercel.app/hardware/rtx-4070-ti/)Gaming · Ada | 12 GB | 504 GB/s | 285 W | $799launch MSRP 2023, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [Radeon RX 6750 XT](https://whatcanirunai.vercel.app/hardware/rx-6750-xt/)Gaming · RDNA 2 · RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. | 12 GB | 432 GB/s | 250 W | $549launch MSRP 2022, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [Radeon RX 6700 XT](https://whatcanirunai.vercel.app/hardware/rx-6700-xt/)Gaming · RDNA 2 · RDNA 2: not on AMD’s official ROCm support list for Radeon cards, though many users run it; llama.cpp Vulkan works without ROCm. | 12 GB | 384 GB/s | 230 W | $479launch MSRP 2021, [source](https://en.wikipedia.org/wiki/List%5Fof%5FAMD%5Fgraphics%5Fprocessing%5Funits) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [GeForce RTX 3060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3060-12gb/)Gaming · Ampere · The usual cheap way into 12 GB on NVIDIA. | 12 GB | 360 GB/s | 170 W | $329launch MSRP 2021, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [GeForce RTX 2060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-2060-12gb/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 12 GB | 336 GB/s | 184 W | No list price (sold through OEMs) | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [GeForce RTX 2080 Ti](https://whatcanirunai.vercel.app/hardware/rtx-2080-ti/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 11 GB | 616 GB/s | 250 W | $999launch MSRP 2018, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [Mistral-Nemo-Instruct-2407](https://whatcanirunai.vercel.app/models/mistral-nemo-instruct-2407/)\~9.5 GB at Q4 | 349 / 656 | | [GeForce RTX 3080 10GB](https://whatcanirunai.vercel.app/hardware/rtx-3080-10gb/)Gaming · Ampere | 10 GB | 760 GB/s | 320 W | $699launch MSRP 2020, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [HyperCLOVAX-SEED-Omni-8B](https://whatcanirunai.vercel.app/models/hyperclovax-seed-omni-8b/)\~8.5 GB at Q4 | 340 / 656 | | [GeForce RTX 3070 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3070-ti/)Gaming · Ampere | 8 GB | 608 GB/s | 290 W | $599launch MSRP 2021, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [Intel Arc A750](https://whatcanirunai.vercel.app/hardware/arc-a750/)Gaming · Alchemist | 8 GB | 512 GB/s | 225 W | $289launch MSRP 2022, [source](https://en.wikipedia.org/wiki/Intel%5FArc) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 2080 Super](https://whatcanirunai.vercel.app/hardware/rtx-2080-super/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 8 GB | 496 GB/s | 250 W | $699launch MSRP 2019, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 3060 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3060-ti/)Gaming · Ampere | 8 GB | 448 GB/s | 200 W | $399launch MSRP 2020, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 3070](https://whatcanirunai.vercel.app/hardware/rtx-3070/)Gaming · Ampere | 8 GB | 448 GB/s | 220 W | $499launch MSRP 2020, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 2060 Super](https://whatcanirunai.vercel.app/hardware/rtx-2060-super/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 8 GB | 448 GB/s | 175 W | $399launch MSRP 2019, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 2070](https://whatcanirunai.vercel.app/hardware/rtx-2070/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 8 GB | 448 GB/s | 175 W | $499launch MSRP 2018, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 2070 Super](https://whatcanirunai.vercel.app/hardware/rtx-2070-super/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 8 GB | 448 GB/s | 215 W | $499launch MSRP 2019, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 2080](https://whatcanirunai.vercel.app/hardware/rtx-2080/)Gaming · Turing · Turing: no FlashAttention 2 support; llama.cpp runs it. | 8 GB | 448 GB/s | 215 W | $699launch MSRP 2018, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 4060 Ti 8GB](https://whatcanirunai.vercel.app/hardware/rtx-4060-ti-8gb/)Gaming · Ada | 8 GB | 288 GB/s | 160 W | $399launch MSRP 2023, [source](https://en.wikipedia.org/wiki/List%5Fof%5FNvidia%5Fgraphics%5Fprocessing%5Funits) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | | [GeForce RTX 3050 8GB](https://whatcanirunai.vercel.app/hardware/rtx-3050-8gb/)Gaming · Ampere | 8 GB | 224 GB/s | 130 W | $249launch MSRP 2022, [source](https://en.wikipedia.org/wiki/GeForce%5F30%5Fseries) | [NVIDIA-Nemotron-Nano-9B-v2](https://whatcanirunai.vercel.app/models/nvidia-nemotron-nano-9b-v2/)\~7.2 GB at Q4 | 315 / 656 | ## Which popular models run, and how fast Cards with 12 GB or more. Estimated decode speed at Q4 in tokens per second ([how we estimate speed](https://whatcanirunai.vercel.app/method/#speed)); MoE models show an upper bound only. Older architectures may land at the low end of the range, and the P40's prompt processing is much slower than its decode speed suggests. __Estimated fit and decode speed at Q4 for popular models on each used GPU__ | Used GPU | [Qwen3-8B](https://whatcanirunai.vercel.app/models/qwen3-8b/)\~6.7 GB | [Qwen3-14B](https://whatcanirunai.vercel.app/models/qwen3-14b/)\~11.3 GB | [gpt-oss-20b](https://whatcanirunai.vercel.app/models/gpt-oss-20b/)\~15.6 GB | [Qwen3.8-27B](https://whatcanirunai.vercel.app/models/qwen3.8-27b/)\~20.4 GB | [Qwen3-30B-A3B](https://whatcanirunai.vercel.app/models/qwen3-30b-a3b/)\~22.3 GB | [Qwen3-32B](https://whatcanirunai.vercel.app/models/qwen3-32b/)\~23.8 GB | | ------------------------------------------------------------------------------- | ------------------------------------- | ---------------------------------------- | -------------------------------------------- | -------------------------------------------- | ------------------------------------------------ | ---------------------------------------- | | [NVIDIA RTX 6000 Ada 48GB](https://whatcanirunai.vercel.app/hardware/rtx-6000-ada/)48 GB · 960 GB/s | Fits116–164 tok/s | Fits64–91 tok/s | Fitsup to 374 tok/s | Fits34–48 tok/s | Fitsup to 449 tok/s | Fits29–41 tok/s | | [NVIDIA RTX A6000](https://whatcanirunai.vercel.app/hardware/rtx-a6000/)48 GB · 768 GB/s | Fits93–131 tok/s | Fits51–73 tok/s | Fitsup to 299 tok/s | Fits27–39 tok/s | Fitsup to 359 tok/s | Fits23–33 tok/s | | [AMD Radeon Instinct MI50 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi50-32gb/)32 GB · 1024 GB/s | Fits124–175 tok/s | Fits69–97 tok/s | Fitsup to 399 tok/s | Fits36–52 tok/s | Fitsup to 479 tok/s | Fits31–44 tok/s | | [AMD Radeon Instinct MI60 32GB](https://whatcanirunai.vercel.app/hardware/instinct-mi60/)32 GB · 1024 GB/s | Fits124–175 tok/s | Fits69–97 tok/s | Fitsup to 399 tok/s | Fits36–52 tok/s | Fitsup to 479 tok/s | Fits31–44 tok/s | | [NVIDIA Tesla V100 32GB (PCIe)](https://whatcanirunai.vercel.app/hardware/tesla-v100-32gb/)32 GB · 900 GB/s | Fits109–154 tok/s | Fits60–85 tok/s | Fitsup to 351 tok/s | Fits32–45 tok/s | Fitsup to 421 tok/s | Fits27–39 tok/s | | [GeForce RTX 3090 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3090-ti/)24 GB · 1008 GB/s | Fits122–173 tok/s | Fits68–96 tok/s | Fitsup to 393 tok/s | Fits36–51 tok/s | Tightup to 471 tok/s | Tight30–43 tok/s | | [GeForce RTX 4090](https://whatcanirunai.vercel.app/hardware/rtx-4090/)24 GB · 1008 GB/s | Fits122–173 tok/s | Fits68–96 tok/s | Fitsup to 393 tok/s | Fits36–51 tok/s | Tightup to 471 tok/s | Tight30–43 tok/s | | [Radeon RX 7900 XTX](https://whatcanirunai.vercel.app/hardware/rx-7900-xtx/)24 GB · 960 GB/s | Fits116–164 tok/s | Fits64–91 tok/s | Fitsup to 374 tok/s | Fits34–48 tok/s | Tightup to 449 tok/s | Tight29–41 tok/s | | [GeForce RTX 3090](https://whatcanirunai.vercel.app/hardware/rtx-3090/)24 GB · 936 GB/s | Fits113–160 tok/s | Fits63–89 tok/s | Fitsup to 365 tok/s | Fits33–47 tok/s | Tightup to 437 tok/s | Tight28–40 tok/s | | [NVIDIA RTX A5000](https://whatcanirunai.vercel.app/hardware/rtx-a5000/)24 GB · 768 GB/s | Fits93–131 tok/s | Fits51–73 tok/s | Fitsup to 299 tok/s | Fits27–39 tok/s | Tightup to 359 tok/s | Tight23–33 tok/s | | [NVIDIA Titan RTX](https://whatcanirunai.vercel.app/hardware/titan-rtx/)24 GB · 672 GB/s | Fits81–115 tok/s | Fits45–64 tok/s | Fitsup to 262 tok/s | Fits24–34 tok/s | Tightup to 314 tok/s | Tight20–29 tok/s | | [NVIDIA Tesla P40 24GB](https://whatcanirunai.vercel.app/hardware/tesla-p40/)24 GB · 346 GB/s | Fits42–59 tok/s | Fits23–33 tok/s | Fitsup to 135 tok/s | Fits12–17 tok/s | Tightup to 162 tok/s | Tight10–15 tok/s | | [Radeon RX 7900 XT](https://whatcanirunai.vercel.app/hardware/rx-7900-xt/)20 GB · 800 GB/s | Fits97–137 tok/s | Fits54–76 tok/s | Fitsup to 312 tok/s | Won’t fit | Won’t fit | Won’t fit | | [GeForce RTX 4080](https://whatcanirunai.vercel.app/hardware/rtx-4080/)16 GB · 717 GB/s | Fits87–123 tok/s | Fits48–68 tok/s | Tightup to 279 tok/s | Won’t fit | Won’t fit | Won’t fit | | [Radeon RX 6950 XT](https://whatcanirunai.vercel.app/hardware/rx-6950-xt/)16 GB · 576 GB/s | Fits70–99 tok/s | Fits39–55 tok/s | Tightup to 224 tok/s | Won’t fit | Won’t fit | Won’t fit | | [Intel Arc A770 16GB](https://whatcanirunai.vercel.app/hardware/arc-a770-16gb/)16 GB · 560 GB/s | Fits68–96 tok/s | Fits38–53 tok/s | Tightup to 218 tok/s | Won’t fit | Won’t fit | Won’t fit | | [Radeon RX 6800](https://whatcanirunai.vercel.app/hardware/rx-6800/)16 GB · 512 GB/s | Fits62–88 tok/s | Fits34–49 tok/s | Tightup to 199 tok/s | Won’t fit | Won’t fit | Won’t fit | | [Radeon RX 6800 XT](https://whatcanirunai.vercel.app/hardware/rx-6800-xt/)16 GB · 512 GB/s | Fits62–88 tok/s | Fits34–49 tok/s | Tightup to 199 tok/s | Won’t fit | Won’t fit | Won’t fit | | [Radeon RX 6900 XT](https://whatcanirunai.vercel.app/hardware/rx-6900-xt/)16 GB · 512 GB/s | Fits62–88 tok/s | Fits34–49 tok/s | Tightup to 199 tok/s | Won’t fit | Won’t fit | Won’t fit | | [NVIDIA RTX A4000](https://whatcanirunai.vercel.app/hardware/rtx-a4000/)16 GB · 448 GB/s | Fits54–77 tok/s | Fits30–43 tok/s | Tightup to 174 tok/s | Won’t fit | Won’t fit | Won’t fit | | [GeForce RTX 3080 Ti](https://whatcanirunai.vercel.app/hardware/rtx-3080-ti/)12 GB · 912 GB/s | Fits110–156 tok/s | Tight61–87 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [GeForce RTX 3080 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3080-12gb/)12 GB · 912 GB/s | Fits110–156 tok/s | Tight61–87 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [GeForce RTX 4070 Ti](https://whatcanirunai.vercel.app/hardware/rtx-4070-ti/)12 GB · 504 GB/s | Fits61–86 tok/s | Tight34–48 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [Radeon RX 6750 XT](https://whatcanirunai.vercel.app/hardware/rx-6750-xt/)12 GB · 432 GB/s | Fits52–74 tok/s | Tight29–41 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [Radeon RX 6700 XT](https://whatcanirunai.vercel.app/hardware/rx-6700-xt/)12 GB · 384 GB/s | Fits46–66 tok/s | Tight26–36 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [GeForce RTX 3060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-3060-12gb/)12 GB · 360 GB/s | Fits43–62 tok/s | Tight24–34 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [GeForce RTX 2060 12GB](https://whatcanirunai.vercel.app/hardware/rtx-2060-12gb/)12 GB · 336 GB/s | Fits41–58 tok/s | Tight23–32 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | ## Before you buy used * **Datacenter cards** (Tesla P40, V100, Instinct MI50/MI60) have passive coolers built for server airflow: plan a fan shroud. Most have no display output, so you need another GPU or integrated graphics for the screen. The P40 takes a CPU-style 8-pin (EPS) power plug, so you may need an adapter. * **Software support:** FlashAttention 2 needs an Ampere or newer NVIDIA GPU ([source](https://github.com/Dao-AILab/flash-attention)); Pascal, Volta and Turing cards still run llama.cpp. AMD's gfx906 cards (MI50, MI60) are not supported by current ROCm releases ([AMD's GPU table](https://rocm.docs.amd.com/projects/install-on-linux/en/develop/reference/system-requirements.html); ROCm 5.7 was the last fully supported version). * **Condition:** ask for a stress-test run with memory temperatures, check the VRAM size in the listing, and prefer sellers with returns. * **Power and size:** check board power, power connectors and card length against your PSU and case in the [planner](https://whatcanirunai.vercel.app/planner/). * **Two cards:** two 24 GB cards give 48 GB. See [multi-GPU for local LLMs](https://whatcanirunai.vercel.app/multi-gpu-for-local-llm/) and the [dual used RTX 3090](https://whatcanirunai.vercel.app/builds/dual-used-rtx-3090-70b/) and [used RTX 3060](https://whatcanirunai.vercel.app/builds/12gb-used-rtx-3060-build/) build recipes. Head-to-head: [RTX 3090 vs RTX 4090](https://whatcanirunai.vercel.app/compare/rtx-3090-vs-rtx-4090/) · [RTX 3090 vs RX 7900 XTX](https://whatcanirunai.vercel.app/compare/rtx-3090-vs-rx-7900-xtx/) · [RTX 4070 Ti Super vs RTX 3090](https://whatcanirunai.vercel.app/compare/rtx-4070-ti-super-vs-rtx-3090/). Have a card we don't list? Enter its VRAM in the [VRAM calculator](https://whatcanirunai.vercel.app/vram-calculator/). ## Quick answers What is the best used GPU for local LLMs? For most people a used GeForce RTX 3090: 24 GB of VRAM and 936 GB/s, enough for 27B–32B-class models at Q4 (Qwen3.8-27B at an estimated 33–47 tok/s), with full CUDA support. A new card with 24 GB or more starts at $949 list (Intel Arc Pro B70 32GB), so the used card only makes sense below that. On a tighter budget, a used 12 GB RTX 3060 runs 8B–14B models. Is the Tesla P40 still worth it for local LLMs? It gives 24 GB for the price of an old datacenter card, but it is slow: 346 GB/s of bandwidth (about 37% of an RTX 3090), weak FP16 math so long prompts process slowly, a passive cooler that needs a fan shroud, no display output and a CPU-style 8-pin power plug. It works in llama.cpp. Est. 42–59 tok/s on Qwen3-8B at Q4. Is the AMD Instinct MI50 good for LLMs? On paper yes: 32 GB of HBM2 at 1024 GB/s. The catch is software: AMD's current ROCm releases no longer support its gfx906 chip (ROCm 5.7 was the last fully supported release), so you rely on community ROCm builds or llama.cpp's Vulkan backend. It is also passively cooled. Buy it only if you are comfortable tinkering. Used RTX A6000 or two RTX 3090s? Both give 48 GB. The NVIDIA RTX A6000 is one 2-slot, 300 W card with 768 GB/s; two 3090s are 700 W of board power across two 3-slot cards, but each has 936 GB/s, so they decode a bit faster. Neither fits Llama 3.3 70B at Q4 with room to spare (50.2 GB on our formula), so 48 GB is really for 32B models with long context or 70B at lower quants. New, the cheapest 48 GB+ card is the NVIDIA RTX Pro 6000 Blackwell 96GB at $8,565 list. Why don't you show used prices? We only publish prices we can source and date. Used prices change daily and vary by condition and seller, so we show each card's launch MSRP for reference and, for each VRAM size, the cheapest new card at list price. That is your ceiling: a used card is only a deal below it. Use the eBay button on each card to see current listings. What should I check before buying a used GPU? Ask for a photo of the card running a GPU stress test and its memory temperatures, check that the VRAM size matches the listing, look for bent fins, missing screws or fan noise, confirm your power supply has the right connectors and enough watts (see the planner), and measure your case. Prefer sellers with returns. [Buying guides](https://whatcanirunai.vercel.app/best-gpu-for/) · [Best GPU (new)](https://whatcanirunai.vercel.app/best-gpu-for-local-llm/) · [Best GPU by budget](https://whatcanirunai.vercel.app/best-gpu-under/) · [Multi-GPU](https://whatcanirunai.vercel.app/multi-gpu-for-local-llm/) · [GPU power](https://whatcanirunai.vercel.app/gpu-power-consumption/) · [Tokens per second](https://whatcanirunai.vercel.app/tokens-per-second/) --- Agent guide: https://whatcanirunai.vercel.app/llms.txt · JSON API: https://whatcanirunai.vercel.app/developers/