{"title":"Placas de vídeo e aceleradoras de IA","description":"\u003cp\u003ePlaca de vídeo ou aceleradora de IA é o jeito de dar mais poder de computação a uma máquina que você já tem. Esta categoria atende três tipos de comprador. O data center encontra placas PCIe como a H200 NVL com 141 GB de HBM3, a H100 80 GB e a Atlas 300I Duo, feita para inferência de vídeo. Estúdios e laboratórios de IA ficam com placas de vídeo para estação de trabalho e para games, da RTX PRO 6000 de 96 GB à RTX 5090, e a computação de borda tem uma aceleradora M.2 de 160 TOPS que cabe num mini PC.\u003c\/p\u003e","products":[{"product_id":"atlas-300i-duo-ai-inference-accelerator-card-96gb-280-tops","title":"Placa aceleradora de inferência de IA Atlas 300I Duo — 96GB, 280 TOPS INT8, PCIe Gen4","description":"\u003cp\u003eA Atlas 300I Duo é uma aceleradora de inferência PCIe Gen4 para servidores que rodam visão computacional com IA, análise de vídeo e cargas de mídia. Cada placa traz 96GB de memória própria e entrega 280 TOPS de processamento em INT8, então um único slot dá conta de lotes grandes e de muitos fluxos de inferência simultâneos sem recorrer à memória do host. Núcleos de processamento de IA, uma CPU integrada e mecanismos dedicados de codec de vídeo por hardware ficam na mesma placa, o que permite decodificar, analisar e codificar vídeo numa só passada, sem devolver os quadros ao host.\u003c\/p\u003e\u003cp\u003eEla atende equipes que constroem plataformas de vídeo inteligente, análise de varejo e de trânsito, moderação de conteúdo e outros serviços de inferência de alto volume, em que a densidade do rack e o orçamento de energia fazem diferença. Como codec, CPU e núcleos de IA são integrados, você concentra numa só placa PCIe o que seriam equipamentos separados de transcodificação e de inferência.\u003c\/p\u003e\u003cul\u003e\u003cli\u003e96GB de memória na placa para modelos grandes e alta concorrência\u003c\/li\u003e\u003cli\u003eDesempenho de inferência de 280 TOPS em INT8\u003c\/li\u003e\u003cli\u003eInterface de host PCIe Gen4\u003c\/li\u003e\u003cli\u003eNúcleos de IA, CPU na placa e mecanismos de codec de vídeo por hardware integrados\u003c\/li\u003e\u003c\/ul\u003e\u003cp\u003eGarantia de 1 ano do fabricante.\u003c\/p\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67947233149027,"sku":"BEP-0865","price":1700.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/4efd6160-a8f6-479d-9bec-95014bd5f9f2.png?v=1794867860"},{"product_id":"rtx-pro-6000-blackwell-max-q-workstation-gpu-96gb-gddr7-ecc","title":"RTX PRO 6000 Blackwell Max-Q Workstation GPU — 96GB GDDR7 ECC","description":"\u003cp\u003eThe RTX PRO 6000 Blackwell Max-Q is a workstation-grade graphics card built for studios and labs that run heavy visual and compute jobs — 3D rendering, CAD and simulation, video production, and local AI inference and fine-tuning. Its 96GB of GDDR7 ECC memory holds large scenes, high-resolution textures and sizable models in VRAM, so artists and engineers spend less time managing memory limits and more time iterating.\u003c\/p\u003e\u003cp\u003eThe Max-Q variant targets a lower power envelope, which suits multi-GPU workstations and rack builds where thermal headroom and acoustics matter. Error-correcting memory guards against silent data corruption during long batch renders and overnight training runs.\u003c\/p\u003e\u003cul\u003e\n\u003cli\u003e96GB GDDR7 ECC memory on a 512-bit interface\u003c\/li\u003e\n\u003cli\u003eBase core clock around 1395 MHz\u003c\/li\u003e\n\u003cli\u003eDirectX 12 support for modern visualization pipelines\u003c\/li\u003e\n\u003cli\u003ePCI Express x16 add-in card for desktop and workstation chassis\u003c\/li\u003e\n\u003cli\u003eFan-cooled Max-Q design for tighter power and thermal budgets\u003c\/li\u003e\n\u003c\/ul\u003e\u003cp\u003eBacked by a manufacturer warranty. Confirm exact power connectors and PCIe generation for your motherboard before ordering.\u003c\/p\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67963144437859,"sku":"BEP-0883","price":11000.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/204e01ac-07de-488e-afbc-d92ee23585b8.jpg?v=1795092647"},{"product_id":"h200-nvl-ai-accelerator-141gb-hbm3-pcie-5-0-hopper-gpu","title":"H200 NVL AI Accelerator — 141GB HBM3, PCIe 5.0 Hopper GPU","description":"\u003cp\u003eThe H200 NVL is a PCIe data center accelerator built for AI training, inference, and high-performance computing. Its 141GB of HBM3 memory on a 5120-bit interface gives large language models and scientific workloads room to run without constant data shuffling, so a single card can hold models that would otherwise spill across several older GPUs.\u003c\/p\u003e\u003cp\u003eFabricated on a 4nm process with a 1410MHz core clock, it keeps throughput high while holding power draw down — the kind of efficiency that matters when a rack runs around the clock. An active fan assembly keeps the card stable under sustained load, and the PCIe 5.0 interface pairs it with current server platforms.\u003c\/p\u003e\u003cul\u003e\n\u003cli\u003e141GB HBM3 memory, 5120-bit interface for memory-bound AI and HPC jobs\u003c\/li\u003e\n\u003cli\u003e4nm Hopper-class GPU, 1410MHz core clock\u003c\/li\u003e\n\u003cli\u003ePCIe 5.0 slot card with DisplayPort output\u003c\/li\u003e\n\u003cli\u003eActive fan cooling for continuous rack operation\u003c\/li\u003e\n\u003c\/ul\u003e\u003cp\u003eIt suits teams building inference clusters, model-training rigs, and HPC nodes where memory capacity per card is the limiting factor. Each unit ships factory-sealed.\u003c\/p\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67963147026531,"sku":"BEP-0884","price":27000.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/1b76d60d-0d0e-46b6-a5f1-00a6439af605.jpg?v=1795092810"},{"product_id":"nvidia-geforce-rtx-5090-founders-edition-32gb-gddr7","title":"NVIDIA GeForce RTX 5090 Founders Edition 32GB GDDR7 Graphics Card","description":"\u003cp\u003eThe GeForce RTX 5090 Founders Edition is NVIDIA's flagship desktop graphics card, built for venues that run AI workloads, 3D rendering and high-resolution video alongside gaming. It pairs 21,760 CUDA cores with 32 GB of GDDR7 memory on a 512-bit bus, giving 1,792 GB\/s of bandwidth for the kind of heavy, high-resolution scenes that stall smaller cards.\u003c\/p\u003e\u003cul\u003e\n\u003cli\u003e\n\u003cstrong\u003e32 GB GDDR7, 512-bit bus\u003c\/strong\u003e — 1,792 GB\/s bandwidth for large models and 8K assets.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e4th-gen RT and 5th-gen Tensor Cores\u003c\/strong\u003e — 318 RT-TFLOPS and 3,352 AI TOPS, with DLSS 4 and Frame Generation.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eCompact 2-slot build\u003c\/strong\u003e — 304 mm long with the Double Flow Through cooler for quiet operation under full load.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eModern connectivity\u003c\/strong\u003e — PCIe 5.0 x16, three DisplayPort 2.1b and one HDMI 2.1b, up to 4 displays at 8K@165Hz.\u003c\/li\u003e\n\u003c\/ul\u003e\u003cp\u003eIt fits gaming cafes and LAN centers that sell high-refresh play, and back-office AI workstations where the 32 GB frame buffer lets a single card train and run larger models. TDP is 575 W, so plan on a 1,000 W PSU and a 16-pin (12VHPWR) power connector.\u003c\/p\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67963150434403,"sku":"BEP-0886","price":4370.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/4dddcf39-f1a5-443c-bc60-e3bbf3a1ad2d.png?v=1795092985"},{"product_id":"m-2-ai-accelerator-card-160-tops-npu-for-edge-llm-inference","title":"M.2 AI Accelerator Card — 160 TOPS NPU for Edge LLM Inference","description":"\u003cp\u003eThis M.2 AI accelerator card is a plug-in NPU module that brings 160 TOPS of inference throughput to a compatible host, so AI workloads run on the device instead of a remote server. In a bar, restaurant, or retail floor that already runs a mini PC or embedded controller for signage, vision analytics, or a local assistant, the card does the heavy math on-premises — no cloud round-trips, no per-call API bills, and no customer data leaving the building.\u003c\/p\u003e\u003cp\u003eThe compact M.2 form factor drops into AI PCs and embedded hardware designs that expose a suitable slot, and the module draws about 10 W, so it fits thermally constrained builds without a dedicated power feed. It is built around a local LLM use case: running a quantized model for a self-order kiosk, a menu assistant, or back-office automation close to the data it works on.\u003c\/p\u003e\u003cul\u003e\n\u003cli\u003e160 TOPS of AI acceleration for edge inference and local LLM workloads\u003c\/li\u003e\n\u003cli\u003eM.2 module that integrates into compatible AI PCs and embedded boards\u003c\/li\u003e\n\u003cli\u003eRoughly 10 W draw — suited to compact, fanless or low-power builds\u003c\/li\u003e\n\u003cli\u003eKeeps inference on-device, so prompts and data stay on your hardware\u003c\/li\u003e\n\u003c\/ul\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67963258273891,"sku":"BEP-0893","price":1630.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/485745e4-c882-4ef5-b39d-2ee261fbe53b.png?v=1795095963"},{"product_id":"rtx-pro-5500-blackwell-workstation-gpu-84gb-gddr7-ecc-pcie","title":"RTX PRO 5500 Blackwell Workstation GPU — 84GB GDDR7 ECC, PCIe 5.0","description":"\u003cp\u003eThe RTX PRO 5500 Blackwell Workstation GPU is a professional accelerator card built for studios, render farms and AI teams that push large datasets through a single workstation. Its 84GB of GDDR7 ECC memory holds big models, high-resolution scenes and complex visual workloads in VRAM, while error-correcting code guards against silent memory faults during long jobs — the kind of reliability professional pipelines depend on.\u003c\/p\u003e\u003cp\u003eThe 384-bit memory interface and 1740MHz core clock back demanding compute-oriented workflows: 3D rendering, simulation, machine-learning training and inference. A factory water-cooling loop manages heat during sustained operation, so the card holds its clocks through overnight renders and multi-hour training runs rather than throttling. The PCIe 5.0 x16 host interface gives it full bandwidth to the rest of the system.\u003c\/p\u003e\u003cul\u003e\n\u003cli\u003e84GB GDDR7 ECC memory for large models and datasets\u003c\/li\u003e\n\u003cli\u003e384-bit memory interface, 1740MHz core clock\u003c\/li\u003e\n\u003cli\u003eWater-cooled for sustained full-load operation\u003c\/li\u003e\n\u003cli\u003ePCIe 5.0 x16 host interface, 600W power\u003c\/li\u003e\n\u003c\/ul\u003e\u003cp\u003eBacked by a 1-year manufacturer warranty.\u003c\/p\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67963486994531,"sku":"BEP-0897","price":23500.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/d3115338-b94a-46bd-8062-5f37f8592df1.png?v=1795097802"},{"product_id":"nvidia-h100-80gb-pcie-ai-accelerator-gpu","title":"NVIDIA H100 80GB PCIe AI Accelerator GPU","description":"\u003cp\u003eThe NVIDIA H100 80GB PCIe is a data-center accelerator for AI training, deep-learning inference and high-throughput compute workloads. It drops into a standard PCIe 4.0 x16 slot, which lets venues add serious model-training capacity to an existing server or workstation chassis without committing to an SXM board and its specialized backplane.\u003c\/p\u003e\u003cp\u003eWith 80 GB of on-board memory, the card holds large models and batch sizes in GPU memory, so fewer transfers to system RAM slow a training run. It fits the back-office and facility setups that now run local AI — a restaurant group training demand-forecasting models, a retail chain scoring camera analytics on-premises, or a production site running vision inference on the line.\u003c\/p\u003e\u003cul\u003e\n\u003cli\u003ePCIe 4.0 x16 interface — installs in standard server and workstation slots\u003c\/li\u003e\n\u003cli\u003e80 GB GPU memory for large models and bigger batch sizes\u003c\/li\u003e\n\u003cli\u003eDirectX 12 Ultimate and active fan cooling\u003c\/li\u003e\n\u003cli\u003eBacked by a 1-year manufacturer warranty\u003c\/li\u003e\n\u003c\/ul\u003e","brand":"BarEquip.Pro","offers":[{"title":"Default Title","offer_id":67963493646435,"sku":"BEP-0898","price":27225.0,"currency_code":"USD","in_stock":true}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0643\/6726\/3843\/files\/547ab326-1e6e-4b3f-b349-a5b4ccb5b0f8.jpg?v=1795097940"}],"url":"https:\/\/barequip.pro\/pt\/collections\/gpus.oembed","provider":"BarEquip.Pro","version":"1.0","type":"link"}