https://1legion.com/blog/ Notes from 1Legion on NVIDIA GPU servers, clusters and data centers. en-us https://1legion.com/blog/70gb-of-weights-four-gpus-whats-left-for-the-kv-cache-at-80-96-141-and-288gb/ https://1legion.com/blog/70gb-of-weights-four-gpus-whats-left-for-the-kv-cache-at-80-96-141-and-288gb/ Tue, 20 Oct 2026 00:00:00 GMT GPUs Load a 70B model in FP8 on 80, 96, 141 and 270GB GPUs and see how much memory each leaves for the KV cache, with the sizing arithmetic worked out. https://1legion.com/blog/8-x-288gb-what-fits-on-a-b300-node/ https://1legion.com/blog/8-x-288gb-what-fits-on-a-b300-node/ Tue, 13 Oct 2026 00:00:00 GMT GPUs What fits on an NVIDIA B300 node: 270GB memory math for 405B and 70B model inference, KV cache sizing, bandwidth ceilings, and where a single GPU stops. https://1legion.com/blog/8-x-96gb-what-fits-on-an-rtx-pro-6000-blackwell-node/ https://1legion.com/blog/8-x-96gb-what-fits-on-an-rtx-pro-6000-blackwell-node/ Tue, 06 Oct 2026 00:00:00 GMT GPUs What fits on an RTX PRO 6000 Blackwell node: the 96GB memory math for 70B model inference and QLoRA fine tuning, KV cache sizing, and where the limit sits. https://1legion.com/blog/foundation-model-training-bottlenecks-memory-bandwidth-not-gpu-count/ https://1legion.com/blog/foundation-model-training-bottlenecks-memory-bandwidth-not-gpu-count/ Tue, 29 Sep 2026 00:00:00 GMT Infrastructure Foundation model training throughput depends on GPU memory capacity, memory bandwidth and interconnect, not GPU count. Here is the data behind that claim. https://1legion.com/blog/whats-driving-rtx-pro-6000-blackwell-prices-higher-in-2026/ https://1legion.com/blog/whats-driving-rtx-pro-6000-blackwell-prices-higher-in-2026/ Wed, 16 Sep 2026 00:00:00 GMT GPUs Why the RTX PRO 6000 Blackwell costs more to buy in 2026, how the GDDR7 memory shortage drives it, and what it means for buy versus rent decisions. https://1legion.com/blog/why-series-a-ai-companies-are-moving-off-aws-gpu-before-they-have-to/ https://1legion.com/blog/why-series-a-ai-companies-are-moving-off-aws-gpu-before-they-have-to/ Wed, 09 Sep 2026 00:00:00 GMT Infrastructure Why Series A AI companies move off AWS GPU before a forcing function decides for them: price spreads, quota walls, no renewal checkpoint and runway math. https://1legion.com/blog/b300-vs-h200-how-to-decide-which-node-your-workload-actually-needs/ https://1legion.com/blog/b300-vs-h200-how-to-decide-which-node-your-workload-actually-needs/ Wed, 02 Sep 2026 00:00:00 GMT GPUs B300 vs H200: a technical decision framework for dedicated bare metal GPU nodes, covering memory capacity, FP4 throughput, power, cooling and workload fit. https://1legion.com/blog/what-a-36-to-60-month-bare-metal-contract-actually-looks-like-and-why-it-changes-how-you-plan/ https://1legion.com/blog/what-a-36-to-60-month-bare-metal-contract-actually-looks-like-and-why-it-changes-how-you-plan/ Tue, 25 Aug 2026 00:00:00 GMT Infrastructure What a 36 to 60 month bare metal GPU contract locks in, what stays flexible, and how it changes capacity planning and budget predictability for AI teams. https://1legion.com/blog/10-questions-that-reveal-the-real-cost-of-your-cloud-gpu-setup/ https://1legion.com/blog/10-questions-that-reveal-the-real-cost-of-your-cloud-gpu-setup/ Wed, 19 Aug 2026 00:00:00 GMT Guides Ten questions to run against your cloud GPU bills, covering utilization, egress, overruns, storage sprawl, support fees and what you really pay per job. https://1legion.com/blog/b300-for-ai-model-labs-what-sustained-training-looks-like-on-dedicated-bare-metal/ https://1legion.com/blog/b300-for-ai-model-labs-what-sustained-training-looks-like-on-dedicated-bare-metal/ Wed, 12 Aug 2026 00:00:00 GMT GPUs What sustained pretraining needs from infrastructure, what dedicated NVIDIA B300 bare metal delivers for AI model labs, and how to configure clusters for it. https://1legion.com/blog/why-long-duration-training-runs-fail-on-shared-cloud-and-what-to-do-about-it/ https://1legion.com/blog/why-long-duration-training-runs-fail-on-shared-cloud-and-what-to-do-about-it/ Wed, 05 Aug 2026 00:00:00 GMT Infrastructure Preemption, noisy neighbors, storage contention and checkpoint egress are built into shared cloud. What long training runs face, and what bare metal fixes. https://1legion.com/blog/10-questions-to-ask-your-gpu-infrastructure-provider-before-you-sign-anything/ https://1legion.com/blog/10-questions-to-ask-your-gpu-infrastructure-provider-before-you-sign-anything/ Wed, 29 Jul 2026 00:00:00 GMT Guides Ten questions to ask a GPU infrastructure provider before you sign: tenancy, performance guarantees, egress, storage, scaling, drivers, SLAs and exit terms. https://1legion.com/blog/why-the-b300-is-the-gpu-serious-ai-teams-are-planning-around-right-now/ https://1legion.com/blog/why-the-b300-is-the-gpu-serious-ai-teams-are-planning-around-right-now/ Thu, 16 Jul 2026 00:00:00 GMT GPUs The NVIDIA HGX B300 is shipping now. What Blackwell Ultra delivers technically, how it compares with H100, and why dedicated bare metal is the way to run it. https://1legion.com/blog/b300-for-production-inference-what-changes-when-you-move-beyond-training/ https://1legion.com/blog/b300-for-production-inference-what-changes-when-you-move-beyond-training/ Thu, 09 Jul 2026 00:00:00 GMT GPUs What the NVIDIA B300's 270GB of memory, FP4 compute and latency profile mean for teams serving LLMs at scale on dedicated bare metal infrastructure. https://1legion.com/blog/isv-certification-and-gpu-infrastructure-why-it-matters-more-than-you-think-for-media-workloads/ https://1legion.com/blog/isv-certification-and-gpu-infrastructure-why-it-matters-more-than-you-think-for-media-workloads/ Thu, 02 Jul 2026 00:00:00 GMT Infrastructure What ISV certification means for GPU infrastructure, and why it matters for broadcast, VFX and media SaaS pipelines running professional software at scale. https://1legion.com/blog/from-h100-to-b300-what-changes-for-teams-already-running-serious-ai-workloads/ https://1legion.com/blog/from-h100-to-b300-what-changes-for-teams-already-running-serious-ai-workloads/ Thu, 18 Jun 2026 00:00:00 GMT GPUs Running H100 in production and evaluating B300? Where each GPU makes sense, what the B300 resolves, and why the infrastructure layer matters for both. https://1legion.com/blog/b300-on-bare-metal-what-the-architecture-means-for-llm-training-at-scale/ https://1legion.com/blog/b300-on-bare-metal-what-the-architecture-means-for-llm-training-at-scale/ Wed, 03 Jun 2026 00:00:00 GMT GPUs The NVIDIA B300 brings 3.4x the memory of the H100 and native FP4 compute. What that means for LLM training at scale, and why bare metal makes the difference. https://1legion.com/blog/comfyui-ai-video-on-bare-metal-gpu/ https://1legion.com/blog/comfyui-ai-video-on-bare-metal-gpu/ Mon, 01 Jun 2026 00:00:00 GMT Infrastructure Run ComfyUI, AI video generation and image pipelines on dedicated bare metal GPU servers, with full VRAM per job, no shared tenants and no egress fees. https://1legion.com/blog/llm-inference-on-bare-metal-gpu/ https://1legion.com/blog/llm-inference-on-bare-metal-gpu/ Mon, 01 Jun 2026 00:00:00 GMT Infrastructure Run LLM inference on dedicated bare metal GPU servers with full VRAM and bandwidth, FP4 and FP8 support, no shared tenants and no virtualization layer. https://1legion.com/blog/1legion-and-classover-form-ousia-compute-what-a-50m-dedicated-gpu-joint-venture-actually-means/ https://1legion.com/blog/1legion-and-classover-form-ousia-compute-what-a-50m-dedicated-gpu-joint-venture-actually-means/ Tue, 26 May 2026 00:00:00 GMT News In May 2026, 1Legion and Classover formed Ousia Compute, a dedicated GPU joint venture. What was announced, and what a venture of this kind involves. https://1legion.com/blog/why-llm-inference-breaks-on-shared-gpu-infrastructure-and-what-to-do-about-it/ https://1legion.com/blog/why-llm-inference-breaks-on-shared-gpu-infrastructure-and-what-to-do-about-it/ Tue, 19 May 2026 00:00:00 GMT Infrastructure Shared GPU cloud adds latency variance you cannot control. Why dedicated bare metal suits production LLM inference, and when it makes sense to switch. https://1legion.com/blog/bare-metal-vs-shared-gpu-infrastructure-what-actually-matters-in-production/ https://1legion.com/blog/bare-metal-vs-shared-gpu-infrastructure-what-actually-matters-in-production/ Tue, 24 Mar 2026 00:00:00 GMT Infrastructure Bare metal vs shared GPU infrastructure: the real impact on performance, consistency and scalability for production AI, and when dedicated hardware wins. https://1legion.com/blog/the-real-cost-of-running-rtx-5090-at-scale-what-ai-teams-need-to-know/ https://1legion.com/blog/the-real-cost-of-running-rtx-5090-at-scale-what-ai-teams-need-to-know/ Thu, 05 Mar 2026 00:00:00 GMT GPUs The real cost of running RTX 5090 GPUs at scale: how data movement, idle compute and infrastructure design shape AI training economics for generative AI. https://1legion.com/blog/nvidia-rtx-pro-6000-blackwell-servers-are-now-available-on-1legion-in-canada/ https://1legion.com/blog/nvidia-rtx-pro-6000-blackwell-servers-are-now-available-on-1legion-in-canada/ Tue, 20 Jan 2026 00:00:00 GMT News 1Legion adds NVIDIA RTX PRO 6000 Blackwell servers in Montreal, Canada: dedicated bare metal for AI inference, VFX and media workloads on 36 to 60 month terms. https://1legion.com/blog/when-you-should-choose-an-h100-a-practical-technical-guide-for-ai-vfx-and-media-workloads/ https://1legion.com/blog/when-you-should-choose-an-h100-a-practical-technical-guide-for-ai-vfx-and-media-workloads/ Wed, 03 Dec 2025 00:00:00 GMT Guides When to move from consumer GPUs like the RTX 5090 to H100-class data center GPUs for frontier AI, multimodal models, distributed training and media pipelines. https://1legion.com/blog/moving-your-transcoding-solution-to-the-cloud-with-the-nvidia-rtx-4000-ada/ https://1legion.com/blog/moving-your-transcoding-solution-to-the-cloud-with-the-nvidia-rtx-4000-ada/ Tue, 25 Nov 2025 00:00:00 GMT GPUs Why the NVIDIA RTX 4000 Ada suits cloud transcoding, with hardware AV1, high stream density and stable performance for live and VOD media workflows. https://1legion.com/blog/the-real-cost-of-training-ai-models-in-the-cloud/ https://1legion.com/blog/the-real-cost-of-training-ai-models-in-the-cloud/ Mon, 10 Nov 2025 00:00:00 GMT Infrastructure Why AI training in the cloud costs more than the hourly rate suggests: storage, egress, idle GPU time, total cost of ownership and the bare metal alternative.