[{"data":1,"prerenderedAt":1310},["ShallowReactive",2],{"blog-post-en-\u002Fblog\u002Fpaiton-minimax-h3-image-radeon-ai-pro-r9700":3,"blog-posts-sidebar-en":724},{"id":4,"title":5,"body":6,"categories":705,"date":711,"description":712,"extension":713,"heading":5,"image":714,"meta":715,"navigation":716,"originalUrl":717,"path":718,"seo":719,"slug":720,"socialImage":714,"stem":721,"updated":722,"video":722,"__hash__":723},"blog\u002Fblog\u002Fpaiton-minimax-h3-image-radeon-ai-pro-r9700.md","MiniMax-H3-Image on a single R9700",{"type":7,"value":8,"toc":688},"minimark",[9,27,30,36,39,91,95,103,106,109,112,115,119,122,129,134,213,218,221,224,227,230,236,241,244,248,251,254,258,261,264,268,271,275,278,281,287,370,391,394,397,400,448,451,457,460,466,469,472,478,484,490,496,499,503,520,524,531,579,582,610,641,654,657,666,669,672,684],[10,11,12,13,20,21,26],"p",{},"MiniMax-H3-Image now runs on one AMD Radeon AI PRO R9700 (32 GB). A 2048 x 2048 image takes 6.9 s and a 4096 x 4096\nimage takes 33.4 s, with the card at its 300 W power limit. The FP8 weights are on\n",[14,15,19],"a",{"href":16,"rel":17},"https:\u002F\u002Fhuggingface.co\u002FEliovpAI\u002FMiniMax-H3-Image-FP8-Paiton-RDNA4",[18],"nofollow","Hugging Face"," and the container is in our\n",[14,22,25],{"href":23,"rel":24},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin",[18],"paiton-vllm-plugin"," repository.",[10,28,29],{},"The BF16 image checkpoint and the first 50 text-encoder layers needed for inference total about 120 GB of weights. This post explains what the\nmodel is, why it does not fit on an R9700, what we built so that it does, and how we checked that the images still\nmatch the original.",[10,31,32],{},[33,34,35],"em",{},"The concert poster above and the generated examples below come from a sandboxed R9700 test environment. Comparison images identify the BF16 reference and our FP8 output.",[10,37,38],{},"In this article",[40,41,42,49,55,61,67,73,79,85],"ul",{},[43,44,45],"li",{},[14,46,48],{"href":47},"#what-minimax-h3-image-is","What MiniMax-H3-Image is",[43,50,51],{},[14,52,54],{"href":53},"#why-it-does-not-fit-on-an-r9700","Why it does not fit",[43,56,57],{},[14,58,60],{"href":59},"#what-we-built","What we built",[43,62,63],{},[14,64,66],{"href":65},"#speed","Speed",[43,68,69],{},[14,70,72],{"href":71},"#quality","Quality",[43,74,75],{},[14,76,78],{"href":77},"#more-examples","More examples",[43,80,81],{},[14,82,84],{"href":83},"#limits-of-this-build","Limits and setup",[43,86,87],{},[14,88,90],{"href":89},"#license","License",[92,93,48],"h2",{"id":94},"what-minimax-h3-image-is",[10,96,97,102],{},[14,98,101],{"href":99,"rel":100},"https:\u002F\u002Fhuggingface.co\u002FMiniMaxAI\u002FMiniMax-H3-Image",[18],"MiniMax-H3-Image"," is MiniMax's open-weight model for text to image\nand image editing. Three things set it apart from the open image models we have run so far.",[10,104,105],{},"It makes large images directly. MiniMax gives a working range of 4 to 20 MP and says it does best around 16 MP, for\nexample 4096 x 4096 (1:1) or 5472 x 3072 (16:9), with no upscaler in the pipeline. MiniMax compares that with the 1 to\n4 MP that is common for open models today.",[10,107,108],{},"It needs only 6 sampling steps, with no classifier-free guidance. The checkpoint is distilled to six transformer passes\nper image, where many models run 20 to 50 steps and two passes per step when guidance is on.",[10,110,111],{},"One checkpoint does both generation and editing, with up to 16 reference images. It is also good at text inside\nimages: posters, signs, menus and handwritten notes come out spelled correctly in most of our tests.",[10,113,114],{},"The architecture is a 33.1B-parameter single-stream diffusion transformer with 50 blocks, hidden size 5376 and 56\nattention heads. The prompt and any reference images go through a Qwen3-VL text encoder; the reference images also go\nthrough the image VAE. Everything then becomes one packed token sequence. At 16 MP that sequence is about 16,400 tokens,\nand every block runs full attention over all of them.",[92,116,118],{"id":117},"why-it-does-not-fit-on-an-r9700","Why it does not fit on an R9700",[10,120,121],{},"The image checkpoint is BF16 and totals 68.65 GB, including its VAE. The first 50 layers of the 32B Qwen3-VL text encoder\nadd about 52 GB. The R9700 has 32 GB.",[10,123,124],{},[125,126],"img",{"alt":127,"src":128},"Bar chart. The BF16 image checkpoint including its VAE is 68.7 GB, plus approximately 52.0 GB for the first 50 text-encoder layers, 120.7 GB in total, far past the R9700's 32 GB. Our build peaks at 29.6 GB on the card: 20.3 GB of FP8 transformer weights and 9.3 GB of activations, VAE and text-encoder layers.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002Fmemory.webp",[10,130,131],{},[33,132,133],{},"Estimated BF16 working weights and observed Paiton peak usage are different quantities. Host RAM is separate from GPU memory.",[135,136,137,154],"table",{},[138,139,140],"thead",{},[141,142,143,147,151],"tr",{},[144,145,146],"th",{},"Memory on the GPU",[144,148,150],{"align":149},"right","BF16 working weights",[144,152,153],{"align":149},"Paiton peak, 16 MP request",[155,156,157,169,180,191,202],"tbody",{},[141,158,159,163,166],{},[160,161,162],"td",{},"Image checkpoint, including VAE",[160,164,165],{"align":149},"68.7 GB",[160,167,168],{"align":149},"See allocations below",[141,170,171,174,177],{},[160,172,173],{},"First 50 text-encoder layers",[160,175,176],{"align":149},"About 52.0 GB",[160,178,179],{"align":149},"Streamed from host RAM",[141,181,182,185,188],{},[160,183,184],{},"Resident FP8 transformer",[160,186,187],{"align":149},"Included above",[160,189,190],{"align":149},"20.3 GB",[141,192,193,196,199],{},[160,194,195],{},"Activations, VAE and resident text-encoder layers",[160,197,198],{"align":149},"Not counted in the weights total",[160,200,201],{"align":149},"9.3 GB",[141,203,204,207,210],{},[160,205,206],{},"Total shown",[160,208,209],{"align":149},"About 120.7 GB",[160,211,212],{"align":149},"29.6 GB",[10,214,215],{},[33,216,217],{},"The Paiton build also uses 24.4 GB of pinned host RAM for the text encoder. 27.6 GiB is approximately 29.6 GB.",[10,219,220],{},"Loading the BF16 weights onto the card stops before the first step. Offloading them to system RAM would move tens of\ngigabytes over PCIe for each of the six steps; we did not try that route. MiniMax lists SGLang, vLLM-Omni,\ndiffusers and ComfyUI among its local deployment options for the BF16 checkpoint.",[10,222,223],{},"Even plain FP8 is not enough on its own. An FP8 transformer and an FP8 text encoder together come to about 47 GB.",[92,225,60],{"id":226},"what-we-built",[10,228,229],{},"We wrote a native runtime for the whole pipeline: text encoder, transformer, VAE and an HTTP server that takes the same\nrequests as MiniMax's own examples. It is HIP and C++ on Paiton's kernels, with no PyTorch and no Triton, and the\ncontainer image is 2.27 GB.",[10,231,232],{},[125,233],{"alt":234,"src":235},"Diagram of one request. The prompt and up to three reference images go into the FP8 text encoder, whose layers stream from 24.4 GB of pinned host RAM. Its tokens go into the transformer, which runs 6 steps with 19.3B FP8 parameters on the card, precomputed AdaLN tables and INT8\u002FFP8 attention, one HIP graph per step. The latent goes to the INT8 VAE decoder, which writes the PNG.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002Fpipeline.webp",[237,238,240],"h3",{"id":239},"adaln-tables-instead-of-adaln-weights","AdaLN tables instead of AdaLN weights",[10,242,243],{},"Each transformer block has a modulation projection (AdaLN) that turns the timestep into scales and shifts for that\nblock. Across the 50 blocks these projections hold about 13B of the 33.1B parameters, and their only input is the\ntimestep. The distilled model always uses the same six timesteps. So we run the projections once, ahead of time, for\nthose six timesteps and store the outputs as small tables. The result is exact, and 13B parameters never reach the\nGPU. That leaves 19.3B parameters on the card.",[237,245,247],{"id":246},"fp8-weights-and-activations-rounded-with-gptq","FP8 weights and activations, rounded with GPTQ",[10,249,250],{},"The remaining projections run in FP8 E4M3: one scale per output channel for the weights and one per token for the\nactivations, so the matrix multiplies run on the R9700's FP8 units. The transformer takes 20.3 GB.",[10,252,253],{},"We tried several ways to choose the FP8 weights against BF16 reference activations. SmoothQuant and clipping by\nmean-squared error gave nothing for FP8 E4M3. GPTQ rounding cut the error of each matrix multiply by 19 to 31 % against\nplain rounding, and that is what we ship. A 4-bit weight variant ran at the same speed with twice the error, so we kept\nFP8.",[237,255,257],{"id":256},"a-text-encoder-that-streams-from-host-memory","A text encoder that streams from host memory",[10,259,260],{},"The text encoder's 50 layers are FP8 too, 24.4 GB in all. They live in pinned host memory and are copied to the card\nfor each request, while up to 12 layers stay on the card when there is room. The encoder runs once per image, so this\ncosts 0.7 to 0.9 s per request.",[10,262,263],{},"Running this close to the card's memory limit caused one problem. In early runs 3 of 54 denoising steps stalled\nbecause the GPU driver evicted our buffers near full VRAM. Capping the runtime at 27.5 GiB removed the stalls (0 of\n60 steps).",[237,265,267],{"id":266},"attention-and-the-vae","Attention and the VAE",[10,269,270],{},"Attention uses INT8 for Q and K and FP8 for P and V. In isolation the last transformer block was the most sensitive to\nthis, so we tested exact attention for single blocks. End to end, the images were no closer to BF16 with it, so every\nblock uses the low-precision kernel. The VAE decoder uses INT8 weights and decodes 256 px tiles with the same overlap\nand blending as the reference.",[237,272,274],{"id":273},"where-the-time-goes","Where the time goes",[10,276,277],{},"Each of the six steps is one HIP graph. At 16 MP a step takes about 5.2 s. In our profile the projection GEMMs took\nabout 3.0 s of a step, running at 65 to 68 % of the card's FP8 peak, and attention about 2.3 s. The text encoder adds\n0.7 to 0.9 s per request and the VAE decoder about 0.75 s at 16 MP. Peak memory is 27.6 GiB.",[92,279,66],{"id":280},"speed",[10,282,283],{},[125,284],{"alt":285,"src":286},"Bar chart of seconds per request on one R9700: text to image 2048 x 2048 in 6.9 s, 4096 x 4096 in 33.4 s; edits at 4096 x 4096 with one reference in 39.9 s, 4736 x 3552 with one reference in 50.1 s, 4096 x 4096 with three references in 88.5 s.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002Ftimes.webp",[135,288,289,305],{},[138,290,291],{},[141,292,293,296,299,302],{},[144,294,295],{},"Request",[144,297,298],{},"Output size",[144,300,301],{"align":149},"Reference images",[144,303,304],{"align":149},"Time",[155,306,307,321,333,346,358],{},[141,308,309,312,315,318],{},[160,310,311],{},"Text to image",[160,313,314],{},"2048 x 2048",[160,316,317],{"align":149},"0",[160,319,320],{"align":149},"6.9 s",[141,322,323,325,328,330],{},[160,324,311],{},[160,326,327],{},"4096 x 4096",[160,329,317],{"align":149},[160,331,332],{"align":149},"33.4 s",[141,334,335,338,340,343],{},[160,336,337],{},"Edit",[160,339,327],{},[160,341,342],{"align":149},"1",[160,344,345],{"align":149},"39.9 s",[141,347,348,350,353,355],{},[160,349,337],{},[160,351,352],{},"4736 x 3552",[160,354,342],{"align":149},[160,356,357],{"align":149},"50.1 s",[141,359,360,362,364,367],{},[160,361,337],{},[160,363,327],{},[160,365,366],{"align":149},"3",[160,368,369],{"align":149},"88.5 s",[10,371,372],{},[33,373,374,375,380,381,380,386,390],{},"One R9700, 300 W limit, six steps. ",[14,376,379],{"href":377,"rel":378},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002Fmain\u002Fmodels\u002FMiniMax-H3-Image\u002Fresults\u002Fcontainer-e2e-gptq-20261111.jsonl",[18],"Published measurements"," · ",[14,382,385],{"href":383,"rel":384},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002Fmain\u002Fmodels\u002FMiniMax-H3-Image\u002Fpaiton_h3i\u002Fserver.py#L92",[18],"Timing boundary",[14,387,389],{"href":388},"\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002Fresults.json","Chart data",".",[10,392,393],{},"Times cover the native model pipeline through the finished PNG, including output resizing, with a warm server and the card at 300 W. Request preprocessing, queueing and response transfer are excluded.\nImages below 4 MP are made on a 4 MP canvas and scaled down, as in MiniMax's pipeline, so a 1024 x 1024 request takes\n7.1 s. Each reference image adds tokens to the sequence, which is why edits take longer.",[92,395,72],{"id":396},"quality",[10,398,399],{},"We compared our build with the BF16 model in diffusers on 30 cases: 24 text-to-image prompts from 4 to 17 MP in several\naspect ratios, and 6 edits with 1 to 3 reference images. Both sides used the same token ids and the same starting\nnoise, so any difference comes from the numerics.",[135,401,402,415],{},[138,403,404],{},[141,405,406,409,412],{},[144,407,408],{},"Metric",[144,410,411],{"align":149},"BF16",[144,413,414],{"align":149},"Our FP8 build",[155,416,417,428,438],{},[141,418,419,422,425],{},[160,420,421],{},"CLIP-L text-image score, mean",[160,423,424],{"align":149},"32.31",[160,426,427],{"align":149},"32.18",[141,429,430,433,435],{},[160,431,432],{},"CLIP-L similarity to the BF16 image, mean \u002F lowest",[160,434],{"align":149},[160,436,437],{"align":149},"0.979 \u002F 0.855",[141,439,440,443,445],{},[160,441,442],{},"DINOv2 similarity to the BF16 image, mean \u002F lowest",[160,444],{"align":149},[160,446,447],{"align":149},"0.945 \u002F 0.699",[10,449,450],{},"The text-image score is within the spread between cases. Most of the differences are in layout: a 6-step sampler takes\na different path after small numeric changes, so some images are composed differently while showing the same content.\nText, faces and product shots rendered correctly in these tests.",[10,452,453],{},[125,454],{"alt":455,"src":456},"Side by side, a chalkboard menu from the BF16 model and from our FP8 build with the same prompt and noise. Both read \"TODAY'S SPECIALS\", Tomato soup 4.50, Grilled cheese 6.00, Caesar salad 7.50, Lemon tart 3.80, Flat white 3.20; only the small chalk drawings differ.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F05-compare-menu.jpg",[10,458,459],{},"Edits stay closest to BF16, with image similarity of 0.96 to 0.999.",[10,461,462],{},[125,463],{"alt":464,"src":465},"Three panels: the input storefront, the BF16 edit and our FP8 edit. The prompt asked to change the sign to \"PADARIA LUA\" and paint the door yellow. Both edits do this and are almost identical.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F06-compare-edit.jpg",[92,467,78],{"id":468},"more-examples",[10,470,471],{},"All made with the released weights at 4096 x 4096, about 34 s each, and shown here scaled to 2048 x 2048. We picked the\nbetter of two seeds for each prompt.",[10,473,474],{},[125,475],{"alt":476,"src":477},"Close-up portrait of an elderly fisherwoman with deep wrinkles, freckles, wind-blown grey hair and a mustard knitted sweater.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F07-portrait.jpg",[10,479,480],{},[125,481],{"alt":482,"src":483},"An old bookshop window at dusk with gold lettering \"OWL & LANTERN BOOKS\" and \"Rare · Used · New since 1952\", and hand-lettered cards in the window.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F08-bookshop.jpg",[10,485,486],{},[125,487],{"alt":488,"src":489},"A 3D render of five shapes placed as the prompt asked: a red cube on a blue sphere, a green cylinder to the left, a yellow cone behind and a purple pyramid in front on the right.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F09-shapes.jpg",[10,491,492],{},[125,493],{"alt":494,"src":495},"A newspaper front page on a café table with the masthead \"The Harbor Gazette\" and the headline \"CITY OPENS NEW HARBOR BRIDGE\".","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F10-newspaper.jpg",[10,497,498],{},"Not everything works. Small body text, such as the newspaper columns and some book spines, looks like writing from a\ndistance but is not real words. Close-up hands, a clock set to three o'clock and a horse riding on an astronaut's back\ncame out wrong in our tries.",[92,500,502],{"id":501},"limits-of-this-build","Limits of this build",[40,504,505,508,511,514,517],{},[43,506,507],{},"Prompts are capped at 4,096 tokens. Each 1024 px reference image costs about 1,000 of them, so an edit takes about\nthree references, not the model's 16. Raising the cap needs more text-encoder memory; we have not measured how much.",[43,509,510],{},"The largest canvas is 16.8 MP (4736 x 3552). MiniMax's 21:9 size at 16 MP (6272 x 2688) is slightly larger and does\nnot fit next to the weights.",[43,512,513],{},"Requests run one at a time.",[43,515,516],{},"The host needs 64 GB of RAM, because the text encoder streams from pinned memory, and 48 GB of disk for the weights.",[43,518,519],{},"The same seed gives the same noise as the diffusers pipeline to within the last bit of each value, so images are\nclose to, but not bit-identical with, BF16 runs from the same seed.",[92,521,523],{"id":522},"try-it","Try it",[10,525,526,527,530],{},"You need Docker, a ROCm-capable AMD driver and Python 3 on the host. Check the ",[14,528,529],{"href":89},"model licence"," before downloading or running it.",[532,533,538],"pre",{"className":534,"code":535,"language":536,"meta":537,"style":537},"language-bash shiki shiki-themes github-light github-dark","git clone --depth 1 https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin.git\ncd paiton-vllm-plugin\n.\u002Fmodels\u002FMiniMax-H3-Image\u002Flaunch.sh\n","bash","",[539,540,541,564,573],"code",{"__ignoreMap":537},[542,543,546,550,554,558,561],"span",{"class":544,"line":545},"line",1,[542,547,549],{"class":548},"sScJk","git",[542,551,553],{"class":552},"sZZnC"," clone",[542,555,557],{"class":556},"sj4cs"," --depth",[542,559,560],{"class":556}," 1",[542,562,563],{"class":552}," https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin.git\n",[542,565,567,570],{"class":544,"line":566},2,[542,568,569],{"class":556},"cd",[542,571,572],{"class":552}," paiton-vllm-plugin\n",[542,574,576],{"class":544,"line":575},3,[542,577,578],{"class":548},".\u002Fmodels\u002FMiniMax-H3-Image\u002Flaunch.sh\n",[10,580,581],{},"In another terminal, from the same repository directory:",[532,583,585],{"className":534,"code":584,"language":536,"meta":537,"style":537},".\u002Fmodels\u002FMiniMax-H3-Image\u002Frun.sh generate \"A red fox in fresh snow at sunrise\" --size 4096x4096 --out fox.png\n",[539,586,587],{"__ignoreMap":537},[542,588,589,592,595,598,601,604,607],{"class":544,"line":545},[542,590,591],{"class":548},".\u002Fmodels\u002FMiniMax-H3-Image\u002Frun.sh",[542,593,594],{"class":552}," generate",[542,596,597],{"class":552}," \"A red fox in fresh snow at sunrise\"",[542,599,600],{"class":556}," --size",[542,602,603],{"class":552}," 4096x4096",[542,605,606],{"class":556}," --out",[542,608,609],{"class":552}," fox.png\n",[532,611,613],{"className":534,"code":612,"language":536,"meta":537,"style":537},".\u002Fmodels\u002FMiniMax-H3-Image\u002Frun.sh edit \"Make \u003CPicture 1> a watercolor painting\" --image photo.png --size 2048x2048 --out edit.png\n",[539,614,615],{"__ignoreMap":537},[542,616,617,619,622,625,628,631,633,636,638],{"class":544,"line":545},[542,618,591],{"class":548},[542,620,621],{"class":552}," edit",[542,623,624],{"class":552}," \"Make \u003CPicture 1> a watercolor painting\"",[542,626,627],{"class":556}," --image",[542,629,630],{"class":552}," photo.png",[542,632,600],{"class":556},[542,634,635],{"class":552}," 2048x2048",[542,637,606],{"class":556},[542,639,640],{"class":552}," edit.png\n",[10,642,643,644,647,648,653],{},"The first start downloads 48 GB of weights from Hugging Face and checks each file against its SHA-256. The server\ntakes requests once it prints ",[539,645,646],{},"READY http:\u002F\u002F0.0.0.0:30000",". The\n",[14,649,652],{"href":650,"rel":651},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Ftree\u002Fmain\u002Fmodels\u002FMiniMax-H3-Image",[18],"guide"," has the full API and the\nsettings.",[92,655,90],{"id":656},"license",[10,658,659,660,665],{},"The model is under the ",[14,661,664],{"href":662,"rel":663},"https:\u002F\u002Fhuggingface.co\u002FMiniMaxAI\u002FMiniMax-H3-Image\u002Fblob\u002Fmain\u002FLICENSE",[18],"MiniMax-H3-Image Community License",". Its standard grant excludes the European Union, the United Kingdom, the Republic of Korea and the United States. The territorial restrictions also cover displaying and distributing generated outputs.",[10,667,668],{},"Commercial products must prominently show \"MiniMax-H3-Image\" in their interface. Hosted services for third parties, or licensees and\u002For affiliates with more than US$1 million in annualized revenue, need prior written authorization from MiniMax.",[10,670,671],{},"The images in this article were generated in a sandboxed test environment to illustrate our work on a single R9700. We do not control or monitor how others use the model. Before using it, readers are responsible for checking the licence and obtaining any required permissions.",[10,673,674,675,679,680,390],{},"Our ",[14,676,678],{"href":677},"\u002Fproducts\u002Fpaiton","Paiton framework"," also runs other image and language workloads on AMD GPUs. For MiniMax's video model, see our separate ",[14,681,683],{"href":682},"\u002Fblog\u002Fpaiton-minimax-h3-radeon-ai-pro-r9700","R9700 video-and-audio article",[685,686,687],"style",{},"html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":537,"searchDepth":566,"depth":566,"links":689},[690,691,692,699,700,701,702,703,704],{"id":94,"depth":566,"text":48},{"id":117,"depth":566,"text":118},{"id":226,"depth":566,"text":60,"children":693},[694,695,696,697,698],{"id":239,"depth":575,"text":240},{"id":246,"depth":575,"text":247},{"id":256,"depth":575,"text":257},{"id":266,"depth":575,"text":267},{"id":273,"depth":575,"text":274},{"id":280,"depth":566,"text":66},{"id":396,"depth":566,"text":72},{"id":468,"depth":566,"text":78},{"id":501,"depth":566,"text":502},{"id":522,"depth":566,"text":523},{"id":656,"depth":566,"text":90},[706,707,708,709,710],"Paiton","AMD Radeon","Local AI","Image Generation","MiniMax","2026-11-12T10:00:00Z","MiniMax-H3-Image on one Radeon AI PRO R9700: 2048 x 2048 in 6.9 seconds, 4096 x 4096 in 33.4 seconds. FP8 weights, image editing and setup.","md","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3-image\u002F01-hero-concert.jpg",{},true,"https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-minimax-h3-image-radeon-ai-pro-r9700","\u002Fblog\u002Fpaiton-minimax-h3-image-radeon-ai-pro-r9700",{"title":5,"description":712},"paiton-minimax-h3-image-radeon-ai-pro-r9700","blog\u002Fpaiton-minimax-h3-image-radeon-ai-pro-r9700",null,"mUsTj2lFyGlYS5gTkNA-Qw-BBdYaU_QgqwPDshC-whw",[725,727,739,749,761,771,780,795,804,813,822,831,844,854,865,874,888,897,909,940,951,973,991,1010,1028,1045,1062,1073,1089,1104,1116,1125,1133,1148,1160,1171,1182,1192,1205,1215,1228,1239,1249,1260,1269,1281,1292,1301],{"path":718,"title":5,"description":712,"date":711,"slug":720,"image":714,"originalUrl":717,"categories":726},[706,707,708,709,710],{"path":728,"title":729,"description":730,"date":731,"slug":732,"image":733,"originalUrl":734,"categories":735},"\u002Fblog\u002Fpaiton-kimi-k31-pro-amd-instinct-mi355x","Kimi-K3.1-Pro on 6 MI355X GPUs at 800 W with Paiton","Kimi-K3.1-Pro on six MI355X GPUs at 800 W each. Paiton throughput, power measurements, a separate 1M-context test and a four-GPU option.","2026-11-11T12:00:00Z","paiton-kimi-k31-pro-amd-instinct-mi355x","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-kimi31\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-kimi-k31-pro-amd-instinct-mi355x",[706,736,737,738],"AMD Instinct","Inference","Kimi",{"path":740,"title":741,"description":742,"date":743,"slug":744,"image":745,"originalUrl":746,"categories":747},"\u002Fblog\u002Fpaiton-idol-radeon-ai-pro-r9700","The Radeon AI PRO R9700 does 3D too: Idol in 16.4 seconds","Idol turns an image into a 3D Gaussian splat in 16.44 seconds on one Radeon AI PRO R9700 with Paiton, versus 39.22 seconds for stock on the same GPU.","2026-11-06T09:00:00Z","paiton-idol-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-idol\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-idol-radeon-ai-pro-r9700",[706,707,708,748],"3D Generation",{"path":750,"title":751,"description":752,"date":753,"slug":754,"image":755,"originalUrl":756,"categories":757},"\u002Fblog\u002Fpaiton-deepseek-v41-flash-amd-instinct-mi350p","Deepseek V4.1 Flash on a single MI350P","DeepSeek V4.1 Flash on one 144 GB MI350P: 77.9 tokens\u002Fs for one request and 1,107.6 in a 512-request batch, with Paiton and CPU expert offload.","2026-11-03T12:00:00Z","paiton-deepseek-v41-flash-amd-instinct-mi350p","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-deepseek-v41\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-deepseek-v41-flash-amd-instinct-mi350p",[706,736,758,759,760],"AI Inference","DeepSeek","Quantization",{"path":762,"title":763,"description":764,"date":765,"slug":766,"image":767,"originalUrl":768,"categories":769},"\u002Fblog\u002Fpaiton-qwen38-flash-next-amd-instinct-mi350p","Qwen3.8 Flash Next on MI350P: twice the speed | Paiton","Qwen3.8 Flash Next on one AMD MI350P: 2.24× the tested vLLM setup's single-request decode speed and about 2.2 million shared KV-cache tokens with Paiton.","2026-10-30T12:00:00Z","paiton-qwen38-flash-next-amd-instinct-mi350p","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-mi350p\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-flash-next-amd-instinct-mi350p",[706,736,758,770,760],"Qwen",{"path":772,"title":773,"description":774,"date":775,"slug":776,"image":777,"originalUrl":778,"categories":779},"\u002Fblog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp","Qwen3.8 Flash Next: 3-bit near 16-bit accuracy, with MTP prefix caching","Paiton's new Qwen3.8 3-bit weights approach 16-bit task scores on two R9700 cards. Native MTP prefix caching resumes a 200K prompt in 0.68 seconds. Results, limits and setup.","2026-10-24T18:45:00Z","paiton-qwen38-3bit-accuracy-prefix-caching-mtp","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-3bit-mtp\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp",[706,707,708,770,760],{"path":781,"title":782,"description":783,"date":784,"slug":785,"image":786,"originalUrl":787,"categories":788},"\u002Fblog\u002Fnvidia-dsx-pod-vera-rubin-modular-data-centers","Nvidia confirms the modular DC direction","NVIDIA DSX POD brings Vera Rubin AI factories to modular data centers. ElioVP explains site readiness, cooling and the thinking behind ModFlex.","2026-10-21T13:00:00Z","nvidia-dsx-pod-vera-rubin-modular-data-centers","\u002Fasset\u002Fimages\u002Fblog\u002Fnvidia-dsx-pod-modflex\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fnvidia-dsx-pod-vera-rubin-modular-data-centers",[789,790,791,792,793,794],"Modular DC","AI Infrastructure","ModFlex","NVIDIA","Vera Rubin","Liquid Cooling",{"path":796,"title":797,"description":798,"date":799,"slug":800,"image":801,"originalUrl":802,"categories":803},"\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700","Qwen3.8 Flash Next in 4-bit on 2 × Radeon R9700: tested against 16-bit","Paiton runs Qwen3.8 Flash Next in 4-bit on two R9700 cards: 96.8% GSM8K, 96.3% HumanEval, 189.8 tokens\u002Fs and an optional RAM prefix cache. Accuracy, speed and setup.","2026-10-12T15:00:00Z","paiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700",[706,707,708,770,760],{"path":805,"title":806,"description":807,"date":808,"slug":809,"image":810,"originalUrl":811,"categories":812},"\u002Fblog\u002Fpaiton-qwen38-flash-next-radeon-ai-pro-r9700","Qwen3.8 Flash Next on 2 × Radeon AI PRO R9700: 216 tokens\u002Fs, 200K context","Paiton runs Qwen3.8 Flash Next with 3-bit experts on two R9700 cards: 216.3 tokens\u002Fs, 565.1 aggregate tokens\u002Fs and a separate 200K mode. Benchmarks, quality and setup.","2026-10-10T10:30:00Z","paiton-qwen38-flash-next-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-flash-next\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-flash-next-radeon-ai-pro-r9700",[706,707,708,770,760],{"path":814,"title":815,"description":816,"date":817,"slug":818,"image":819,"originalUrl":820,"categories":821},"\u002Fblog\u002Fpaiton-qwen38-w3a4-radeon-ai-pro-r9700","Qwen3.8 27B on 1 × Radeon AI PRO R9700: 3-bit weights, 20% faster decode","Qwen3.8 27B on one R9700: 19.9% faster weighted decode, a new 4-bit cache, 200K context and optional vision. Benchmarks and current Paiton setup.","2026-09-27T09:00:00Z","paiton-qwen38-w3a4-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-w3a4\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-w3a4-radeon-ai-pro-r9700",[706,707,708,770,760],{"path":823,"title":824,"description":825,"date":826,"slug":827,"image":828,"originalUrl":829,"categories":830},"\u002Fblog\u002Fpaiton-qwen-image-21-radeon-ai-pro-r9700","Qwen-Image 2.1 on 1 × Radeon AI PRO R9700: 2048×2048 images in 103 seconds","Generate 2048×2048 Qwen-Image 2.1 images locally with 1 × Radeon AI PRO R9700. The released Paiton v1.0.2 container measured 103.29 seconds per warm request, through PNG delivery.","2026-09-23T09:00:00Z","paiton-qwen-image-21-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen-image-21\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen-image-21-radeon-ai-pro-r9700",[706,707,708,709,770],{"path":832,"title":833,"description":834,"date":835,"slug":836,"image":837,"originalUrl":838,"categories":839},"\u002Fblog\u002Fpaiton-qwen38-mxfp4-dflash2-r9700","Qwen3.8: 400.7 tok\u002Fs on R9700 | Paiton","Qwen3.8 on one R9700: 400.7 aggregate tok\u002Fs with ROCm 10 and vLLM 0.29, plus public 200K\u002F220K chat profiles. Benchmarks, limits and launch commands.","2026-09-16T07:30:00Z","paiton-qwen38-mxfp4-dflash2-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-mxfp4\u002Fupdate-2026-09-19\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-mxfp4-dflash2-r9700",[706,707,840,841,842,843],"vLLM","Qwen3.8","Inference Optimization","DFlash2",{"path":845,"title":846,"description":847,"date":848,"slug":849,"image":850,"originalUrl":851,"categories":852},"\u002Fblog\u002Fpaiton-qwen38-neo-gguf-vllm-r9700","Qwen3.8 GGUF in vLLM: Faster Responses on One Radeon","Run the original NEO CODER MAX GGUF in vLLM with Paiton on an R9700. Explore measured latency gains, image input and local deployment.","2026-09-14T07:30:00Z","paiton-qwen38-neo-gguf-vllm-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-neo-gguf\u002F00-hero-neo-gguf-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-neo-gguf-vllm-r9700",[706,707,708,853,840],"GGUF",{"path":682,"title":855,"description":856,"date":857,"slug":858,"image":859,"originalUrl":860,"categories":861},"MiniMax H3 on Radeon: 15-Second Video With Native Sound","Paiton generates a 15-second MiniMax H3 video with stereo audio on one Radeon AI PRO R9700 in 5m 33s, with 16.7% lower latency than matched stock.","2026-09-09T07:30:00Z","paiton-minimax-h3-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3\u002F00-featured-minimax-h3-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-minimax-h3-radeon-ai-pro-r9700",[706,707,708,862,863,864],"Video Generation","MiniMax H3","ComfyUI",{"path":866,"title":867,"description":868,"date":869,"slug":870,"image":871,"originalUrl":866,"categories":872},"\u002Fblog\u002Fpaiton-flux2-klein-radeon-ai-pro-r9700","Local FLUX.2 klein on Radeon AI PRO R9700: Faster Image Generation with Less VRAM","Paiton generates 1024×1024 FLUX.2 klein images in 1.054 seconds on an R9700: 16.2% lower warm latency and 33.4% lower peak Torch allocation.","2026-09-07T09:00:00","paiton-flux2-klein-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-flux2-klein\u002Ffox-paiton.webp",[706,707,708,709,873,864],"FLUX",{"path":875,"title":876,"description":877,"date":878,"slug":879,"image":880,"originalUrl":881,"categories":882},"\u002Fblog\u002Fpaiton-ornith15-radeon-ai-pro-r9700","Ornith 1.5 at 44.6 tok\u002Fs on One Radeon AI PRO R9700","Paiton serves Ornith 1.5 35B A3B at 44.63 output tok\u002Fs on one Radeon AI PRO R9700, 27% faster than tuned stock vLLM, with 21.3% lower modeled cost.","2026-09-05T09:00:00","paiton-ornith15-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-ornith15\u002F00-featured-ornith15-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-ornith15-radeon-ai-pro-r9700",[706,883,707,758,884,885,842,886,840,887],"Artificial Intelligence","GPU Performance","Inference Latency","Large Language Models","Cost Efficiency",{"path":889,"title":890,"description":891,"date":892,"slug":893,"image":894,"originalUrl":895,"categories":896},"\u002Fblog\u002Fpaiton-qwen38-radeon-ai-pro-r9700","Paiton: 21% More Qwen3.8 Throughput on Radeon AI PRO R9700","Paiton serves AMD’s Qwen3.8 27B at 39.77 output tokens\u002Fs on one Radeon AI PRO R9700, delivering 21% more throughput and 17.4% lower modeled cost.","2026-09-04T09:00:00","paiton-qwen38-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-r9700\u002F00-featured-paiton-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-radeon-ai-pro-r9700",[706,883,707,758,884,885,842,886,840,887],{"path":898,"title":899,"description":900,"date":901,"slug":902,"image":903,"originalUrl":722,"categories":904},"\u002Fblog\u002Fai-data-center-power-requirements-gpu-per-megawatt","AI Data Center Power Requirements: The GPU\u002FMW Illusion","Why do AI data-center proposals quote different GPU capacities? See how PUE, peak loads, storage, networking and cooling determine deployable compute.","2026-07-27T23:52:00","ai-data-center-power-requirements-gpu-per-megawatt","\u002Fasset\u002Fimages\u002Fblog\u002Fai-data-center-power-requirements-gpu-per-megawatt\u002Fgpu-per-megawatt-illusion.webp",[905,790,906,791,907,908],"All","Data Centers","HPC","AMD Helios",{"path":910,"title":911,"description":912,"date":913,"slug":914,"image":915,"originalUrl":916,"categories":917},"\u002Fblog\u002Fpaiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x","Wan2.2 Video Generation: Paiton on AMD MI355X","Compare Wan2.2-T2V-A14B video generation on AMD MI355X with Paiton and NVIDIA B200 using Diffusers, and explore our diffusion optimization approach.","2026-06-10T14:04:04","paiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonwan2.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x\u002F",[905,883,706,918,919,920,921,922,923,924,925,926,927,928,737,929,930,931,932,933,706,934,935,936,937,938,939],"14B","AMD","B200","Benchmarks","Blackwell","Compute","Diffusion","Eliovp","Generative AI","GPU","Hardware","Instinct","MI355x","NVidia","On-Premise","Optimization","Sovereign AI","T2V","Text-to-Video","Tuning","Video-Generation","Wan2.2",{"path":941,"title":942,"description":943,"date":944,"slug":945,"image":946,"originalUrl":947,"categories":948},"\u002Fblog\u002Ffrom-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure","ElioVP in De Tijd: Chip Optimization and Data Centers","Read about De Tijd's coverage of ElioVP, from its origins in chip optimization to its work on modular data centers and high-density cooling.","2026-02-10T20:48:12","from-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fphysicalnewspaper.webp","https:\u002F\u002Feliovp.com\u002Ffrom-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure\u002F",[905,883,789,949,919,950,789,931],"Uncategorized","De Tijd",{"path":952,"title":953,"description":954,"date":955,"slug":956,"image":957,"originalUrl":958,"categories":959},"\u002Fblog\u002Fprivacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit","AI Privacy: A Strategic Priority for Benelux Businesses","Explore generative AI privacy risks, trust, data retention and governance, and why Benelux businesses need a strategic approach to secure AI.","2026-01-29T13:51:11","privacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fheaderimage.webp","https:\u002F\u002Feliovp.com\u002Fprivacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit\u002F",[905,883,960,949,961,962,963,964,965,966,967,968,925,969,926,970,971,972],"Trending","AI Act","Anthropomorphism","AVG","Benelux","ChatGPT","Cybersecurity","Data Governance","Data Security","GDPR","Microsoft Copilot","Privacy","Shadow AI",{"path":974,"title":975,"description":976,"date":977,"slug":978,"image":979,"originalUrl":980,"categories":981},"\u002Fblog\u002Fitsme-bij-ons-is-het-its-not-me-en-dit-is-waarom","Why We Do Not Use itsme: Privacy and Data Sovereignty","Why ElioVP does not use itsme: our assessment of identity metadata, cloud dependence, data sovereignty and authentication risks.","2025-11-27T09:32:14","itsme-bij-ons-is-het-its-not-me-en-dit-is-waarom","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffrontimage.webp","https:\u002F\u002Feliovp.com\u002Fitsme-bij-ons-is-het-its-not-me-en-dit-is-waarom\u002F",[905,982,983,984,966,985,986,987,969,988,989,990,971],"AWS","Belgian Mobile ID","Cloud Act","Data Sovereignty","Digital Identity","eIDAS","itsme","Liberty Global","MyGov.be",{"path":992,"title":993,"description":994,"date":995,"slug":996,"image":997,"originalUrl":998,"categories":999},"\u002Fblog\u002Ffield-report-the-reality-of-building-agentic-ai-in-2025","Field Report. The Reality of Building Agentic AI in 2025","Lessons from building on-premise AI agents in 2025 cover workflow design, observability, model training, hallucinations and GPU memory limits.","2025-11-25T14:03:39","field-report-the-reality-of-building-agentic-ai-in-2025","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffieldreport.webp","https:\u002F\u002Feliovp.com\u002Ffield-report-the-reality-of-building-agentic-ai-in-2025\u002F",[905,883,1000,960,1001,1002,1003,1004,1005,1006,1007,1008,934,1009],"Solutions","Agentic AI","AI Engineering","AI Strategy","Autonomous Agents","Enterprise AI","Local LLM","Model Fine-Tuning","On-Premise AI","VRAM Optimization",{"path":1011,"title":1012,"description":1013,"date":1014,"slug":1015,"image":1016,"originalUrl":1017,"categories":1018},"\u002Fblog\u002Fthe-synthetic-unicorn-bubble","The Synthetic Unicorn Bubble","An analysis of AI neocloud investment risks, examining circular financing, infrastructure claims, contract terms and due diligence.","2025-11-24T19:22:28","the-synthetic-unicorn-bubble","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fsyntheticunicorn.webp","https:\u002F\u002Feliovp.com\u002Fthe-synthetic-unicorn-bubble\u002F",[905,883,960,790,1019,1020,1021,1022,1023,1024,1025,1026,1027],"AI Neocloud","Circular Financing","GPU Cloud","Investment Risks","Startup Valuation","Synthetic Bubble","Tech Analysis","Vaporware","Venture Capital",{"path":1029,"title":1030,"description":1031,"date":1032,"slug":1033,"image":1034,"originalUrl":1035,"categories":1036},"\u002Fblog\u002Fbuilding-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power","NVIDIA GB300 NVL72: A Four-Month Modular Data Center Plan","Explore a modular data center design for NVIDIA GB300 NVL72, covering redundant power, hybrid cooling and a four-month deployment plan.","2025-11-20T14:10:19","building-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fsuperpodmodflexfrontimage.webp","https:\u002F\u002Feliovp.com\u002Fbuilding-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power\u002F",[905,789,949,1037,790,1038,1039,794,1040,1041,1042,1043,1044],"150kW Rack","DLC","High Density","Modular Data Center","NVIDIA Blackwell Ultra","NVIDIA GB300","NVL72","Rapid Deployment",{"path":1046,"title":1047,"description":1048,"date":1049,"slug":1050,"image":1051,"originalUrl":1052,"categories":1053},"\u002Fblog\u002Fwhy-cuda-translation-wont-unlock-amds-real-potential","Why “CUDA” Translation Won’t Unlock AMD’s Real Potential","Why CUDA compatibility is not the same as AMD performance: explore ROCm, HIP, kernel tuning and the case for hardware-specific optimization.","2025-11-12T14:48:37","why-cuda-translation-wont-unlock-amds-real-potential","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fchatgpt-image-nov-11-2025-09_16_10-pm-1.webp","https:\u002F\u002Feliovp.com\u002Fwhy-cuda-translation-wont-unlock-amds-real-potential\u002F",[905,883,706,949,1054,883,1055,1056,1057,1058,1059,1060,706,1061],"AMD MI300X","CUDA Translation","FP8","GPU Optimization","High Performance Computing","HIP","Kernel Tuning","ROCm",{"path":1063,"title":1064,"description":1065,"date":1066,"slug":1067,"image":1068,"originalUrl":1069,"categories":1070},"\u002Fblog\u002Fpaiton-the-simplest-way-to-supercharge-ai-inference","Paiton: The Simplest Way to Supercharge AI Inference","Learn how Paiton integrates with existing inference stacks, with AMD MI300X benchmark results and performance-per-dollar comparisons.","2025-11-11T10:31:22","paiton-the-simplest-way-to-supercharge-ai-inference","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaiton-powaaah.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-the-simplest-way-to-supercharge-ai-inference\u002F",[905,883,706,758,736,1054,887,1071,842,1060,706,1072,840],"High Throughput","SGLang",{"path":1074,"title":1075,"description":1076,"date":1077,"slug":1078,"image":1079,"originalUrl":1080,"categories":1081},"\u002Fblog\u002Fstop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens","Paiton MoE Benchmarks: MI300X vs H200 and B200","Compare Qwen3-30B-A3B MoE inference with Paiton on MI300X against H200 and B200, including throughput and cost per million tokens.","2025-09-26T13:36:18","stop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fhulkvshulkpaitonwins.webp","https:\u002F\u002Feliovp.com\u002Fstop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens\u002F",[905,883,706,1082,1054,1083,842,1084,1085,1086,1087,706,1088],"AI Benchmarks","Cost per Token","Mixture of Experts","MoE","NVIDIA B200","NVIDIA H200","Qwen3",{"path":1090,"title":1091,"description":1092,"date":1093,"slug":1094,"image":1095,"originalUrl":1096,"categories":1097},"\u002Fblog\u002Fagentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en","Local Agentic AI: From Inbox to Action","Local-first AI agents turn email, documents and images into tickets, reports and actions, using models tailored to your data and systems.","2025-09-16T13:09:00","agentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffrontfotoblog.webp","https:\u002F\u002Feliovp.com\u002Fagentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en\u002F",[905,883,1000,949,1001,1098,1099,1100,1101,1006,1008,934,1102,1103],"Damage Detection","Document Processing","Email Automation","Invoice Extraction","Ticket Automation","Workflow Automation",{"path":1105,"title":1106,"description":1107,"date":1108,"slug":1109,"image":1110,"originalUrl":1111,"categories":1112},"\u002Fblog\u002Fmi300x-fp8-data-parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach","MI300X FP8 Benchmarks: GPU Partitioning with Paiton","Explore Llama 3.1 8B FP8 benchmarks on partitioned MI300X GPUs with Paiton, comparing throughput and latency against NVIDIA H200 and B200.","2025-07-31T13:32:57","mi300x-fp8-data-parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fimage-2-1.webp","https:\u002F\u002Feliovp.com\u002Fmi300x-fp8-data%e2%80%91parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach\u002F",[905,883,706,949,1113,919,920,1114,1115,930,931,706,840],"AI","H200","MI300X",{"path":1117,"title":1118,"description":1119,"date":1120,"slug":1121,"image":1122,"originalUrl":1123,"categories":1124},"\u002Fblog\u002Fapplicable-ai-for-businesses","Applicable AI for Businesses","Explore ElioVP's approach to local AI for business workflows, including custom model training and automated damage detection for logistics.","2025-07-09T21:35:30","applicable-ai-for-businesses","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fscherm_afbeelding-2025-07-09-om-23.30.17.webp","https:\u002F\u002Feliovp.com\u002Fapplicable-ai-for-businesses\u002F",[905,883,1000,1001,1098,1099,1100,1101,1006,1008,934,1102,1103],{"path":1126,"title":1127,"description":1128,"date":1129,"slug":1130,"image":537,"originalUrl":1131,"categories":1132},"\u002Fblog\u002Fintroducing-paitons-free-evaluation-models","Introducing Paiton’s Free Evaluation Models","Test Paiton with free evaluation models for AMD GPUs. Compare text, vision and image generation performance using your own workloads.","2025-07-07T11:26:13","introducing-paitons-free-evaluation-models","https:\u002F\u002Feliovp.com\u002Fintroducing-paitons-free-evaluation-models\u002F",[905,883,706],{"path":1134,"title":1135,"description":1136,"date":1137,"slug":1138,"image":1139,"originalUrl":1140,"categories":1141},"\u002Fblog\u002Fpaiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b","Llama 3.1 405B: Faster Startup with Paiton on MI300X","See Paiton benchmarks for Llama 3.1 405B on eight AMD MI300X GPUs, covering model startup, tensor parallelism, throughput and latency.","2025-06-12T20:15:23","paiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fservingscreenshot.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b\u002F",[905,883,706,949,758,1054,1142,1056,1143,1144,1145,706,1146,1147],"Cold Start","Graph Compilation","Llama 3.1 405B","LLM Optimization","Startup Latency","Tensor Parallelism",{"path":1149,"title":1150,"description":1151,"date":1152,"slug":1153,"image":1154,"originalUrl":1155,"categories":1156},"\u002Fblog\u002Fpaiton-fp8-beats-nvidias-h200-on-amds-mi300x","Paiton FP8 Beats NVIDIA’s H200 on AMD’s MI300X","Compare Paiton on AMD MI300X with NVIDIA H200 for Llama 3.1 70B FP8, including throughput, first-token delay and latency across batch sizes.","2025-06-08T19:12:40","paiton-fp8-beats-nvidias-h200-on-amds-mi300x","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fblognewfp8.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-fp8-beats-nvidias-h200-on-amds-mi300x\u002F",[905,883,706,949,1054,1157,1005,926,884,885,886,1144,1158,1159],"Cold Start Optimization","Model Serving","vLLM Optimization",{"path":1161,"title":1162,"description":1163,"date":1164,"slug":1165,"image":1166,"originalUrl":1167,"categories":1168},"\u002Fblog\u002Fmi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm","MI300X vs H200 vs RX 7900 XTX vs Tenstorrent n300s with vLLM","Compare MI300X, H200, RX 7900 XTX and Tenstorrent n300s on Llama 3 8B with vLLM, including throughput, modeled token costs and hardware limits.","2025-05-09T14:03:58","mi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcomparisontenstor.webp","https:\u002F\u002Feliovp.com\u002Fmi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm\u002F",[905,883,706,1000,949,919,1115,931,1169,1170],"RX7900XTX","tenstorrent",{"path":1172,"title":1173,"description":1174,"date":1175,"slug":1176,"image":1177,"originalUrl":1178,"categories":1179},"\u002Fblog\u002Fclusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights","ClusterP&L: Financial Modeling for GPU Clusters","Explore how ClusterP&L models GPU cluster costs, profitability and investment scenarios, with ROI metrics, risk simulations and exportable reports.","2025-05-03T10:52:22","clusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcomparisonscenarios.webp","https:\u002F\u002Feliovp.com\u002Fclusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights\u002F",[905,883,789,1000,920,1114,1180,931,1181],"MI325x","pnl calculator",{"path":1183,"title":1184,"description":1185,"date":1186,"slug":1187,"image":1188,"originalUrl":1189,"categories":1190},"\u002Fblog\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200","AMD MI300X vs. NVIDIA H200: Qwen3-32B with Paiton","Compare Qwen3-32B benchmarks on Paiton-optimized AMD MI300X and NVIDIA H200, covering throughput, latency and hardware costs.","2025-05-02T21:10:30","cranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002F3ac59a73-2466-4422-b7e5-ef2e4a8ca58e.webp","https:\u002F\u002Feliovp.com\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200\u002F",[905,883,706,1113,919,1114,1191,931,706,840],"MI300",{"path":1193,"title":1194,"description":1195,"date":1196,"slug":1197,"image":1198,"originalUrl":1199,"categories":1200},"\u002Fblog\u002Fpower-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw","Modular Data Centers for NVIDIA NVL: 1 to 2 MW","Explore modular data center designs for NVIDIA NVL systems, covering power capacity, liquid cooling, redundancy and deployment planning.","2025-05-02T14:09:59","power-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Feliovp_critical-1mw-pod_rev-2_transparent.webp","https:\u002F\u002Feliovp.com\u002Fpower-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw\u002F",[905,789,1201,790,1039,907,794,1040,1202,1203,1043,1204],"1-2MW Data Center","NVIDIA Blackwell","NVIDIA NVL","Precision Cooling",{"path":1206,"title":1207,"description":1208,"date":1209,"slug":1210,"image":1211,"originalUrl":1212,"categories":1213},"\u002Fblog\u002Fexamining-ai-agents-in-the-medical-field-ai-that-speaks-dicom","Examining AI agents in the medical field: AI that speaks DICOM","Explore a local AI agent demo for DICOM workflows, from patient and study retrieval to a comparison of vision models using anonymized medical images.","2025-04-11T14:45:48","examining-ai-agents-in-the-medical-field-ai-that-speaks-dicom","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fhealthcareblog-1.webp","https:\u002F\u002Feliovp.com\u002Fexamining-ai-agents-in-the-medical-field-ai-that-speaks-dicom\u002F",[905,883,1000,949,1113,919,1214],"Healthcare",{"path":1216,"title":1217,"description":1218,"date":1219,"slug":1220,"image":1221,"originalUrl":1222,"categories":1223},"\u002Fblog\u002Feliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs","U.S. Tariffs and AI Supply Chain Resilience: April 2025","Read ElioVP's April 2025 perspective on U.S. tariffs and supply chain resilience for AI servers, HPC systems and modular data centers.","2025-04-04T10:01:27","eliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ftariffsshipping.webp","https:\u002F\u002Feliovp.com\u002Feliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs\u002F",[905,960,1113,919,1224,1225,1226,1227],"import","Taiwan","Tariffs","Trump",{"path":1229,"title":1230,"description":1231,"date":1232,"slug":1233,"image":1234,"originalUrl":1235,"categories":1236},"\u002Fblog\u002Fwhy-ai-agents-are-the-future","Why AI Agents Are the Future","Explore AI agents for ERP, CRM, finance and customer support, with practical use cases and a path from workflow assessment to pilot and deployment.","2025-03-23T22:06:59","why-ai-agents-are-the-future","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ferp2.jpeg","https:\u002F\u002Feliovp.com\u002Fwhy-ai-agents-are-the-future\u002F",[905,883,1000,1113,1237,1238],"AI Agents","ERP",{"path":1240,"title":1241,"description":1242,"date":1243,"slug":1244,"image":1245,"originalUrl":1246,"categories":1247},"\u002Fblog\u002Fthe-rise-of-open-source-ai-model-optimization","The Rise of Open-Source AI Model Optimization","Explore open-source AI optimization trends, from quantization and mixture-of-experts models to hardware-aware tuning, RAG and edge deployment.","2025-03-22T20:59:23","the-rise-of-open-source-ai-model-optimization","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Friseofopensource.jpeg","https:\u002F\u002Feliovp.com\u002Fthe-rise-of-open-source-ai-model-optimization\u002F",[905,883,960,1248,919,927,931],"AI news",{"path":1250,"title":1251,"description":1252,"date":1253,"slug":1254,"image":1255,"originalUrl":1256,"categories":1257},"\u002Fblog\u002Fintroducing-our-benchmarking-tool-powered-by-dstack","Introducing Our Benchmarking Tool: Powered by dstack","Explore our dstack-powered tool for reproducible vLLM benchmarks, automated parameter sweeps and performance reports across local and cloud GPUs.","2025-03-20T14:21:59","introducing-our-benchmarking-tool-powered-by-dstack","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fbenchmarktool.jpeg","https:\u002F\u002Feliovp.com\u002Fintroducing-our-benchmarking-tool-powered-by-dstack\u002F",[905,883,706,1113,919,1258,1259,1115,706],"benchmark","LLM",{"path":1261,"title":1262,"description":1263,"date":1264,"slug":1265,"image":1266,"originalUrl":1267,"categories":1268},"\u002Fblog\u002Foptimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200","Optimizing QwQ-32B (by Qwen): AMD MI300X vs. NVIDIA H200","Compare QwQ-32B throughput and latency on AMD MI300X with Paiton and NVIDIA H200, from small batches to higher concurrency.","2025-03-19T21:41:44","optimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaiton4.jpeg","https:\u002F\u002Feliovp.com\u002Foptimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200\u002F",[905,883,706],{"path":1270,"title":1271,"description":1272,"date":1273,"slug":1274,"image":1275,"originalUrl":1276,"categories":1277},"\u002Fblog\u002Feliovp-featured-on-amd-tech-talk-podcast","Eliovp Featured on AMD “Tech Talk” Podcast","Listen to Elio Van Puyvelde and Jim Greene on AMD's Tech Talk podcast, discussing ElioVP's origins and its AI hardware and software services.","2025-03-19T07:53:39","eliovp-featured-on-amd-tech-talk-podcast","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ftechtalkjimgreene.jpeg","https:\u002F\u002Feliovp.com\u002Feliovp-featured-on-amd-tech-talk-podcast\u002F",[905,919,1278,1279,1280],"Jim Greene","Podcast","Tech Talk",{"path":1282,"title":1283,"description":1284,"date":1285,"slug":1286,"image":1287,"originalUrl":1288,"categories":1289},"\u002Fblog\u002Ffurther-optimizing-amd-powered-inference-with-paiton","Further Optimizing AMD-Powered Inference with Paiton","Explore Paiton's DeepSeek R1 Distill Llama 8B benchmarks on AMD MI300X, focusing on throughput and first-token latency at smaller batch sizes.","2025-03-13T06:18:30","further-optimizing-amd-powered-inference-with-paiton","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost3.webp","https:\u002F\u002Feliovp.com\u002Ffurther-optimizing-amd-powered-inference-with-paiton\u002F",[905,883,706,919,1290,1291,1114,1115,1180,706,840],"Deepseek","H100",{"path":1293,"title":1294,"description":1295,"date":1296,"slug":1297,"image":1298,"originalUrl":1299,"categories":1300},"\u002Fblog\u002Fa-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b","Paiton Benchmarks: DeepSeek R1 Distill Llama 3.1 8B","Compare stock and Paiton-optimized DeepSeek R1 Distill Llama 3.1 8B on AMD MI300X, with throughput and latency benchmarks across batch sizes.","2025-01-31T09:11:02","a-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost2.webp","https:\u002F\u002Feliovp.com\u002Fa-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b\u002F",[905,883,706,919,1290,1291,1114,1115,1180,706,840],{"path":1302,"title":1303,"description":1304,"date":1305,"slug":1306,"image":1307,"originalUrl":1308,"categories":1309},"\u002Fblog\u002Fai-model-optimization-with-paiton","AI Model Optimization with Paiton","Learn how Paiton uses model compilation, custom kernels and kernel fusion to optimize AI inference on AMD GPUs.","2025-01-30T19:53:25","ai-model-optimization-with-paiton","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost1.webp","https:\u002F\u002Feliovp.com\u002Fai-model-optimization-with-paiton\u002F",[905,883,706,919,1291,1114,1115,1180,706,840],1794475005492]