{"id":35480,"date":"2025-02-08T22:53:24","date_gmt":"2025-02-09T01:53:24","guid":{"rendered":"https:\/\/mastertrend.info\/?p=35480"},"modified":"2025-03-01T22:49:52","modified_gmt":"2025-03-02T01:49:52","slug":"neural-texture-compression","status":"publish","type":"post","link":"https:\/\/mastertrend.info\/en\/la-compresion-de-texturas-neuronales\/","title":{"rendered":"Neural Texture Compression: 96% Less VRAM! \ud83e\udde0\ud83d\udcbe"},"content":{"rendered":"<h2>Nvidia&#039;s neural texture compression revolutionizes! \ud83d\ude80<\/h2>\n<div id=\"article-body\">\n<p>Nvidia&#039;s Neural Texture Compression (NTC) has finally been put to the test, showing the technology&#039;s capabilities in a real-world 3D workload. <a href=\"https:\/\/www.youtube.com\/watch?v=Z9VAloduypg\" target=\"_blank\" rel=\"noopener\" data-analytics-id=\"inline-link\" data-url=\"https:\/\/www.youtube.com\/watch?v=Z9VAloduypg\" data-hl-processed=\"none\" data-schema-attribute=\"mentions\">Compose<\/a> on YouTube tested Nvidia&#039;s memory compression technology on an RTX 4090, at 1440p and 4K resolutions, revealing an impressive reduction in the size of NTC textures by the 96% compared to conventional texture compression techniques. \ud83d\ude80<\/p>\n<p>Compusemble tested NTC in two modes: \u201cNTC transcoded to BCn\u201d and \u201cInference on Sample.\u201d The former transcodes textures to BCn upon loading, while the latter only decompresses the individual texels needed to render a specific view, further reducing the texture memory size. \ud83c\udfae<\/p>\n<aside class=\"hawk-nest\" data-block-type=\"embed\" data-render-type=\"fte\" data-skip=\"dealsy\" data-widget-type=\"seasonal\"><\/aside>\n<p>At 1440p with DLSS upscaling enabled, the \u201cNTC transcoded to BCn\u201d mode reduced the <a href=\"https:\/\/mastertrend.info\/en\/chatbit\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3199\">application test memory size<\/a> on a 64%, going from 272MB to 98MB. However, the \u201cNTC inference on sample\u201d mode significantly decreased the texture size to just 11.37MB, which is a 95.8% reduction compared to non-neural compression and an 88% reduction compared to the previous neural compression mode. \ud83d\ude32<\/p>\n<p>Compusemble metrics revealed that performance takes a slight dip when enabling RTX neural texture compression. \u201cNTC mode transcoded to BCn showed a minimal reduction in average FPS compared to NTC disabled, though the lowest 1% FPS was noticeably better than conventional texture compression with NTC disabled. NTC inference mode on samples suffered a steeper drop, going from average FPS of 1,600 to around 1,500 FPS. Low 1% dropped significantly to around 840 FPS.\u201d \ud83d\udcc9<\/p>\n<div class=\"youtube-video youtube-facade\" data-nosnippet=\"\">\n<div class=\"video-aspect-box\">\n<div class=\"watch-on-iframe-Z9VAloduypg\" title=\"First Look At RTX Neural Texture Compression (BETA) On RTX 4090 - YouTube\" data-yt-video-token=\"Z9VAloduypg\"><span class=\"youtube-video-title\">First look at RTX Neural Texture Compression (BETA) on RTX 4090 \u2013 YouTube<\/span><\/div>\n<div title=\"First Look At RTX Neural Texture Compression (BETA) On RTX 4090 - YouTube\" data-yt-video-token=\"Z9VAloduypg\">\n<div class=\"jeg_video_container jeg_video_content\"><iframe title=\"First Look At RTX Neural Texture Compression (BETA) On RTX 4090\" width=\"500\" height=\"281\" src=\"https:\/\/www.youtube.com\/embed\/Z9VAloduypg?feature=oembed\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" referrerpolicy=\"strict-origin-when-cross-origin\" allowfullscreen><\/iframe><\/div>\n<\/div>\n<\/div>\n<\/div>\n<p>The memory size reduction is the same at 1440p with TAA instead of DLSS upscaling, but the performance behavior of the <a href=\"https:\/\/mastertrend.info\/en\/a-d3d11-compatible-gpu-is-required\/\" title=\"\u26a0\ufe0f A GPU compatible with D3D11 is required: Quick Fix!\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"4198\">GPU<\/a> varies. All modes performed significantly faster than DLSS, operating at nearly 2000 FPS. The lowest 1% in \u201cNTC inference on sample\u201d mode hovered in the 1,300 FPS range, a big jump from 840 FPS. \ud83c\udf89<\/p>\n<p>Not surprisingly, increasing the resolution to 4K makes the <a href=\"https:\/\/mastertrend.info\/en\/ultrabooks-and-premium-laptops\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3196\">performance<\/a> drop dramatically. With DLSS upscaling enabled, it shows an average FPS in the 1,100 FPS range in \u201cNTC transcoded to BCn\u201d mode and just under 1,000 FPS in \u201cNTC inference on sample\u201d mode. The lowest 1% for both modes were in the 500 FPS range.<\/p>\n<p>Disabling DLSS in favor of native resolution with TAA shows an average FPS increase in the 1,700 FPS range with \u201cNTC transcoded to BCn\u201d mode and an average FPS increase in the 1,500 range with \u201cNTC inference on sample.\u201d The lowest 1% for the former NTC mode was just shy of 1,100 FPS, while the latter mode had low 1%s just shy of 800 FPS. \ud83d\udcca<\/p>\n<p>Finally, Compusemble tested cooperative vectors with the \u201cNTC inference in sample\u201d mode at <a href=\"https:\/\/mastertrend.info\/en\/ps5-and-xbox-series-x-glitches-2\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3192\">4K resolution<\/a> with TAA. Cooperative Vectors enabled resulted in an average frame rate in the 1,500 range, while with it disabled, the average FPS plummeted to just under 650 FPS. The lowest 1%s were also just under 750 FPS, with Cooperative Vectors enabled; with it disabled, the lowest 1%s were just above 400 FPS, respectively. \ud83d\udd04<\/p>\n<p>Compusemble&#039;s RTX NTC tests show that the <a href=\"https:\/\/mastertrend.info\/en\/tips-to-speed-up-windows-startup\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3384\">technology<\/a> Nvidia&#039;s neural compression can greatly reduce the texture memory footprint of a 3D application, but at the cost of performance, especially in &quot;inference-on-sample&quot; mode. \u26a0\ufe0f<\/p>\n<p>The most interesting aspect of the results of <a href=\"https:\/\/mastertrend.info\/en\/master-your-gpu\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3195\">performance is DLSS<\/a> versus native resolution. The significant increase in frame rate at native resolution shows that the <a href=\"https:\/\/mastertrend.info\/en\/geforce-rtx-5090-with-32gb\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3193\">tensor cores<\/a> used to process NTC are being pushed hard, likely to the point where DLSS upscaling performance is being compromised, enough to potentially limit shader cores. If this were not to happen, we should see DLSS mode operating at a higher frame rate than native TAA tests at 4K. \ud83d\udd0d<\/p>\n<p>RTX Neural Texture Compression has been in development for at least a few years. The new <a href=\"https:\/\/mastertrend.info\/en\/free-malware-removal-tools\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3342\">technology<\/a> uses the tensor cores of modern Nvidia GPUs to compress textures for 3D applications and games instead of the traditional Block Truncation Code. RTX NTC represents the first major improvement in <a href=\"https:\/\/mastertrend.info\/en\/3d-nanoscale-transistors\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3197\">technology<\/a> texture compression since the 1990s, allowing up to four times higher resolution textures than GPUs are capable of processing today. \ud83d\udcc8<\/p>\n<p>The <a href=\"https:\/\/mastertrend.info\/en\/alternatives-to-shareit-for-windows-2\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3198\">technology<\/a> is in beta and there is no release date. Interestingly, the <a href=\"https:\/\/mastertrend.info\/en\/20-great-games-for-integrated-graphics\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3194\">minimum requirements<\/a> for NTC seem to be surprisingly low. The GitHub page for <a href=\"https:\/\/mastertrend.info\/en\/laptop-rtx-50\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3417\">Nvidia<\/a> for RTX NTC confirms that the minimum GPU requirement is an RTX 20 series GPU. However, the <a href=\"https:\/\/mastertrend.info\/en\/how-to-clear-cache-in-arc-browser-2\/\" target=\"_blank\" rel=\"noopener\" data-wpil-monitor-id=\"3341\">technology<\/a> It has also been validated to work on GTX 10-series GPUs, AMD Radeon RX 6000 GPUs, and Arc A-series GPUs, suggesting we could see this technology become commonplace on non-RTX GPUs and even consoles. \ud83c\udf89<\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>Neural texture compression reduces VRAM on a 96%. Discover Nvidia&#039;s innovative technology that transforms your graphics \ud83c\udfae\u2728<\/p>","protected":false},"author":1,"featured_media":35481,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","iawp_total_views":34,"jnews-multi-image_gallery":[],"jnews_single_post":[],"jnews_primary_category":[],"jnews_social_meta":[],"jnews_review":[],"enable_review":"","type":"","name":"","summary":"","brand":"","sku":"","good":[],"bad":[],"score_override":"","override_value":"","rating":[],"price":[],"jnews_override_counter":[],"footnotes":""},"categories":[880],"tags":[1445,1709,1603],"class_list":["post-35480","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ia","tag-evergreencontent","tag-gpu","tag-nvidia"],"_links":{"self":[{"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/posts\/35480","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/comments?post=35480"}],"version-history":[{"count":0,"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/posts\/35480\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/media\/35481"}],"wp:attachment":[{"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/media?parent=35480"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/categories?post=35480"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mastertrend.info\/en\/wp-json\/wp\/v2\/tags?post=35480"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}