https://www.nextplatform.com/2020/11/12/intels-first-discrete-xe-server-gpu-aimed-at-hyperscalers/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Tue, 04 Oct 2022 13:40:15 +0000 hourly 1 https://wordpress.org/?v=6.7.1 https://www.nextplatform.com/2020/11/12/intels-first-discrete-xe-server-gpu-aimed-at-hyperscalers/#comment-157178 Mon, 16 Nov 2020 16:50:25 +0000 http://www.nextplatform.com/?p=137404#comment-157178 I wonder whether Amdahl’s lawn is what makes this 4x baby-GPU configuration viable. Namely that speedup of solving large single problems is limited by sequential bottle-necks; so if there are enough concurrent problems (or separable sub-problems) an element of coarse-grained parallelism can increase system-level performance/$. That’s analogous to CPU application architectures which use SIMD and threading within each NUMA node, and MPI between NUMA nodes.

]]> https://www.nextplatform.com/2020/11/12/intels-first-discrete-xe-server-gpu-aimed-at-hyperscalers/#comment-157015 Thu, 12 Nov 2020 20:49:13 +0000 http://www.nextplatform.com/?p=137404#comment-157015 The Xe update yesterday also listed Xe-HP status as “Sampling”. The announcement also stated it is available on their devcloud.

]]>