<?xml version="1.0" encoding="UTF-8"?>
<!-- generator="bbPress/1.0.2" -->
<rss version="2.0"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>k-Wave User Forum &#187; Topic: Hint to make GPU version faster for fixed setup</title>
		<link>http://www.k-wave.org/forum/topic/hint-to-make-gpu-version-faster-for-fixed-setup</link>
		<description>Support for the k-Wave MATLAB toolbox</description>
		<language>en-US</language>
		<pubDate>Wed, 13 May 2026 06:06:59 +0000</pubDate>
		<generator>http://bbpress.org/?v=1.0.2</generator>
		<textInput>
			<title><![CDATA[Search]]></title>
			<description><![CDATA[Search all topics from these forums.]]></description>
			<name>q</name>
			<link>http://www.k-wave.org/forum/search.php</link>
		</textInput>
		<atom:link href="http://www.k-wave.org/forum/rss/topic/hint-to-make-gpu-version-faster-for-fixed-setup" rel="self" type="application/rss+xml" />

		<item>
			<title>Bradley Treeby on "Hint to make GPU version faster for fixed setup"</title>
			<link>http://www.k-wave.org/forum/topic/hint-to-make-gpu-version-faster-for-fixed-setup#post-7796</link>
			<pubDate>Wed, 09 Sep 2020 19:38:58 +0000</pubDate>
			<dc:creator>Bradley Treeby</dc:creator>
			<guid isPermaLink="false">7796@http://www.k-wave.org/forum/</guid>
			<description>&#60;p&#62;Hi mattspr,&#60;/p&#62;
&#60;p&#62;Are you using the C++/CUDA code? This doesn't require MATLAB to run. You can create the input HDF5 file using any programming language with HDF5 support (there are some python examples on the forum). Altnernatively, you can compile MATLAB code (e.g., using &#60;code&#62;mcc&#60;/code&#62;) and then running it only requires the MATLAB runtime (which can be freely distributed).&#60;/p&#62;
&#60;p&#62;What is the slow part in your simulation? If it's the actual calculation (i.e., running k-Wave), then there's not much to be done unfortunately, as the C++/CUDA code is already pretty optimised. If it's saving / reading the input and output files, then in principle you could modify the C++/CUDA code so that the required input matrices are generated internally, rather than being loaded from a HDF5 file, although that would take quite a bit of work.&#60;/p&#62;
&#60;p&#62;Hope that helps,&#60;/p&#62;
&#60;p&#62;Brad
&#60;/p&#62;</description>
		</item>
		<item>
			<title>mattspr on "Hint to make GPU version faster for fixed setup"</title>
			<link>http://www.k-wave.org/forum/topic/hint-to-make-gpu-version-faster-for-fixed-setup#post-7769</link>
			<pubDate>Tue, 25 Aug 2020 07:29:58 +0000</pubDate>
			<dc:creator>mattspr</dc:creator>
			<guid isPermaLink="false">7769@http://www.k-wave.org/forum/</guid>
			<description>&#60;p&#62;Also can we and how could we throw away matlab?
&#60;/p&#62;</description>
		</item>
		<item>
			<title>mattspr on "Hint to make GPU version faster for fixed setup"</title>
			<link>http://www.k-wave.org/forum/topic/hint-to-make-gpu-version-faster-for-fixed-setup#post-7768</link>
			<pubDate>Tue, 25 Aug 2020 07:25:53 +0000</pubDate>
			<dc:creator>mattspr</dc:creator>
			<guid isPermaLink="false">7768@http://www.k-wave.org/forum/</guid>
			<description>&#60;p&#62;Hello, we are building hardware for ultrasonic diagnostics. We got fixed grid of emiters and detectors, algorythm to build image (or 3d model) from diagnostics is to minimize error bitween diagnostics result and simulation of kWaveFirstOrder, so we need to run kWaveFirstOrder multiple times. Do you have any hints to make GPU code faster for our case?
&#60;/p&#62;</description>
		</item>

	</channel>
</rss>
