<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Hardware on wporter.org</title>
    <link>https://wporter.org/categories/hardware/</link>
    <description>Recent content in Hardware on wporter.org</description>
    <generator>Hugo</generator>
    <language>en-US</language>
    <lastBuildDate>Sat, 18 Apr 2026 14:30:00 +0000</lastBuildDate>
    <atom:link href="https://wporter.org/categories/hardware/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Linux md RAID benchmarks on 4 Kioxia PM5s</title>
      <link>https://wporter.org/linux-md-raid-benchmarks-on-4-kioxia-pm5s/</link>
      <pubDate>Sat, 18 Apr 2026 14:30:00 +0000</pubDate>
      <guid>https://wporter.org/linux-md-raid-benchmarks-on-4-kioxia-pm5s/</guid>
      <description>&lt;p&gt;I had access to some machines with a load of brand new Kioxia PM5 (read-intensive SAS3) SSDs, so after finishing my production benchmarking I decided to grab some academic numbers to follow up on &lt;a href=&#34;https://wporter.org/linux-md-raid-layouts/&#34;&gt;my prior &lt;code&gt;md&lt;/code&gt; RAID layouts writeup&lt;/a&gt;.&lt;/p&gt;&#xA;&lt;p&gt;I&amp;rsquo;ll be benchmarking just four drives here, as the PCIE 3.0x8 HBA they&amp;rsquo;re hooked up to is a limiting factor for anything wider. These drives really could use a NVME interface (that&amp;rsquo;s what the CM5 is!)&lt;/p&gt;</description>
    </item>
    <item>
      <title>Linux md RAID layouts</title>
      <link>https://wporter.org/linux-md-raid-layouts/</link>
      <pubDate>Sat, 11 Apr 2026 21:15:00 +0000</pubDate>
      <guid>https://wporter.org/linux-md-raid-layouts/</guid>
      <description>&lt;p&gt;I was working on setting up some big mdadm RAID arrays, so I had to do some reading up on RAID layouts to make sure I was setting them up correctly. The result? This nice (if I say so myself) write-up! Yes, I went a little overboard.&lt;/p&gt;&#xA;&lt;p&gt;If you&amp;rsquo;re &lt;em&gt;not&lt;/em&gt; familiar with RAID levels, let&amp;rsquo;s go just deep enough:&lt;/p&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;RAID 0 = &lt;strong&gt;blocks &amp;ldquo;striped&amp;rdquo; across all drives for best performance, no redundancy&lt;/strong&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;Calling this RAID isn&amp;rsquo;t strictly accurate.. it&amp;rsquo;s not redundant at all!&lt;/li&gt;&#xA;&lt;li&gt;Blocks of data are distributed between all member disks. This means excellent performance (both writes and reads can be highly parallelized).&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;RAID 1 = &lt;strong&gt;blocks &amp;ldquo;mirrored&amp;rdquo; between all drives for redundancy&lt;/strong&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;Mirrors one drive to n replicas. As such, you only get the capacity of a single drive.&lt;/li&gt;&#xA;&lt;li&gt;Dumb mirror. Very simple, very fast reads (if you throw enough threads at it, you can read from n drives simultaneously), but writes will never be faster than one drive.&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;RAID 4 = &lt;strong&gt;one dedicated parity (redundancy) drive, data striped between the rest of the drives.&lt;/strong&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;One drive is reserved for redundancy. Parity data is all kept on one drive.&lt;/li&gt;&#xA;&lt;li&gt;Parity bytes are an XOR combination of data bytes (e.g., P = a XOR b XOR c XOR d XOR e - you can figure out what &amp;ldquo;a&amp;rdquo; was given b, c, d, e and P) which is how you get redundancy without a mirror. This is called &amp;ldquo;erasure coding&amp;rdquo; - you can reconstruct missing data if you know exactly which blocks are gone. If a drive fails, you know which blocks are gone.&lt;/li&gt;&#xA;&lt;li&gt;As parity data lives on one dedicated drive, performance of the array is limited by that drive.&lt;/li&gt;&#xA;&lt;li&gt;Because RAID4 (and 5, and 6) rely on parity, small I/O is punished. If you write a small chunk, you can&amp;rsquo;t just write. You have to read the old data block, read the old parity block, XOR old data, new data, old parity to get your new parity value, write the new data block, then write the new parity block.&lt;/li&gt;&#xA;&lt;li&gt;Large sequential writes are less affected, because you don&amp;rsquo;t need to read old data - you can just dump a new set of data blocks and parity block to disk all at once.&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;RAID 5 = &lt;strong&gt;single set of parity data distributed across the array, data striped between all drives&lt;/strong&gt;.&#xA;&lt;ul&gt;&#xA;&lt;li&gt;The equivalent of one drive&amp;rsquo;s capacity is reserved for redundancy. Parity data is distributed across the array.&lt;/li&gt;&#xA;&lt;li&gt;Like RAID4, but, as the parity blocks are distributed between all members, RAID5 allows more parallelism (all drives store parity and data blocks).&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;RAID 6 = &lt;strong&gt;two sets of parity data distributed across the array, data striped between all drives&lt;/strong&gt;.&#xA;&lt;ul&gt;&#xA;&lt;li&gt;The equivalent of two drives&amp;rsquo; capacity is reserved for redundancy. Parity data is distributed across the array.&lt;/li&gt;&#xA;&lt;li&gt;While RAID5 offers just one set of parity blocks, RAID 6 has traditional XOR parity blocks (the &amp;ldquo;P&amp;rdquo; parity data) &lt;em&gt;and&lt;/em&gt; a second set of parity data that&amp;rsquo;s reversible with some more advanced algebra (&amp;ldquo;Q&amp;rdquo; parity data). If you want to read about how that second set of parity data works, &lt;a href=&#34;https://igoro.com/archive/how-raid-6-dual-parity-calculation-works/&#34;&gt;check out Igor Ostrovsky&amp;rsquo;s blog post on the topic&lt;/a&gt;. He does an excellent job at explaining it.&lt;/li&gt;&#xA;&lt;li&gt;The result is that your machine can solve for the missing values using the two parity equations and live data, so you can always figure out what bytes you had on the two disks you&amp;rsquo;re able to lose.&lt;/li&gt;&#xA;&lt;li&gt;If two drives are lost and you need to use the Q parity to reconstruct, performance will suffer significantly, as every read will require reading from all surviving drives and doing relatively expensive calculations.&lt;/li&gt;&#xA;&lt;li&gt;However, if you only lose one drive and can reconstruct the missing data from the P parity algorithm or live data (e.g., only had one drive failure) the array will continue to perform relatively well.&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;RAID 10 = &lt;strong&gt;striped mirrors = pairs of two drives are mirrored, then the mirrored pairs are all striped together.&lt;/strong&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;Half of your drives are used for redundancy. Data is mirrored across a configurable number of members of the array (by default, 2). As such, you&amp;rsquo;ll lose at least 50% of your disk space.&lt;/li&gt;&#xA;&lt;li&gt;Simpler than erasure coding. Very fast. Works just like you&amp;rsquo;d think it does with &amp;ldquo;near&amp;rdquo; distribution, which is what is typically used. More on that later.&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;p&gt;ZFS and LVM support some other fun things like linear volumes or triple parity, but those are different beasts entirely.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Flash Mellanox ConnectX-3 Pro firmware (CX314A-BCCT to CX354A-FCCT)</title>
      <link>https://wporter.org/flash-mellanox-connectx-3-pro-firmware-cx314a-bcct-to-cx354a-fcct/</link>
      <pubDate>Sat, 31 Jan 2026 19:00:00 +0000</pubDate>
      <guid>https://wporter.org/flash-mellanox-connectx-3-pro-firmware-cx314a-bcct-to-cx354a-fcct/</guid>
      <description>&lt;h2 id=&#34;introduction&#34;&gt;Introduction&lt;/h2&gt;&#xA;&lt;p&gt;The ConnectX-3 Pro (MT27520) was sold in a few different configurations (e.g., dual SFP+, QSFP+, dual QSFP+, Ethernet only, Infiniband and Ethernet). However, all the ASICs are the same, so you can easily cross-flash them (It&amp;rsquo;s possible to flash VPI firmware to an Ethernet card if you want to unlock Infiniband.)&lt;/p&gt;&#xA;&lt;p&gt;You can also typically flash OEM cards to stock Mellanox firmware with little hassle.&lt;/p&gt;&#xA;&lt;p&gt;The most recent batch I purchased are CX314 Ethernet-only cards (about $10 each) so I figure I&amp;rsquo;ll crossflash them to the VPI cards before pressing them into service as IB HCAs (for that sweet, sweet RDMA).&lt;/p&gt;</description>
    </item>
    <item>
      <title>1, 10, 25, 40 gig NIC shopping list</title>
      <link>https://wporter.org/1-10-25-40-gig-nic-shopping-list/</link>
      <pubDate>Wed, 14 Jan 2026 20:50:00 +0000</pubDate>
      <guid>https://wporter.org/1-10-25-40-gig-nic-shopping-list/</guid>
      <description>&lt;h2 id=&#34;gigabit&#34;&gt;Gigabit&lt;/h2&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;BCM5719 (quad-port Broadcom PCIe 2.0x4, ~$15, SR-IOV, VMq)&lt;/li&gt;&#xA;&lt;li&gt;BCM5720 (dual-port Broadcom PCIe 2.0x1, ~$10, SR-IOV, VMq)&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;p&gt;Cheap OEM models of Broadcom NICs:&lt;/p&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;HP 331T (~$10 - 12)&#xA;&lt;ul&gt;&#xA;&lt;li&gt;HP 332T - BCM5720 (&amp;lt;$10, some for ~$5)&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;Dell 0YGCV4 (not cheap, but included for completeness, ~$15)&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;blockquote&gt;&#xA;&lt;p&gt;Note that the HPE or Dell OEM NICs (especially HPE) you may have to tape off the SMBus pins (B5 and B6, lay the card flat with the rear/ports pointed to your left and go right four, don&amp;rsquo;t tape the back pins off, label tape, kapton tape, electrical tape will work fine) on the PCI-E finger of the card to get it to boot in things that are not a server. If you do not do this, you may encounter memory errors including failure to POST or missing memory channels.&lt;/p&gt;</description>
    </item>
    <item>
      <title>IBM V7000 Expansion Module 2076-224 24SFF SAS2 disk shelves (Xyratex EB-2425-E6EBD)</title>
      <link>https://wporter.org/ibm-v7000-expansion-module-2076-224-24sff-sas2-disk-shelves-xyratex-eb-2425-e6ebd/</link>
      <pubDate>Wed, 14 Jan 2026 20:45:00 +0000</pubDate>
      <guid>https://wporter.org/ibm-v7000-expansion-module-2076-224-24sff-sas2-disk-shelves-xyratex-eb-2425-e6ebd/</guid>
      <description>&lt;p&gt;I recently decommissioned an IBM V7000 SAN (dual controller, each is commodity Xeon, probably LGA 1366, running what is, as best as I can tell, Linux with a Java webapp for management). Good riddance.&lt;/p&gt;&#xA;&lt;p&gt;I had a feeling that the disk shelves were just SAS, so I grabbed them for some science. Well, turns out, the Gen 1 IBM V7000 expansion modules are perfectly normal SAS2 disk shelves. Nothing special about them. They are white-labeled Xyratex EB-2425-E6EBD enclosures (as sold by Pure, NetApp, Dell, HPE, Cray, Seagate.. so, if you need parts, they&amp;rsquo;re very easy to find) and pretty quiet for 24x 600gb 10K spinnies per shelf. I hooked them up to a Dell H200E (LSI 2008) HBA and the disks came right up.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Quick thoughts on &amp; power consumption of Supermicro 6028 with X10DRU-TR4T&#43;</title>
      <link>https://wporter.org/quick-thoughts-on-power-consumption-of-supermicro-6028-with-x10dru-tr4t-/</link>
      <pubDate>Mon, 23 Jun 2025 15:30:00 +0000</pubDate>
      <guid>https://wporter.org/quick-thoughts-on-power-consumption-of-supermicro-6028-with-x10dru-tr4t-/</guid>
      <description>&lt;p&gt;I got a pair of these for storage servers. Still haven&amp;rsquo;t used them for more than some quick tinkering yet. Bit of a waste.&lt;/p&gt;&#xA;&lt;h2 id=&#34;some-findings&#34;&gt;Some findings&lt;/h2&gt;&#xA;&lt;p&gt;You can drop a peripheral (Molex) to SATA power cable and SFF 8087 to SATA breakout cable in these to shove some SATA drives in the back of the box. I&amp;rsquo;ve done this with both of mine.&lt;/p&gt;&#xA;&lt;p&gt;By combining this with some U.2 risers you could feasibly fit eight internal SATA drives, twelve spinners in the front, and twelve more NVMEs in PCI-E slots. Since the system has a quartet of built-in Intel X540 10gbe NICs, you could actually consider using &lt;em&gt;all&lt;/em&gt; of the remaining PCI-E for drives.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Dell Latitude 7320 Detachable (PC tablet) review - hardware, Windows 11, Fedora 42</title>
      <link>https://wporter.org/dell-latitude-7320-detachable-pc-tablet-review-hardware-windows-11-fedora-42/</link>
      <pubDate>Sun, 22 Jun 2025 14:30:00 +0000</pubDate>
      <guid>https://wporter.org/dell-latitude-7320-detachable-pc-tablet-review-hardware-windows-11-fedora-42/</guid>
      <description>&lt;h2 id=&#34;table-of-contents&#34;&gt;Table of Contents&lt;/h2&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;Introduction&lt;/li&gt;&#xA;&lt;li&gt;Hardware&#xA;&lt;ul&gt;&#xA;&lt;li&gt;Display&lt;/li&gt;&#xA;&lt;li&gt;Build&lt;/li&gt;&#xA;&lt;li&gt;Peripherals&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;/li&gt;&#xA;&lt;li&gt;Performance&lt;/li&gt;&#xA;&lt;li&gt;Windows (11 24H2 LTSC)&lt;/li&gt;&#xA;&lt;li&gt;Linux (Fedora 42, with GNOME)&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;h2 id=&#34;introduction&#34;&gt;Introduction&lt;/h2&gt;&#xA;&lt;p&gt;I&amp;rsquo;ve always liked the idea of a modern Windows tablet. It can double as a desktop, connected to a Thunderbolt dock, do the job of a laptop, with a (hopefully) decent keyboard and a kickstand, and do the job of a tablet - either for entertainment, or for a portable drawing pad. It would be perfect! See, I really don&amp;rsquo;t want to throw a locked-down iPad into my mix of devices, and I really don&amp;rsquo;t want a buggy, slow Android tablet, either. And if you don&amp;rsquo;t like Windows, GNOME provides a pretty decent tablet interface, too!&lt;/p&gt;</description>
    </item>
    <item>
      <title>Poking at the Intel Management Engine (vPro Enterprise, AMT) for IPMI-like remote access on a M920q</title>
      <link>https://wporter.org/poking-at-the-intel-management-engine-vpro-enterprise-amt-for-ipmi-like-remote-access-on-a-m920q/</link>
      <pubDate>Wed, 23 Apr 2025 18:30:00 +0000</pubDate>
      <guid>https://wporter.org/poking-at-the-intel-management-engine-vpro-enterprise-amt-for-ipmi-like-remote-access-on-a-m920q/</guid>
      <description>&lt;p&gt;I&amp;rsquo;m sure that if you&amp;rsquo;re reading this post on my backwards sysadmin blog, you&amp;rsquo;re at least vaguely familiar with the Intel Management Engine. For shoots and giggles, let&amp;rsquo;s chat about it anyway.&lt;/p&gt;&#xA;&lt;p&gt;Essentially, it&amp;rsquo;s a tiny embedded computer (like your good old ASpeed BMC) that&amp;rsquo;s tightly integrated with almost any recent (2008+) Intel Platform Controller Hubs (PCHs - the chipset on a modern Intel board) or on the SoC itself (for Atom mobile processors - these have an integrated Security Engine). It has the potential to give you full remote access to the system at a hardware level (potentially far greater access than a separate BMC). The ME runs its own operating system (MINIX, in this generation, though this changes every so often). It runs its own applications that do.. stuff - its full scope of access/functionality is not documented, but extra fun features have been used by malware in the past.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Maxxing out my P1 Gen 4 with an RTX A5000 16gb and an i9-11950H (for science)</title>
      <link>https://wporter.org/maxxing-out-my-p1-gen-4-with-an-rtx-a5000-16gb-and-an-i9-11950h-for-science/</link>
      <pubDate>Sat, 05 Apr 2025 12:12:59 +0000</pubDate>
      <guid>https://wporter.org/maxxing-out-my-p1-gen-4-with-an-rtx-a5000-16gb-and-an-i9-11950h-for-science/</guid>
      <description>&lt;h2 id=&#34;update&#34;&gt;Update&lt;/h2&gt;&#xA;&lt;p&gt;That wonderful 16g of VRAM was on its way out. Back to the baby GPU for me.&lt;/p&gt;&#xA;&lt;h2 id=&#34;introduction&#34;&gt;Introduction&lt;/h2&gt;&#xA;&lt;p&gt;My P1 Gen 4 is just about my favorite laptop. It&amp;rsquo;s got a big, bright, reasonably high resolution 16&amp;quot; screen (2560x1600, 400 nits, 100% sRGB, 60hz.. more on this Soon™ - there are some real nice 4K 120hz 16&amp;quot; panels for cheap), no number pad, lots of memory, a nice glass trackpad, and it has a TrackPoint.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
