TechFuel HQ

I added direct downloads to the TechFuelHQ dataset hub: CSV measurements, raw evidence manifests, and reproduction files. Each dataset links its method and license. Price files are dated observations, not live quotes.

Home Assistant on Proxmox: make it a VM, not an LXC. HA doesn't document HAOS in a container, and the container install has no Supervisor or Apps. No ISO: HAOS ships as a qcow2 disk image you import into an empty UEFI VM. Documented minimum is 2 vCPU and 2 GB RAM. techfuelhq.com/tutorials/ho...

Video calls stutter whenever someone uploads? Usually that's bufferbloat, and it's fixable. Run a bufferbloat test first. If it grades badly, turn on SQM with CAKE set just under your line speed. What that costs on GL.iNet, MikroTik and UniFi routers: techfuelhq.com/articles/bes... #homelab

Proxmox hosts: Proxmox staff say the PwnEscape KVM escape fix is in proxmox-kernel-7.0.14-24-pve, and only the 7.0.14 series gets it. EOL releases like PVE 8 get nothing. Reboot to load it. Still on 8.x? My 8-to-9 preflight list: techfuelhq.com/tutorials/pr... #Proxmox

Plex Media Server before 1.43.3.10861: the Plex2Shell chain is being exploited, and CISA added its four CVEs to KEV on Oct 20. Update first. Eyeing Jellyfin this week? What changes is cost, transcoding, and remote access you set up yourself. techfuelhq.com/articles/jel... #selfhosted

Start self-hosting with Pi-hole. Every device gets ad blocking, on 512 MB of RAM. Then Uptime Kuma. techfuelhq.com/articles/sel... #selfhosted

Mini PC to first Proxmox VM: ~45 minutes. The map flags DD mode, the target disk, and the no-subscription repo. techfuelhq.com/articles/pro... #Proxmox

4 x 4 TB drives: RAID 5 gives 12 TB; RAID 6 gives 8 TB, before filesystem overhead. Compare your NAS layout. RAID isn't a backup. #Homelab techfuelhq.com/tools/nas-st...

Model weights are only part of the VRAM budget. Our calculator separates weights, KV cache and runtime overhead. Estimates, not guaranteed fit. techfuelhq.com/tools/llm-vr... #LocalLLM

If every TV and player direct-plays your files, a Plex or Jellyfin server needs no GPU. If not, an Intel N100 or N150 iGPU decodes HEVC 10-bit and tone-maps HDR. ASUS's NUC 14 Essential N150 barebone was $239.99 at Newegg on September 24. #selfhosted #homelab

SkatterBencher's 9800X3D at -40 with PBO, Scalar 10X, +200 MHz and EXPO I averaged 84.6°C at 157.0 W in OCCT SSE, against 87.0°C at 149.0 W stock, while clocking 240 MHz higher. The catch is Scalar 10X, which raises the allowed FIT budget tenfold.

The 7800X3D and 9800X3D both have 96MB of L3. The newer chip's gaming lead comes from Zen 5 IPC, higher clocks, and moving the cache die beneath the cores, which lets it hold those clocks. Same AM5 board after a BIOS update. #buildapc #PCGaming

A 9800X3D doesn't game faster on an X870E board. It's an eight-core, 120W chip and a decent B850 already supplies the power. Spend above that only on ports or lanes your build will actually use. #buildapc #PCGaming

A 16-core 9950X3D games on eight cores by design. AMD's driver parks the die without V-Cache while you play, which is how TechSpot's 12-game average landed it level with the 9800X3D. It was 77% faster in Cinebench multi-core. Games only? 9800X3D. #PCGaming #buildapc

Jellyfin remote access starts with two checks: can the TV run your VPN client, and are you behind CGNAT? My decision guide: techfuelhq.com/tutorials/je... #Jellyfin

How to tell what kind of stutter you have. Watch GPU usage when the frametime graph spikes. Pinned near 99% and still hitching points at shaders, VRAM or frame pacing. Dropping to 50-70% points at the CPU, RAM or a background thread stealing time. #pcgaming #gamingpc #pcbuild

How the 2026 CPU gaming tiers were scored. Tom's Hardware's 1080p table, an RTX 5090 Founders Edition at High and Ultra, the geometric mean of 17 games, normalized to the Ryzen 7 9850X3D at 100. The slowest of 30 chips, the Core i5-14400, still scores 58. #pcgaming #cpu #pcbuild

DDR4 vs DDR5 on Tom's Hardware's Aug 24, 2026 index update. 32GB of DDR4-3200 was $179, 32GB of DDR5-6000 was $402. The catch is the platform. New DDR4 kits top out near DDR4-3600 and AM4's CPU line ended with Zen 3. Good for AM4 owners, wrong for a build you plan to grow. #ddr5 #ssd #pcbuild

Resizable BAR greyed out? Run msinfo32 before you touch CSM. BIOS Mode has to read UEFI. On a Legacy/MBR install, disabling CSM (which ReBAR needs) can stop Windows from booting. A stale BIOS or VBIOS is the other usual cause. #pcgaming #gpu #pcbuild

No HAGS switch on a Radeon RX 6000 card is expected. AMD's HAGS release notes (20.5.1 Beta, 23.12.1, 24.1.1) name the RX 5600, RX 5700, RX 7600 and RX 7700 to 7900, and no RX 6000 card. Intel does not support it on Arc A-Series either. #pcgaming #windows11 #gpu

9800X3D or 9950X3D comes down to what the PC does besides games. TechSpot measured the 9950X3D 77% faster in Cinebench multi-core and essentially tied in its 12-game average. If the box only plays, the second die sits parked while you play. #pcgaming #amd #ryzen

I logged my ROG Astral RTX 5080 OC every five seconds through a 25-minute gaming session on June 28, 2026. 308 samples, zero throttle events. On my hand-tuned profile the core averaged 3,202 MHz at 261 W and 56 C, so the clocks it showed were the clocks it kept. #rtx5080 #pcgaming #pcbuild

Path-traced Doom: The Dark Ages at 1440p ran at 31 fps on the RTX 5070 Ti in Steven Walton's 52-game TechSpot test, 63% ahead of the RX 9070 XT and still short of playable. The same test found the 9070 XT level with the Ti or ahead with ray tracing on in Pragmata and Monster Hunter Wilds. #pcgaming

A self-hosting order that builds skills. Pi-hole first (512MB RAM, about three minutes with Docker Compose). Uptime Kuma second, so you know when it breaks. Then Jellyfin, Nextcloud, Immich and finally Home Assistant, whose recommended install is its own OS. #selfhosted #homelab #docker

The old warning that ZFS eats half the RAM on a Proxmox box is out of date. A fresh PVE 8.1+ install caps ARC at 10% of RAM (max 16 GiB) in /etc/modprobe.d/zfs.conf. An older box upgraded to 9 has no entry there, and ZFS 2.3+ then defaults to 62.5% of RAM. #proxmox #homelab #minipc

Desk-depth check before a 48-inch OLED monitor. At 4K, 48 inches works out to about 92 PPI, roughly a 24-inch 1080p monitor. At 24 to 30 inches away that size is overwhelming. At 36 to 48 inches it reads as the right size. techfuelhq.com/articles/aor... #oled #pcgaming #gamingmonitor

Ethan Sholly's 2026 Self-Host User Survey is open at selfh.st: 45 questions, including a new AI section. If you self-host, this is a chance to document how you actually use it. selfh.st/survey/2026-...

Plex/Jellyfin hardware starts with the client: Direct Play, codecs, tone mapping and stream count. techfuelhq.com/articles/bes...

A UPS self-test can pass while the shutdown chain is still untested. I added a blank record for transfer, guest shutdown and host power-off times. Follow your model's approved test procedure. An estimated runtime is not a measured reserve.

EQ14 or G3 Plus? Start with the network variant, drive slots and RAM support. Our comparison includes a configuration worksheet. techfuelhq.com/homelab/beel...

New RAM in or CMOS cleared, and the screen stays black? Wait. ASUS puts memory training at about 1 to 3 minutes and asks you not to force the machine off while it runs. I don't read debug lights or touch the power button until three minutes are up. After a memory change, a minute proves nothing.

Six RX 9070 and 9070 XT owners have filed coil whine reports in the census I keep. All six rated it 2 or 3 on a 0 to 4 scale. Four tested a frame cap, and all four said it changed the sound. No Sapphire Pulse 9070 XT on file yet. If you own one I want your report, silent ones most of all.

Four cores at 3.6 GHz doesn't tell you it's an N150. The N97 and N200 are also four-core parts that turbo about that high. Only the SKU name identifies it. techfuelhq.com/homelab/n150...

Rated endurance doesn't follow the tiers. The DRAM-less Lexar NM790 is rated 1,000 TBW at 1TB. Both Gen5 flagships in my table are rated 600. Tiered 14 drives on published vendor specs only. techfuelhq.com/articles/ssd...

Every forum answers "how much GPU sag is too much" with a feeling. I couldn't find one dataset that records how far real cards droop, in millimetres, against card length, mount and months installed. So I opened one. One row, one card, one ruler. Yours takes two minutes.

Everything on the site is measured or cited, never sponsored. The free tools live at techfuelhq.com/tools and the open datasets (coil whine, GPU sag, daily GPU and SSD street prices) at techfuelhq.com/data. Ask me a build or homelab question here. I answer them.

Every forum answers "how much GPU sag is too much" with a feeling. I couldn't find a dataset that records how far real cards droop, in millimetres, against card length, mount and months installed. So I opened one. One row, one card, one ruler. techfuelhq.com/data/gpu-sag-census/

Everything on the site is measured or cited, never sponsored. The free tools live at techfuelhq.com/tools and the open datasets (coil whine, GPU sag, daily GPU and SSD street prices) at techfuelhq.com/data. Ask me a build or homelab question here. I answer them.