If your home server needs iGPU hardware transcoding for Plex or Jellyfin, buy the AMD Ryzen 5 5600G. Its integrated Vega 7 delivers usable Jellyfin hardware acceleration via VAAPI. If you're skipping transcoding entirely — direct-play only, containers, compile jobs, self-hosted tools — pick the Ryzen 7 5700X for two extra cores. That's the whole decision.
The no-discrete-GPU server case and why the iGPU changes the math
Home servers are a build class where "buy fewer parts" often beats "buy the more powerful part". A dedicated GPU adds cost, heat, idle power, PCIe slots you now have to think about, and a card that spends 99% of its life waiting for a Docker container to poll it. That's why the iGPU-vs-more-cores question is the real fork for AM4 home-server builders in 2026.
On paper the Ryzen 7 5700X and Ryzen 5 5600G look like tiers of the same product. In practice they're different chips solving different problems. The 5600G is a Cezanne-generation APU with a real integrated GPU (Vega 7) that can encode and decode H.264/HEVC/AV1 via VAAPI. The 5700X is a Vermeer-generation CPU with 8 cores / 16 threads and no display output at all — you either add a GPU or run headless. Per Tom's Hardware's CPU hierarchy, the 5700X is a stronger multi-threaded chip; per AMD's Ryzen product portal, the 5600G is the one with the iGPU.
The decision comes down to a single question: does your server transcode video?
Key takeaways
- 5600G wins if transcoding matters. Vega 7 iGPU handles 2-4 simultaneous Plex/Jellyfin transcodes.
- 5700X wins if it doesn't. Two extra cores for Docker, compile jobs, LLM CPU inference, and container density.
- Idle power essentially ties. Both sit near 25-35 W idle on a bare server board.
- 5600G is PCIe 3.0 only (not 4.0), which matters if you plan multiple NVMe drives.
- Neither needs a big cooler. Any $30-60 tower cooler handles both quietly.
Step 0 diagnostic: does your server need iGPU transcoding at all?
Answer this honestly before you spend money:
- Plex or Jellyfin with mostly direct-play clients (Apple TV 4K, Nvidia Shield, modern TVs, native Jellyfin apps) — you rarely need transcoding. iGPU is optional.
- Plex/Jellyfin with older clients (Chromecast Gen 2, browser tabs, legacy smart TVs) — you'll transcode frequently. iGPU is valuable.
- Off-network streaming (Plex Remote, Jellyfin over VPN with bandwidth caps) — you'll transcode to lower bitrates. iGPU is very valuable.
- No media server; you're running containers, compile jobs, Home Assistant, Nextcloud — you don't need iGPU. More cores help.
If two of your three answers point to "no transcoding", buy the 5700X. If two point to "yes transcoding" and you don't own a modern GPU, buy the 5600G.
5-column spec-delta table
| Chip | Cores/Threads | iGPU | TDP | Idle power (typical) |
|---|---|---|---|---|
| Ryzen 5 5600G | 6C / 12T | Vega 7 (7 CUs) | 65 W | ~26 W wall |
| Ryzen 7 5700X | 8C / 16T | None | 65 W | ~28 W wall |
| Ryzen 5 5600 (non-G) | 6C / 12T | None | 65 W | ~28 W wall |
| Ryzen 7 5800X | 8C / 16T | None | 105 W | ~34 W wall |
Note: PCIe generation matters if you plan multiple NVMe SSDs. The 5600G runs PCIe 3.0; the 5700X runs PCIe 4.0. For a single boot NVMe like the Samsung 970 EVO Plus, you won't notice. For two NVMe drives plus a networking card, the 5700X's PCIe 4.0 lanes give more usable bandwidth.
Ryzen 5 5600G: the built-in Vega iGPU advantage for a headless box
The 5600G is basically a headless-server chip that also happens to boot without a discrete GPU. That matters more than it sounds: you can build a case-fanless mini-ITX server in a small case, no video card, no GPU-side heat, no PCIe budget spent, and still transcode video for the household.
Jellyfin and Plex both support VAAPI hardware acceleration on the Vega iGPU. In practice: 2-4 simultaneous 1080p H.264 → 720p transcodes work fine, and 1080p HEVC decode is comfortable. 4K HDR-to-SDR tone-mapping is the workload where the Vega 7 starts showing its age; it works but with visible CPU spike. If your library is mostly 1080p HEVC or below, the 5600G handles a household's transcode load quietly.
The catch: only 6 cores. If you also run Docker Compose stacks with 15+ containers, RAG indexing on a self-hosted Ollama, or heavy compile workloads, you'll notice the shortfall.
Ryzen 7 5700X: two extra cores for containers and compile jobs, but no iGPU
The 5700X is what you buy when your server does everything except transcode video. Two extra cores plus 4 extra threads at the same 65 W TDP means more Docker containers running comfortably, better compile throughput for anyone using this as a dev/build server, and headroom for occasional CPU-side LLM inference on small models.
Where it hurts: no video output. You either add a cheap GPU for boot/troubleshooting, or you run fully headless. Headless is fine 99% of the time; when your BIOS won't come up because a firmware update went sideways, you'll wish for HDMI-out. Some AM4 boards ship without proper serial console fallback either, which raises the recovery bar.
The 5700X also handles Jellyfin/Plex transcode fine via CPU-only x264/x265, but you'll burn all 8 cores for a single 4K → 1080p transcode. That's the difference between "iGPU does 3 transcodes at 15 W" and "CPU does 1 transcode at 55 W".
Benchmark table: Jellyfin/Plex transcode and container throughput
Approximate performance on a 5600G vs 5700X home server, Docker + Jellyfin + typical container stack:
| Workload | 5600G | 5700X |
|---|---|---|
| 1080p H.264 → 720p transcode (single) | ~40 fps (iGPU) | ~50 fps (CPU) |
| 1080p HEVC → 1080p H.264 transcode | ~35 fps (iGPU) | ~30 fps (CPU) |
| 4K HDR → 1080p SDR tone-map | ~15 fps (iGPU) | ~10 fps (CPU) |
| Concurrent 1080p transcodes | 3-4 | 2 (CPU-bound) |
| Docker container density (idle) | ~20 comfortable | ~30 comfortable |
| Nginx + Postgres + 3 web apps QPS | strong | stronger |
| Nightly repo compile (Linux kernel) | ~30% slower | baseline |
The iGPU-transcoding rows are where the 5600G decisively wins; the multi-threaded compile and container-density rows are where the 5700X earns its price.
Storage and boot: NVMe cache vs bulk SATA for a home server
A well-built home server pairs a fast NVMe boot/cache drive with bulk SATA storage. On this class of build:
- Boot + Docker + config: Samsung 970 EVO Plus 250 GB NVMe. Fast, cheap, reliable. Runs the OS and container config; database files go here for lower latency.
- Bulk media / backups: Crucial BX500 1 TB SATA SSD for smaller libraries, spinning-rust drives for larger. The BX500 is a fine budget SATA option for a "everything I actually watch fits in 1 TB" household.
- Cache tier (if you use Unraid or similar): a second small NVMe.
The 5700X's PCIe 4.0 support means the 970 EVO Plus runs at full bandwidth. On the 5600G, PCIe 3.0 caps the same drive at ~3,500 MB/s — plenty for any home-server workload but not the drive's peak.
Verdict matrix
- Get the 5600G if… you run Plex/Jellyfin as a primary use case, want to skip a dedicated GPU entirely, don't need more than 6 cores, and value the quieter/smaller thermal envelope.
- Get the 5700X if… you don't care about hardware transcode (direct-play only, or you already own a GPU with NVENC), you run dense container stacks, or you compile/build regularly.
Recommended pick
For most self-hosters starting a home server from scratch in 2026 with Plex or Jellyfin as a real use case, buy the Ryzen 5 5600G. The iGPU saves you a GPU purchase, saves the PCIe slot for something more useful, and handles a household's transcode load. Pair with a Samsung 970 EVO Plus 250 GB NVMe for the boot drive and a Crucial BX500 1 TB SATA SSD as bulk storage. If you know your transcode load is negligible and you want raw container/compile throughput, the Ryzen 7 5700X is the smarter buy for the same money.
Common pitfalls
- Buying the 5600G, then adding a discrete GPU. You paid for an iGPU you now won't use. Buy the 5700X instead.
- Buying the 5700X, then discovering you need transcode. You now need a GPU purchase you weren't planning on.
- Skimping on RAM. 32 GB is the sweet spot; 16 GB is tight once you start running RAG or database-heavy containers.
- Cheap PSU. Home servers run 24/7. Efficiency losses on a bad PSU cost more than the price gap to a better one over 3 years.
- Assuming the iGPU covers 4K HDR tone-mapping. It works but eats CPU; if that's your daily driver, a modern GPU with proper tone-map ASICs is worth the money.
Related guides
- Ryzen 7 5800X vs 5700X gaming
- Best cooler for Ryzen 7 5800X — Noctua NH-U12S vs ML240L
- Build a low-power Jellyfin server on a Raspberry Pi 4 8 GB
- Best NVMe SSD for local LLM model storage
- ARM64 vs AMD Ryzen for Linux local AI desktops
Citations and sources
- AMD Ryzen desktop product portal
- Jellyfin hardware acceleration documentation
- Tom's Hardware — CPU hierarchy
This piece is editorial synthesis based on publicly available information. No independent first-party benchmarking is reported.
FAQ
Should I use a Ryzen 5600G or 5700X for a home server? Buy the 5600G if you rely on Plex or Jellyfin hardware transcoding and don't want to add a discrete GPU. Buy the 5700X if you skip transcoding — direct-play only clients, containers, compile jobs — and value the two extra cores. That framing decides the build regardless of budget; both chips cost within $30 of each other at typical prices.
Does the 5700X really lack an iGPU entirely? Yes. It's a Vermeer-die Zen 3 chip with no integrated graphics. You either add a cheap GPU for boot and troubleshooting (any modern low-profile card works, even a used GT 710 for pure display-out), or you run fully headless with serial-console fallback via a compatible motherboard. Some AM4 boards handle headless boot gracefully; a few do not — check the manual before buying.
How well does the Vega iGPU handle 4K HDR tone-mapping? It works via VAAPI in both Jellyfin and Plex, but 4K HDR-to-SDR is the workload where the Vega 7 shows its age — expect 15-20 fps for a single 4K HDR → 1080p SDR transcode with tone-mapping, which is enough for playback but not much headroom. For heavy 4K HDR households, a modern discrete GPU (Intel Arc for AV1 encode, or NVIDIA for NVENC) is a real upgrade.
Can either chip run local LLM inference on the CPU? Both do, but slowly. A Ryzen 7 5700X on 8-core CPU inference lands around 5-10 tok/s on a 7B q4 model — usable for chat, not for coding-assistant workloads. The 5600G is closer to 4-7 tok/s. Neither is a substitute for a GPU-based LLM rig; both are fine if you occasionally want to run a small model as a background service.
What about the newer 5600GT or 5700G3D? The 5600GT is a mild refresh of the 5600G with slightly higher boost clocks and the same Vega iGPU — treat it as a "same guide applies" upgrade if you find it cheaper than a 5600G. AMD's Zen 3 X3D APU variants trade cache for iGPU strength; for a pure server they add little over the 5600G. Stick with the 5600G unless the GT variant is priced below it.
