GPU and software readiness
Synology Drive AI Search has stricter hardware requirements than DSM Agent. Synology’s current documentation requires a supported storage system with a compatible GPU, and the current Universal Search 2.0 release notes expose semantic-indexing support on a narrow enterprise model set. This means a DiskStation can support DSM 7.4 and DSM Agent while still not qualifying for Drive AI Search.
Quick answer
Treat AI Search readiness as six independent gates
Confirm the exact Synology system, GPU compatibility, DSM/package versions, Universal Search semantic-indexing availability, memory headroom and eligible Team Folders. If any one of those gates fails, the correct result is not ready rather than “probably works.”
Live Amazon products
Synology systems relevant to this AI and DSM decision
Product cards are discovery aids only. They never override Synology's live applied-model lists, GPU compatibility list, DSM release notes or package requirements.
Technical decision
Buy hardware for a verified support path, not a hopeful experiment
GPU support on Synology is model-specific and vendor-supported combinations matter. Check the live graphics-card compatibility list immediately before purchase, then confirm that the current Universal Search release lists the storage system for semantic indexing. Unsupported GPUs or unlisted systems may boot, but they should not be treated as a supported AI Search deployment.
Interactive AI / DSM tool
AI Search Readiness Checker
Use this as a planning aid. It applies the inputs you provide, but final model support, GPU compatibility, package requirements, beta status and DSM behavior must be checked against Synology's current official documentation before purchase or production changes.
Deployment checklist
Four checks before a Synology AI rollout
Match the exact model
Do not infer support from a nearby model number, generation or shared CPU. Use Synology current current applied-model list.
Verify the DSM build
DSM Agent requires DSM 7.4 or later, while AI Search also depends on newer package versions and semantic indexing support.
Separate AI feature paths
DSM Agent, website AI Advisor, AI Console and Drive AI Search have different hardware, privacy and deployment requirements.
Pilot before rollout
Measure memory, indexing time, permissions and recovery implications on representative data before enabling AI features broadly.
Gate 1: exact storage-system support
Start with the current Universal Search and GPU support lists. A model that supports DSM 7.4 is not automatically an AI Search model, because semantic indexing is tied to the GPU-capable enterprise path. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gate 1: exact storage-system support adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Gate 2: a compatible GPU
AI Search needs a supported graphics card in a supported chassis. Use Synology’s compatibility list for the exact model and GPU combination rather than buying from generic PCIe specifications alone. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gate 2: a compatible gpu adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Gate 3: required package versions
The AI Search software chain has minimum versions. AI Console, Drive, Office, NVIDIA Runtime Library, Synology AI Runtime and Container Manager must meet the current documented baselines. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gate 3: required package versions adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Gate 4: Universal Search semantic indexing
The indexing engine must expose Semantic indexing. Universal Search 2.0 introduced the settings that let administrators choose Team Folders and monitor semantic-indexing progress. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gate 4: universal search semantic indexing adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Gate 5: memory headroom
AI models reserve meaningful system memory. AI Console specifications recommend leaving substantial RAM for the system and list separate memory consumption for embedding, OCR, speech-to-text and image-captioning models. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gate 5: memory headroom adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Gate 6: eligible data
Folder type and indexing scope matter. Hybrid Share folders are currently excluded from semantic indexing, and file-type support varies according to the enabled AI model. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gate 6: eligible data adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Current enterprise model reality
The present public support path is enterprise-oriented. FS3420, FS6420 and several 2026 xs-series RackStation models appear in the current Universal Search 2.0 semantic-indexing release notes reviewed for this Sprint. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on current enterprise model reality adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Why a DS920+ result can be confusing
DSM Agent and Drive AI Search have different requirements. DS920+ may appear on the DSM Agent beta list, yet that does not satisfy the dedicated GPU and Universal Search semantic-indexing requirements for Drive AI Search. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on why a ds920+ result can be confusing adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Memory planning by enabled model
Not every AI capability needs to be loaded. Synology lists about 2 GB for embedding, 2 GB for OCR, 3 GB for speech-to-text and 3 GB for image captioning, while also recommending system-memory reserve. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on memory planning by enabled model adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
GPU compatibility can change
Compatibility pages are living documents. Graphics-card support and applied models can change after a package update, so procurement should include a same-day compatibility check and documentation of the exact supported combination. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on gpu compatibility can change adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Pilot indexing before capacity planning
Indexing speed depends on more than file count. Content type, enabled models, storage performance, GPU capability and existing services all affect the time and resource impact of building semantic indexes. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on pilot indexing before capacity planning adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
A readiness checker should be conservative
False positives are expensive in enterprise hardware. This tool returns ready only when every selected gate aligns with the current documented path; anything else is a prompt to verify before purchase or deployment. Readiness has to be proven across hardware and software at the same time. A model name, a physically installed GPU or a current DSM release by itself cannot satisfy the supported path when Universal Search, AI Console and runtime dependencies have their own version requirements.
The section on a readiness checker should be conservative adds a separate planning question. For this hardware-readiness topic, procurement should follow the live compatibility list rather than generic PCIe assumptions. Record the exact chassis, GPU, memory level and package baseline before purchase so supportability can be demonstrated later instead of reconstructed after a failed deployment.
Questions people ask
AI Search hardware questions
Does Synology AI Search require a GPU?
Yes. Current Synology documentation requires a supported storage system with a compatible GPU.
Can a normal DiskStation Plus model use AI Search?
Do not assume so. The current public Universal Search 2.0 semantic-indexing release is limited to a narrow GPU-capable enterprise set reviewed for this Sprint.
Is DSM Agent compatibility enough for AI Search?
No. DSM Agent has its own broad applied-model list, while AI Search adds GPU, package and Universal Search semantic-indexing requirements.
How much memory should I plan?
Synology AI Console recommends reserving at least 16 GB for system memory in GPU-powered AI model scenarios and lists additional model-specific memory consumption.
How much memory does the embedding model use?
Synology lists approximately 2 GB for the embedding model in its AI Console specifications.
How much memory does OCR use?
Synology lists approximately 2 GB for the OCR model, with additional system and service memory still required.
How much memory does speech-to-text use?
The AI Console specification lists approximately 3 GB for the STT model.
How much memory does image captioning use?
The AI Console specification lists approximately 3 GB for image captioning.
Can I use any NVIDIA card that physically fits?
No. Synology directs users to its GPU compatibility list. Physical fit and PCIe electrical compatibility do not establish vendor support.
What is the safest purchase process?
Check the current AI Search tutorial, Universal Search release notes and GPU compatibility list for the exact system on the day you order hardware.
Official references and methodology
Verify the exact model, DSM build, package versions and support state
The readiness checker deliberately separates DSM Agent support from AI Search support. Current AI Search model confirmation follows Synology’s Universal Search 2.0 release notes and AI Search tutorial, while GPU purchases must be revalidated against the live compatibility list.
- Drive AI Search setup requirements
- Universal Search 2.0 release notes
- GPU compatibility list
- AI Console resource specifications
As an Amazon Associate, Cloudzat may earn from qualifying purchases. Prices, product availability, DSM builds, beta eligibility, package versions, GPU compatibility and AI feature requirements can change. Verify the exact Synology model and current official release notes before purchasing hardware or changing a production system.