Synology PAS7700 Active-Active NVMe Enterprise Storage

Price on request · Talk to a Sirap engineer

The enterprise end of Synology

The PAS7700 sits at the very top of the Synology line and behaves nothing like the DiskStation or RackStation units most people picture when they hear the brand. It is a 4U, all-NVMe array with two controllers that both serve data at the same time, designed for workloads where downtime is unacceptable and where the storage has to keep pace with modern GPU servers and dense virtualisation. If a typical Synology NAS is shared storage for a business, the PAS7700 is infrastructure for a data centre. It runs DSM Enterprise, so the management experience is familiar, but the engineering underneath is a different class of system.

How the PAS7700 is different from a normal Synology NAS

Four design choices separate the PAS7700 from the rest of the range, and each one matters for a specific reason.

Two active-active controllers, not one. A standard DiskStation or RackStation has a single controller: if it fails, the unit is offline until it is fixed. Synology’s SA-series high-availability units add a second controller, but in active-passive mode, where the second one waits idle and takes over only on failure. The PAS7700 runs both controllers active-active, so they share the workload during normal operation and either one can carry the full load if the other fails. The result is higher sustained performance and failover with no service interruption, which is what enterprise and round-the-clock workloads require. Hosts reach both controllers over multiple paths (Multipath I/O, using ALUA for LUNs and ANA for NVMe namespaces), so if a controller, port or cable fails the host simply carries on over the surviving path. That is what makes failover transparent to the application rather than merely fast.

All-NVMe, not SATA or hard disks. Every one of its drive bays takes a U.3 NVMe SSD connected over PCIe, not the SATA interface used across the consumer and SMB range. NVMe removes the protocol overhead that limits SATA flash, which is why the array can sustain millions of IOPS and tens of gigabytes per second rather than the hundreds of megabytes a typical NAS delivers.

Block storage over NVMe-over-Fabrics, not just file shares. Ordinary Synology units serve files over SMB and NFS. The PAS7700 also presents block storage over NVMe-oF, letting servers treat the array as if the flash were local while it sits across the network. For databases and virtualisation, this combination of low latency and shared capacity is the point.

A direct path to the GPU. The array supports a data path that moves data from the SSDs to GPU memory without routing every byte through the host CPU and its memory first. For AI and machine-learning training, where GPUs are the expensive resource and must be kept fed, removing that CPU bottleneck is the difference between GPUs that are working and GPUs that are waiting on storage.

What it is for: USPs and main use cases

The PAS7700 is bought to solve problems that defeat a conventional NAS. Its core strengths:

• No single point of failure: active-active controllers, plus redundant power and cooling, mean a component can fail and the service keeps running. This is for workloads that genuinely cannot stop.
• Performance at scale: in the region of two million IOPS and around 30 GB/s of throughput, enough to serve hundreds of virtual machines or a busy analytics database without the storage becoming the bottleneck.
• Built for GPU and AI pipelines: NVMe-oF and the direct-to-GPU path keep expensive accelerators supplied with training data instead of stalling on slow storage.
• Grows to petabyte scale: 48 NVMe bays in the head unit, expandable to 216 bays across expansion units, with a single volume that scales into the petabytes given sufficient controller memory.
• Familiar DSM management: the same operating system and tooling as the rest of the Synology range, so snapshots, replication and integration are not a separate skill set.

The main use cases follow directly: AI and machine-learning training storage, large-scale server virtualisation and virtual desktop infrastructure, high-transaction databases and real-time analytics, and media or research workloads that need very high sustained throughput with continuous availability.

Deployment examples

Feeding a GPU cluster for AI training

An organisation training models on a rack of GPU servers has a storage problem that ordinary NAS units cannot solve: the GPUs consume data far faster than file-share storage can deliver it, so the accelerators sit idle waiting on reads. A PAS7700 connected to the GPU servers over 100GbE, presenting storage with NVMe-oF and using the direct-to-GPU data path, keeps the training data flowing without the host CPU becoming a chokepoint. The expensive part of the cluster, the GPUs, stays busy, which is the entire economic justification for the array.

Consolidating a large virtualisation estate

A business running hundreds of virtual machines needs storage that is both fast under random load and continuously available, because an outage takes every VM down at once. The PAS7700’s active-active controllers mean a controller fault does not interrupt service, while all-NVMe performance handles the random IO of many VMs without the latency spikes a hybrid array would suffer. One array can replace a rack of slower storage and remove the single-controller risk at the same time.

A high-transaction database or real-time analytics platform

Transactional databases and analytics engines live or die on storage latency and consistency. Running them on the PAS7700 gives them NVMe latency on shared, redundant capacity, so query and transaction performance stays predictable as load rises, and a hardware fault does not become a database outage. For workloads where every second of downtime has a cost, the no-single-point-of-failure design is as important as the speed.

One design detail to plan around: capacity is split per controller

One characteristic of the dual-controller design is worth understanding before you size the array, because it catches out people who expect a single large pool. The PAS7700’s total capacity is divided equally between its two controllers and is not shared across them: each controller owns its own storage pool and serves its own volumes. In practice a volume has to fit within one controller’s share of the capacity rather than the combined total, so sizing is done per controller, not as one undivided number. This is not a flaw so much as a consequence of how active-active redundancy works, but it does change how capacity is planned, and it is exactly the kind of thing we work through before quoting rather than after delivery.

Technical specifications

Specifications confirmed against the Synology PAS7700 datasheet. Synology recommends configuring both controllers identically, and up to 512 GB of memory per controller for demanding, high-availability deployments, so that performance and failover behaviour stay consistent. Sirap sizes memory and capacity to your workload as part of the quote.

Download the full datasheet · See all Synology storage

What is in scope from Sirap

An array of this class is not a self-install. As Malta’s only Synology Platinum Partner, Sirap delivers the PAS7700 as an engineered solution, and for this unit specifically that means:

• Architecture and sizing: we work out usable capacity against raw (Synology’s guidance reserves headroom for snapshots, file-system metadata and rebuild operations on top of RAID 6 parity), the per-controller memory your workload needs (up to 512 GB per controller for demanding deployments), and the network design, all against your actual GPU, virtual-machine or database targets.
• Network and fabric design: planning the 100GbE and NVMe-oF connectivity to your servers so the array’s performance reaches the hosts rather than being throttled at the network.
• High-availability commissioning: configuring and testing active-active failover so you have seen it work before it is ever needed in production.
• Data protection design: snapshots, replication and a backup strategy to a separate target, because even a no-single-point-of-failure array still needs a backup that is independent of it.
• Local enterprise support: specified, deployed and supported by engineers based in Malta, with Platinum-backed service levels rather than a distant support queue.

Often paired with

• 100GbE switching and host adapters: to carry NVMe-oF traffic between the array and your GPU or virtualisation hosts at full speed.
• A separate Synology backup target: a RackStation or second array holding independent, ideally immutable copies, so a logical disaster or ransomware event cannot take out production and backup together.
• Enterprise Eaton UPS and power management: clean, monitored power for a system that is expected to run continuously. Sirap has been an Eaton power partner since 1988.

Could this be grant-funded?

An investment in AI infrastructure or high-availability data systems can align well with Malta Enterprise schemes aimed at digital and productive investment. Whether the Smart & Sustainable Investment Grant or another instrument is the right fit depends on the project and the business. Read our funding guide and see if your project qualifies, or talk to a Sirap engineer about structuring the case.

When a different Synology is the right call

The PAS7700 is deliberate overkill for most businesses, and we will say so when it is. If you need shared storage for a team, file serving, backup or a handful of virtual machines, a RackStation or DiskStation does the job at a fraction of the cost and complexity. If you need continuous availability but not extreme performance, the active-passive SA-series high-availability units are the more proportionate choice. The PAS7700 earns its cost only when you genuinely need all three of its defining traits together: no single point of failure, NVMe-class performance, and scale into the petabytes, typically driven by GPU and AI pipelines, dense virtualisation or large databases. Tell us the workload and we will recommend the right tier honestly, even when it is a smaller box.

Frequently asked questions

How is active-active different from the SA-series high-availability units?
The SA-series runs two controllers in active-passive mode: one works while the other waits as a standby and takes over only on failure. The PAS7700 runs both controllers active-active, so both serve data during normal operation and either can carry the full load if the other fails. You get more usable performance day to day and failover without a service interruption, rather than a brief switchover.

What does the direct-to-GPU data path actually do?
It lets data move from the array’s NVMe SSDs into GPU memory without being copied through the host server’s CPU and main memory on the way. In AI training, the CPU and its memory bandwidth are often the bottleneck that leaves expensive GPUs waiting on data. Removing that hop keeps the GPUs supplied and the training running at full rate.

Do we need NVMe-over-Fabrics and 100GbE to use it?
Not necessarily, but to reach the array’s full performance you want a fast fabric. NVMe-oF over 100GbE is what lets remote servers use the storage almost as if it were local NVMe. The exact network design depends on your hosts and workload, which is part of what we size during planning.

Does it run the same DSM as other Synology units?
Yes & no, it runs DSM Enterprise which is DSM with the enterprise feature set, so snapshots, replication, monitoring and management follow the same model as the rest of the range. The hardware is a different class, but the software and the skills to run it are familiar.

Is the array a backup on its own?
No. Redundant controllers protect against hardware failure, not against ransomware, accidental deletion or a logical fault, all of which would affect the live data on the array. A genuine strategy still needs an independent, separate backup, ideally with immutable copies. We design that alongside the deployment.

How much usable capacity will we actually get?
Less than the raw drive total, by design, as with any enterprise array. Some capacity goes to RAID 6 parity (two drives’ worth of protection), and Synology’s guidance also reserves space for snapshots and file-system metadata plus roughly 20% free headroom so performance and rebuilds stay healthy. On top of that, the dual-controller architecture splits capacity between the two controllers. We calculate real usable capacity for your exact configuration up front, so the number you plan around is the number you get.

How far does it scale?
The head unit holds 48 NVMe drives and expands to 216 bays across expansion units. A single volume reaches 200 TB at base controller memory and scales into the petabytes as you add memory, using RAID 6 or triple-parity protection. We size this to your capacity and growth plan.

Talk to a Sirap engineer

Tell us about the workload, the servers it has to feed and your availability targets, and we will tell you honestly whether the PAS7700 is the right platform or whether a more proportionate Synology would serve you better. Specified and supported here in Malta.

Request a quote or a consultation · Price on request