SC99 — Dual-Socket 48-Core CFD HPC Node with 256 GB Memory and 22 TB Storage for ANSYS Fluent

The SC99 node is the first dual-socket configuration in the lineup, and it changes the balance in three ways at once. Two AMD EPYC 9255 processors give 48 cores and 96 threads with a 4.0 GHz all-core boost. Memory doubles to 256 GB. And storage jumps to six 3.75 TB SATA SSDs — around 22 TB of usable capacity, several times anything else here.

Two sockets also means two sets of memory controllers: 24 memory channels in total, roughly 1.2 TB/s of aggregate bandwidth, and 256 MB of L3 cache across the pair. For CFD, that combination of capacity and bandwidth is often more valuable than raw core count.

Specifications

Compared with our single-socket 48-core node, SC99 has the same core count but twice the memory, twice the memory bandwidth, and roughly six times the storage capacity. Per core, that's about 5.3 GB of RAM — a far healthier ratio than the high-core single-socket tiers, and the reason this node suits larger cases rather than just more of them.

What this node is built for

SC99 is the capacity-and-bandwidth tier. It suits:

• Larger meshes that don't fit in 128 GB — 256 GB roughly doubles the cell count you can hold, which is the hard limit no amount of extra cores can work around.

• Bandwidth-bound solves — 24 memory channels across two sockets is the widest memory path available here, and CFD solvers are often limited by exactly that.

• Heavy physics on large cases — multiphase, combustion, many transported scalars: cases that need both memory capacity and bandwidth, not just cores.

• Long transient campaigns with large output — 22 TB of storage holds months of time-step data, full DOE sweeps, or several large cases' complete results without constant housekeeping.

• Cases that scale well to around 48 ranks — enough parallel spread to be genuinely fast, without pushing past where most CFD problems scale.

Understanding the dual-socket layout

A two-socket machine is not simply a bigger one-socket machine. Each processor has its own memory controllers and its own local RAM, and a core reaching data attached to the other socket pays a latency penalty crossing the interconnect between them.

For Fluent this is manageable and worth knowing about. Partitioning the mesh so each socket works mostly on locally-held data — and pinning MPI ranks accordingly — keeps that cross-socket traffic low. Run without any NUMA awareness and you can lose some of the bandwidth advantage the second socket provides. We configure this for you as part of the setup, so in practice it's a detail we handle rather than one you have to manage.

Choosing SC99

Choose SC99 when:

• Your case needs more than 128 GB of memory, or

• Your solve is bandwidth-bound and would benefit from 24 memory channels, or

• You're generating far more result data than a 3.76 TB node can comfortably hold.

If your case fits in 128 GB and scales past 48 cores, one of our higher-core single-socket nodes will give you more parallel spread. If it's gated by single-thread speed, a high-clock 8-core tier remains the faster answer. SC99's advantage is capacity and memory throughput, not core count.

As a rough guide, Fluent needs about 1 GB of RAM per million cells for a typical steady single-phase case, so 256 GB opens up meshes roughly twice the size of what the 128 GB nodes can hold — and with 48 cores, those larger meshes still give each rank plenty of cells to work on. That balance is what this node is for.

Storage: capacity over peak speed

Six 3.75 TB SATA SSDs is a different trade-off from the two NVMe drives in our other nodes. Per-drive throughput is lower than NVMe, but total capacity is roughly six times larger, and six drives working in parallel spread the load well for the write patterns CFD actually produces — periodic autosaves and time-step output rather than sustained peak-rate streaming.

For a long transient run writing thousands of time steps, or a large DOE sweep where every case's full result set needs keeping, capacity is usually the constraint that bites first. SC99 is sized for that.

How parallel solving works on this node

When you run Fluent in parallel, the solver partitions your mesh across all 48 cores (using algorithms such as METIS), each core solving its share of cells and exchanging boundary data every iteration via MPI. On a dual-socket node, most of that communication still stays on-board rather than crossing a network — a real advantage over two separate machines — while each socket's 12 memory channels feed its own cores from local memory. Sensible partitioning and rank placement keep cross-socket traffic to a minimum, which is what lets the doubled bandwidth actually show up in your solve times.

Why rent instead of buy

How it works — you're in full control

SC99 is yours to drive directly. You connect to the node over a secure remote desktop connection and run everything yourself, exactly as you would on a local workstation:

1. Connect via remote desktop — you get full remote access to the SC99 node.

2. Set up your simulation — load your geometry and case, configure your solver settings, and run ANSYS Fluent with all 48 cores at your command.

3. Run and monitor in real time — start, pause, adjust, and watch the solution as it progresses, with complete control throughout.

4. Keep your results — post-process on the node or download your solution and data files whenever you're ready.

You have full authority over the node for the duration of your rental: we provide the hardware and the support, and you run the simulations your way.

Frequently asked questions

How is SC99 different from a single-socket 48-core node? Same core count, but twice the memory (256 GB), twice the memory bandwidth (24 channels), and around six times the storage. If your case fits in 128 GB and is compute-bound, the single-socket node is simpler and likely comparable. If you need capacity or bandwidth, SC99 is the better match.

Does the second socket slow anything down? Only if the job is set up without regard for it. Cores reaching memory attached to the other socket pay a latency penalty, so partitioning and rank placement matter. We configure this as part of your setup.

Is SATA storage a downgrade from NVMe? Per drive, yes — but there are six of them and roughly six times the total capacity. For CFD write patterns, which are periodic rather than sustained at peak rate, capacity is usually the binding constraint. If your workflow genuinely needs peak sequential throughput, tell us and we'll point you at an NVMe tier.

How large a mesh can it handle? With 256 GB of RAM, plan for meshes roughly double what a 128 GB node holds — comfortably into the hundreds of millions of cells for simpler steady single-phase cases, fewer for transient or multiphase physics (about 1 GB of RAM per million cells is a good starting estimate). Send us your case for a precise answer.

Is my data kept confidential? Yes.

Get started with SC99

If your cases have outgrown 128 GB, or you're bandwidth-bound, or you simply need somewhere to keep tens of terabytes of results, SC99 is built for exactly that.

Contact MR CFD to run your ANSYS Fluent case on SC99 — and we'll help you choose between capacity, bandwidth, and core count for your simulation.