SC103 — 128-Core CFD HPC Node with 512 MB Cache and 12-Channel Memory for ANSYS Fluent

The SC103 node puts a small cluster's worth of cores into a single socket. It runs one AMD EPYC 9755 — 128 cores and 256 threads, all of them at 4.1 GHz — with 512 MB of L3 cache and twelve DDR5 memory channels behind it. Paired with 128 GB of RAM and 2 × 1.88 TB NVMe SSD, it delivers more total compute per hour than any other node here, with no network between ranks.

Worth noting up front: on this chip, AMD's single-core boost and all-core boost are the same 4.1 GHz. There's no drop-off when every core is loaded — 128 cores under a full parallel job run at the same clock a single core does on its own.

Specifications

512 MB of L3 is the largest cache in the lineup by a wide margin, and 4.1 GHz across all 128 cores matches the all-core clock of our 48- and 96-core tiers — the extra cores carry no clock penalty. Twelve memory channels deliver roughly 600 GB/s of bandwidth.

Sizing note: read this before choosing SC103

SC103 carries 128 GB of RAM across 128 cores — roughly 1 GB per core, the tightest ratio here. Its 600 GB/s of memory bandwidth is also the same figure our 64-core node has, shared among twice as many cores. Both point to the same conclusion: this node is built to run many jobs at once, not one job as fast as possible.

For a single Fluent case, 128 MPI ranks is well past where most CFD problems scale usefully. As a rough guide, Fluent needs about 1 GB of RAM per million cells for a typical steady single-phase case, so a mesh that fits in 128 GB leaves each of 128 ranks with a small share of cells — the point at which exchanging boundary data between partitions starts to cost more than the solving itself. A 24- or 32-core node will often finish that same case sooner.

Where SC103 has no competition is concurrency: eight cases on 16 cores each, four on 32, or a DOE sweep run in parallel batches. Aggregate throughput per hour is unmatched; it's how that throughput is distributed that differs from the smaller tiers.

If you have one very large case rather than many, ask us before booking. Depending on the mesh, a 32- or 64-core tier may well solve it faster.

What this node is built for

SC103 is a maximum-capacity throughput tier. It suits:

• Running many cases concurrently — the clearest use of 128 cores: a full DOE sweep, a mesh-independence study, or a batch of design variants, several at a time on separate core groups.

• Teams sharing one node — several engineers running their own cases simultaneously without queuing behind each other.

• Very large cases that genuinely scale past 96 ranks — uncommon, but this is the node for them when they appear.

• Time-critical batch work — compressing a week of sequential runs into a day.

• Cluster-scale parallelism without a cluster — 128 ranks on one socket, no network hop, no MPI-over-network tuning.

• Heavy I/O batch runs — NVMe storage handles many concurrent jobs writing data files and autosaves.

Choosing SC103

Choose SC103 when:

• You have many cases to run rather than one, or

• You have a genuinely enormous case that's been shown to scale past 96 ranks and fits within 128 GB.

For a single moderate case, almost any smaller tier will serve you better — that's the honest positioning of this node rather than a caveat buried at the end. Fluent scales well only while each core has enough cells to stay busy relative to the data it exchanges with its neighbors, and 128 ranks pushes past that for most CFD work. The value here is aggregate throughput.

Tell us how many cases you're running and how large they are, and we'll tell you whether SC103 or a smaller tier gets you results faster.

Fast NVMe storage

The two 1.88 TB NVMe SSDs give substantially higher read/write throughput than SATA SSDs, plus nearly 3.76 TB of combined capacity. On a node running many jobs at once that matters more than usual: several cases writing data files and autosaves at the same time would saturate slower storage, and the combined capacity holds a full sweep's output without constant clearing.

How parallel solving works on this node

When you run Fluent in parallel, the solver partitions your mesh across the cores you assign it (using algorithms such as METIS), each core solving its share of cells and exchanging boundary data each iteration via MPI. On SC103 you'll often want to assign a subset — 16 or 32 cores per case — and run several cases side by side rather than spreading one across all 128. When you do use the full core count on a single case, the single-socket layout is a genuine advantage: MPI traffic stays on fast internal channels instead of crossing a network, keeping overhead lower than the same rank count spread across separate machines.

Why rent instead of buy

A 128-core server is a large purchase that most CFD teams would leave underused between projects. Renting it for the batch weeks and dropping back to a smaller tier the rest of the time matches how this work actually arrives.

How it works — you're in full control

SC103 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 SC103 node.

2. Set up your simulations — load your geometry and cases, configure solver settings, and run ANSYS Fluent with all 128 cores at your command, on one case or several.

3. Run and monitor in real time — start, pause, adjust, and watch your solutions as they progress, 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 SC103 different from our 64- or 96-core nodes? More cores and considerably more cache, at the same 4.1 GHz all-core clock and the same total memory bandwidth. Since that bandwidth and the 128 GB of RAM are shared among more cores, SC103's edge is running more jobs at once rather than running one job faster.

Will 128 cores make my single case faster? Usually not. Most CFD cases stop scaling well long before 128 ranks, as communication between partitions starts to outweigh the solving work. Send us your case and we'll check — often a 32- or 64-core node is the faster answer for one case.

So what is this node actually for? Volume. A DOE sweep, a batch of design variants, a mesh-independence study, or a team sharing one machine — several cases running concurrently on separate core groups, which is where 128 cores delivers throughput nothing else here matches.

Is 128 GB enough memory for 128 cores? For running multiple moderate cases concurrently, yes. For one very large case, memory becomes the binding limit before core count does. If your mesh needs more than 128 GB, our high-memory nodes are the right tier.

Does the clock drop when all 128 cores are working? No — AMD rates this chip at 4.1 GHz for both single-core boost and all-core boost. The clock you see on one core is the clock you get on all of them.

Is my data kept confidential? Yes.

Get started with SC103

If you have a batch of cases rather than a single one — a sweep, a study, or a team's worth of concurrent runs — SC103 gives you more total throughput than any other node here.

Contact MR CFD to run your ANSYS Fluent cases on SC103 — and we'll help you decide whether 128 cores or a smaller, faster-per-core tier gets your results sooner.