Small and Short Simulations

If the simulation is small (doesn't need much RAM nor many threads) and short (will take less than 30 minutes to run), it is often easiest to run it in your own machine.

Small/Medium but Longer Simulations

If you're running a small/medium but long simulation (something that might run for a few hours or days), running it in your own machine is not practical. In this case, you should use the desktop in your office. The desktop machines typically have:

• 12 threads (1 Intel Core i5-12500T, with 6 cores, each with 2 threads).
• 16Gb RAM.

These are not huge machines, but they're not small either. They are under-utilised, most are always idle. In fact, the department turns on all of them automatically at 8AM on working days, and then turns off all the idle ones at 10AM.

You can access these machines remotely using the Remote Access Portal, or via SSH . You can also access other people's desktops, so you are free to use several at once. Here is a complete list of all the desktop machines. Using the same link, you can wake any of them remotely if they have been turned off.

Large Simulations

If your simulation needs more than 12 threads or 16Gb RAM, you should use the Mathematical Institute computing nodes. These machines have:

• 96 / 72 threads (2 Intel Xeon Gold 6342 / 6240, each with 24 / 18 cores, each with 2 threads).
• 2048Gb / 1535Gb / 768Gb RAM.

Sample performance plot.

These are big machines, but hugely oversubscribed. Here is a complete list, where you can also see in real time how much of each machine's capacity is currently in use. If you see a machine that is relatively idle, go ahead and run your jobs! However, if the machines are really busy, running simulations on your office desktop will be faster.

Sample performance plot.

Very Large Simulations

If your simulations are struggling to run in the Institute's computing nodes, and they are too large to run in a desktop, then you should use Vertigo. Vertigo is simular to the Institute's computing nodes, as it has:

• 96 (2 Intel Xeon Gold 6342, each with 24 cores, each with 2 threads).
• 2048Gb RAM.

However, the main difference is that Vertigo uses a queuing system. You tell it how many threads and how much RAM your simulation needs, and Vertigo reserves them for you. You are guaranteed that no other simulation will compete with yours for resources, which can often be a problem in the Institute's computing nodes. However, the price you pay is that your jobs might have to wait before they start, as resources may not be immediately available.