Looking for someone to run a benchmark test for me on a 1-4x 980 Pros


Limp Gawd
Jul 3, 2017
I have a prosumer/hobbyist workload that I could use some assistance running the built in benchmark against (can be done at no cost with a trial license and software available on Windows, Linux, and MacOS; although Linux is the ideal choice). I'm hoping someone on here can help me. If anyone has a 905P, I'm interested in how that performs as well.

The workload in question is PixInsight. This is a astrophotography tool that you can get a free license for. Linux is the preferred OS, but Windows can work. There's a built in benchmark that is short (too short IMHO); it should take <1min to run, whereas real processing steps can often take tens of minutes or more to execute. Results are time and a score. The benchmark summary can be found here: https://www.pixinsight.com/benchmark/

Now, while there are lots of scores, there's no real consistency in documenting the storage subsystem or RAM drive usage. That makes it next to impossible to draw any reasonable conclusions about what the ideal setup is. That's why I'm asking for help.

PI uses an internal metadata/intermediate processing "swap" file. This is not the OS swap. It's easy to configure, but it's based on parallel IO and thus you'd need to create multiple threads. The general rule is one thread per physical CPU core with what appears to be good scaling up to 16+ cores. Those can be pointed to a single drive or balanced across multiple. RAM disks using tmpfs on Linux are common, and faster in general, but non-volatile storage can be an issue. Plus, it limits the dataset size since you're carving out RAM.

What's really driving this question is, I really need to update my rig for this workload. I'm hoping for Zen3 TR 5000 early next year with a ton of RAM. However, beyond having a lot of cores/RAM, optimizing the storage for this tool is a big deal. However, the latest drives like the 980 Pro aren't really Pro models any more; it's basically an EVO and if you have workstation like workloads, it's going to get full, run out of cache, and get rather slow. Latency tends to be an issue for JVM applications as well, something these TLC drives tend to be terrible at. Intel might have a prosumer Optane 2.0 SSD for Rocket Lakes release in March 2021 (rumor). So options:

1) Use 1-4x 980 Pros
2) Use 1-4x 970 Pros (if the 980 Evo Pro flaws harm this use case)
3) Use an older 905P (maybe two smaller versions)
4) Wait for Optane 2.0
5) Find a used enterprise SSD

Any help is appreciated. Thanks!