add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube

Ryzen 7 2700X or go with ThreadRipper

whand

7 months ago

Not a gamer, this would be for processing astro-images involving the stacking and correlating 100's of images. One of the creators of the software (PixInsite) says the software is core hungry and will use all the ram it can get its hands on. So I don't need a killer video card, just thinking M 2.0 1TB Evo 970, 4TB WD Black, 32GB ram for the Ryzen 7 or 128GB for the ThreadRipper, water cooler for overclocking.

Comments

  • 7 months ago
  • 4 points

If the software is multi-core aware then Threadripper would be the way to go. If you're running it on Linux, and have the cash, I'd go for the full 32-core 2990WX. (On Windows, I might be tempted to stop at 16 cores, as Windows doesn't do NUMA architectures nearly as well as Linux.)

Note that if you do choose the 2700X you can still run 64GB memory, 4x16, you just won't be able to run it at top speed. This is likely a case of more memory beating faster memory, though.

  • 7 months ago
  • 2 points

Fellow Astronomer here, done plenty of image stacking with my 140mm TEC Refractor and Celestron 14" SCT. I went with Intel i9-9980XE with 64GB Ram. You can definitely jump in with the Threadripper here. 64GB RAM plenty. 16 Core Threadripper should do it. Most important thing, above the PC you want to build, is well collimated optics and a good EQ mount.

  • 7 months ago
  • 1 point

The optics and mount are the first link in the chain, easy to spend more money on them then the computer. This is something of a balancing act, since you can spend hours getting useful data and having to wait for good seeing which can be days or weeks, what else would the computer be used for without it being gross overkill. Plus there is the cost factor, Allan M if money were no object then your points are well taken and kschendel Linux is a blackhole for RAM the more you throw at it the happier it is and it loves more cores. But alas the money trough is only so deep.

[comment deleted]
[comment deleted]
[comment deleted]
  • 7 months ago
  • 2 points

Might be worth looking into just how much RAM is actually helpful...

If your workload can scale into using say, 400GB of RAM, and winds up hitting a RAM bandwidth bottleneck when it has to use scratch space instead of real RAM, then do all the cores wind up dropping to idle while waiting on the scratch dist?

It may not be easy to answer this question without research and testing of your intended use. If more than 128GB RAM can actually be used, and actually has a bit benefit, then you may want to look at building around an enterprise platform that supports registered memory.

There are examples of these image combining workloads, where having say, 512GB RAM, with a 6-12 core server CPU with relatively low clock speeds, winds up being faster than having 8-18+ fast workstation cores limited to 64-128GB RAM.

  • 7 months ago
  • 2 points

Another option would be using an Optane NMVe for that scratch space. Looks like Optane runs about 1/3 the price of standard memory (registered DDR4 sounds scary) while working well enough in most "huge" loads. Considering the cost of the threadripper, I'd assume you would max out the memory and then want some seriously fast swap space (the new DDR4 optane is Xeon only and requires custom re-coding).

  • 7 months ago
  • 2 points

This thread sounds nerdy AF and I love it. Please continue. :)

Sort

add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube