I'm considering purchasing a new GPU and would like one that can crunch tasks faster than my i3 can. Does anyone have a recommendation for a GPU that works well?

I'm aware that there are probably lots of GPUs that work. I'm looking for something in the $300 - $500 range.

Climate Prediction Net doesn't use GPUs. That is fortunate, since GPU prices are ridiculous anyway. I bought a GTX 1070 a year ago for $330 USD, which sells today for $800. Wait until prices come down.

You'd have to be quick on the draw with ordering GPUs in that range to get close to MSRP from an etailer. A 1060 or 580 should be in that range. 1050/Ti's are as overpriced as some other cards.

Did you mean CPU? You mentioned an i3 witch is a Intel CPU, I would think a CPU with a lot of cores and multithreading.

Did you mean CPU? You mentioned an i3 witch is a Intel CPU, I would think a CPU with a lot of cores and multithreading.

Good catch. I have a Ryzen 1700 and am very happy with it. It usually does a bit better per core than my Haswells (i7-4770 and i7-4790) on most projects. On WCG, Einstein, LHC, MilkyWay, and GPUGrid (to name most of them) it does well. Also, on TNGrid (if you can get the work), the FMA version does better on the Ryzen 1700 than the AVX version does on my i7-4770. With a more recent Intel chip (i7-8700 for example), you should get more output per core. But with 8 cores for the Ryzen, I expect the total output to be greater, though I have no way of measuring it myself.

So AMD is the direction I am heading in the foreseeable future, and I will be building a Ryzen+ later this year. It does not do well on Rosetta for some reason, with low output and a high error rate. But I have noted that on their forum, for what it is worth. And I don't know how it does on CPDN, since my machine is on Linux, and there is no Linux work here at the present. I would expect it to do well though.

So AMD is the direction I am heading in the foreseeable future, and I will be building a Ryzen+ later this year. It does not do well on Rosetta for some reason, with low output and a high error rate. But I have noted that on their forum, for what it is worth. And I don't know how it does on CPDN, since my machine is on Linux, and there is no Linux work here at the present. I would expect it to do well though.

I have a Ryzen 1500+ and it works fine in cpdn, as long as I turn off multi-threading (AMD's hyperthreading). However, with it on, many errors occurred on cpdn. It's all about some Linux problem with something in Ryzen behavior under heavy load. This is a thread on the AMD forums about gcc compilation segment faults, but other tasks that put it under heavy load can show problems as well. https://community.amd.com/thread/215773 126 pages in that thread.

Many Linux users have sent their Ryzen's in for RMA and received later versions that don't have the problem. The new generation of Ryzen's is not supposed to have this problem.

Get it RMA'd. No seg faults with my TR.

I wonder if it is an instruction set issue or microcode maybe. I have a 1920X and have had 3 errors so far, do these problems effect the Threadripper?

I wonder if it is an instruction set issue or microcode maybe. I have a 1920X and have had 3 errors so far, do these problems effect the Threadripper?

Supposedly not. https://www.phoronix.com/scan.php?page=news_item&px=Ryzen-Segv-Response

Also, Ryzens on Windows aren't supposed to be affected.