Thread for LTT Labs Test Suggestions

[PPush]

On 4/20/2024 at 6:46 PM, Xemor said:

Dear Labs Team,

 

I'm not sure if this has been suggested before, but the internet is in desperate need of an objective set of machine learning benchmarks for a wide variety of graphics cards. Ideally for training and inference, it doesn't have to be perfect as there is little competing information on the internet so anything would be an improvement. Ideally training a well known NLP model for some number of steps, and doing some amount of inference with the model and seeing how long it takes for a variety of graphics cards. There's a rapidly growing audience of people interested in which graphics cards are the best for machine learning, which is horrifically underserved by the current tech media landscape.

I had proposed using performance with LCZero (formally Leela Chess Zero), which is a deep neural network-based chess engine that is completely open source.
It allows using the CUDA backend for NVIDIA cards and the ONNX-DML for other cards.

[Deeezy]

Thought it would be a cool idea to have a thread with peoples ideas of what they want to see get tested by the LTT labs. I did i a quick search on the forum to see if one existed already and didnt see anything immediate but let me know if there is a post for that already.

 

Anyways I personally was interested to see some thermal compounds get put through the paces and whether or not certain compounds shined better in different environments (e.g. household/gamer pcs, server boards, etc). I also recently watched the Lumafield video showcasing the CT scan capabilities and wondered how mobile devices look in the scan. Spent the better part of an hour looking for a clean scan of an iphone 15 to make as a wallpaper (let me know if you find one of those too!). But those were just a few ideas that I'd be interested to see what about you guys?

[fedoralinuxrulez]

If you can get it try some thermalxr on a heatsink and see what happens 

[Anders Q Liden]

I don't know how many other people would like this but I would like a test that show the difference with one CPU compared to the major memory sticks that are available for that CPU.

 

For instance if I run CPU X today how will it perform with DDR5 compared to DDR4 (don't think we need to compare to DDR3 today) and perhaps even DDR5 4500 and 6400.

Now, I'm no hardware expert so I don't know exactly how to represent this, but I would love to get more knowledge about how faster/slower memory effect a CPUs performance. Perhaps it's the fact that I'm *not* knowledgeable about these things that makes me interested in it.

Good/bad idea? I'm not certain. But I would love to know.

Thanks for listening!

Edited Monday at 07:17 AM by Anders Q Liden
Wrong Spelling

[Andrew651]

What do you think about testing surge protectors and Ups/surge protectors? Do they actually work?