News:

Members can see all forum boards and posts. Non members can only see a few boards.  If you have forgotten your password use this link to change it.
https://forum.boinc-australia.net/index.php?action=reminder

Main Menu

Intel NPU & Ryzen AI

Started by Abruraspingi, February 28, 2024, 06:53:45 PM

Previous topic - Next topic

Abruraspingi

With all this talk about how all PC's will soon become AI PC's I kept wondering how it would affect BOINC and Folding@home. So far there is little information on it and all I could really find was a forum thread on the now completed MLC@Home project
https://www.mlcathome.org/mlcathome/forum_thread.php?id=4

I guess unless the programmers design code for these chips not much will change but like the Tensor cores in Nvidia chipsets I'm sure it will be just a matter of time.

It seems that the Intel NPU and Ryzen AI just save on power consumption at the moment. Here are a couple of articles to read
https://www.tomshardware.com/laptops/i-tested-intels-meteor-lake-cpus-on-ai-workloads-and-amds-chips-sometimes-beat-them

https://www.intel.com/content/www/us/en/newsroom/news/core-ultra-client-computing-news-1.html

https://www.amd.com/en/products/processors/consumer/ryzen-ai.html

On another note using BOINC to crack passwords is an interesting use of the software
https://www.sciencedirect.com/science/article/abs/pii/S1742287619301446

ryzenmulti

Thanks Ab. There are only so many ways to move bits through a processor and there is a crudload ($trillions$) invested in existing hardware and software development. I doubt a paradigm shift will happen fast unless it supports and/or rebuilds existing legacy code. Given $AI is $dollar driven, I doubt it will be a sunami of change. ... and unless they can stop robbing m/b real estate with slot hog GPU's, things need to get smaller before they get any bigger.

Besides, $AI is just a series of chained if/then condition statements. It requires trillions more logic gates to be truly neural, and unbounded output makes things like what Musk posted on twitter today (i.e it's a farce). Just like hyperthreading is/was. My 20c.

The further back you look, the further forward you can see.

The gang is all here: 160 intel cores (no H/T), 212 epyc/ryzen cores (no SMT) 6 RTX GPU's, 10 TitanV GPUs and more than 1TB RAM. Coming soon ... Instinct MI100

Home cooked twin primes using python ... it started out with 256 digits of pi and eulers number and progressed at light speed to 299792458^1146+101436145 / +101436147 (9715 digits) is a twin prime. Current PB (PiBest) is 314159265^1615+962400452 / +962400454 (13723 digits) is a twin prime!

Abruraspingi

Completely agree with you on that one. There is a lot of money in AI and BOINC doesn't really produce the dollars. Still cheaper then time on a supet computer.

Google has its Alpha Fold AI for folding proteins bht that doesn't mean that power will be in consumer hands.

I remember why the PhysX chip first came out. It was a separate card that plugged into your M/B and took some of the heat off of the CPU. PhysX is kinda standard now but it's up to programmers to utilise it.

What I see as a boon for BOINC when it comes to NPU's is if the chip recognises a trend in the WU and skips a couple of calculations saving time and shortening the WU