I don't know much about these newer forms of processing modules.
The specs don't say a lot about them, other than it's got 2Tflops (probably at half or quarter ops, but not sure).
Could these modules be viable hardware upgrades for programs like fah? (More thinking about boinc or other folding programs using lower atom counts).
https://www.aaeon.com/en/p/ai-edge-comp ... re-xm-2280
Intel ai cores and folding?
Moderator: Site Moderators
Forum rules
Please read the forum rules before posting.
Please read the forum rules before posting.
Re: Intel ai cores and folding?
Hardware designed for AI isn't particularly useful to FAH. While Half Precision Floating Point seems to be well suited for AI -- and can be somewhat faster than Single Precision, it is not useful to FAH. The FAHCores use mostly Single Precision and somewhat limited Double Precision Floating Point calculations.
I have no comments about what BOINC needs or doesn't need. This forum supports FAH, not BOINC.
I have no comments about what BOINC needs or doesn't need. This forum supports FAH, not BOINC.
Posting FAH's log:
How to provide enough info to get helpful support.
How to provide enough info to get helpful support.
-
- Posts: 2573
- Joined: Mon Feb 16, 2009 4:12 am
- Location: Greenwood MS USA
Re: Intel ai cores and folding?
I suspect that they mean this: "as a dedicated neural network accelerator" and that it is not a general purpose device.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Re: Intel ai cores and folding?
They're like tensor cores.
Extra computing cores, running at lower frequencies (like older IGPs, but with more cores).
The fact that they come without cooling, makes me think they're not made for 100% load all the time, and won't go very fast.
Extra computing cores, running at lower frequencies (like older IGPs, but with more cores).
The fact that they come without cooling, makes me think they're not made for 100% load all the time, and won't go very fast.