Ampere Announces Altra Max 128-Core Server Processor

[phoronix]

Phoronix: Ampere Announces Altra Max 128-Core Server Processor

Back in March Ampere Computing detailed their next-gen Altra ARM-based server CPU with up to 80 cores per socket. Today the company is revealing more roadmap details including the forthcoming Altra Max that offers 128 cores per socket.

http://www.phoronix.com/vr.php?view=29312

[danmcgrew]

If the industry keeps progressing like this, with more and more cores at faster and faster speeds, it won't be many years before it will be within our reach to build the ultimate, long-desired, fastest computing machine possible: the analogue computer.

[reavertm]

If the industry keeps progressing like this, with more and more cores at faster and faster speeds, it won't be many years before it will be within our reach to build the ultimate, long-desired, fastest computing machine possible: the analogue computer.

It's already within your reach. Find another human, of opposite sex, spend together some quality time and wait nine months. Its neural network will require some supervised learning at the beginning. Then, you can throw virtually any complex problem at it.

[CochainComplex]

It's already within your reach. Find another human, of opposite sex, spend together some quality time and wait nine months. Its neural network will require some supervised learning at the beginning. Then, you can throw virtually any complex problem at it.

I wish I could give you 2 likes for this one

[cynic]

It's already within your reach. Find another human, of opposite sex, spend together some quality time and wait nine months. Its neural network will require some supervised learning at the beginning. Then, you can throw virtually any complex problem at it.

I've done that. My two units have some issues with audio cards: sometimes they cry loud AF Beside that SMP works nicely, they are quite expandible and the builtin JIT compiler makes their learning performances constantly growing. 100% satisfied.

[CochainComplex]

I've done that. My two units have some issues with audio cards: sometimes they cry loud AF Beside that SMP works nicely, they are quite expandible and the builtin JIT compiler makes their learning performances constantly growing. 100% satisfied.

Common Bug - My units do also have issues with the volume control/ I think it is just kernel panic. Speech recognition is also sometimes malfunctioning not sure of it is more a "priority of tasks" issue. Tasks given by the "owner" are often not real time processed....but luckily they are windows free...

[starshipeleven]

It's already within your reach. Find another human, of opposite sex, spend together some quality time and wait nine months. Its neural network will require some supervised learning at the beginning. Then, you can throw virtually any complex problem at it.

Define "complex". The average human neural network is unable to process the amount of information needed to drive properly a car. Trusting it with anything "complex" is insane.

The only real edge is on burst image processing.

[starshipeleven]

If the industry keeps progressing like this, with more and more cores at faster and faster speeds, it won't be many years before it will be within our reach to build the ultimate, long-desired, fastest computing machine possible: the analogue computer.

I don't know if you are talking of human brains, but others seem to imply you do. Biological neural tissue isn't analog, it's still a digital system, even if it is using a different architecture than electronic computers. Firing signals in the neural network happens only after a certain threshold on the signal sender node(s) is reached. This is NOT what an analog system does.
This is how transistors (base of all modern electronics) also work, among other things, unless you provide them enough voltage to reach their threshold, they won't switch their own signal on or off.

[starshipeleven]

Common Bug - My units do also have issues with the volume control/ I think it is just kernel panic. Speech recognition is also sometimes malfunctioning not sure of it is more a "priority of tasks" issue. Tasks given by the "owner" are often not real time processed....but luckily they are windows free...

You are just "raising them wrong".

[bridgman]

Define "complex". The average human neural network is unable to process the amount of information needed to drive properly a car. Trusting it with anything "complex" is insane.

Are you sure ? My impression was that the problem was insufficient training, not insufficient processing capacity.