Question A new CPU paradigm from Efficient.Computer

[igor_kavinski]

Explore our Computing Solutions | Efficient Computer

Computing solutions redefined: Efficient's reconfigurable dataflow Fabric architecture delivers software‑programmable, ultra‑energy‑efficient performance for edge‑intelligent applications.

efficient.computer

Startup claims 100x more efficient processor than current CPUs, secures $16M in funding

Efficient Computer emerged from stealth mode this week, revealing their "Fabric" chip design. According to the company, this novel CPU can provide performance on par with current...

www.techspot.com

Even if it is successful, it's gonna be 5 years or more before we see this in consumer devices. Developing software for it from scratch is gonna be no simple task. Unless people suddenly start using Linux.

 

[moinmoin]

They should ship with Gentoo Linux.

 

[Abwx]

Explore our Computing Solutions | Efficient Computer

Computing solutions redefined: Efficient's reconfigurable dataflow Fabric architecture delivers software‑programmable, ultra‑energy‑efficient performance for edge‑intelligent applications.

efficient.computer

Startup claims 100x more efficient processor than current CPUs, secures $16M in funding

Efficient Computer emerged from stealth mode this week, revealing their "Fabric" chip design. According to the company, this novel CPU can provide performance on par with current...

www.techspot.com

Even if it is successful, it's gonna be 5 years or more before we see this in consumer devices. Developing software for it from scratch is gonna be no simple task. Unless people suddenly start using Linux.

That s just outrageous claims to have money from the unsuspicious crowds flowing in their pockets, more likely a scam than anything else.

 

[Nothingness]

If it sounds too good to true, it is too good to be true.

 

[Hitman928]

Sounds eerily similar to both the approach and claims that Tachyum made with their original Prodigy CPU before their roadmap blew up in their faces and they had to completely redesign the CPU from the ground up, dramatically change their claims, and push the release date years out (with still no silicon to speak of to date).

 

[soresu]

Explore our Computing Solutions | Efficient Computer

Computing solutions redefined: Efficient's reconfigurable dataflow Fabric architecture delivers software‑programmable, ultra‑energy‑efficient performance for edge‑intelligent applications.

efficient.computer

Startup claims 100x more efficient processor than current CPUs, secures $16M in funding

Efficient Computer emerged from stealth mode this week, revealing their "Fabric" chip design. According to the company, this novel CPU can provide performance on par with current...

www.techspot.com

Even if it is successful, it's gonna be 5 years or more before we see this in consumer devices. Developing software for it from scratch is gonna be no simple task. Unless people suddenly start using Linux.

Something something dataflow architecture.

Sounds like the EDGE ISA that MS Research was pursuing for a while under the E2 name back in the 2010s, sadly that went nowhere and I don't think any dataflow architecture has ever managed to break in since.

 

[soresu]

This blurb makes me dubious.



Call me crazy, but aren't FPGA's basically reconfigurable logic gate arrays?

Therefore potentially as software programmable as whatever processor configuration you can manage to stuff into the gate limits of the SKU?

Makes me think they are trying to smack talk a competitor before they are even competing, which doesn't bode well.

 

[igor_kavinski]

If they can do decent AI work at 1000x less power than Blackwell, that's all they gonna need.

 

[Markfw]

So, they are going to to this at 0.256 watts ?? I don't think so....


+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.105.17 Driver Version: 525.105.17 CUDA Version: 12.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce ... Off | 00000000:01:00.0 Off | N/A |
| 41% 64C P2 256W / 320W | 689MiB / 16376MiB | 89% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 2645 G /usr/lib/xorg/Xorg 120MiB |
| 0 N/A N/A 3283 G cinnamon 19MiB |
| 0 N/A N/A 3552529 G ...2gtk-4.0/WebKitWebProcess 12MiB |
| 0 N/A N/A 3758478 C ....3/Core_23.fah/FahCore_23 532MiB |
+-----------------------------------------------------------------------------+

 

[igor_kavinski]

It depends on the AI algorithm and also how the efficient cores distribute the workload amongst themselves.

Low power neural network mimicry isn't science fiction: https://news.mit.edu/2022/analog-deep-learning-ai-computing-0728

I have a vested interest in seeing AI on GPU become obsolete. The faster it happens, the faster we can get back to a world where GPUs are used primarily for super realistic graphics rendering, their actual and original purpose.