Join us at the #ChipletSummit on 1/25 @ 4:30PM for the Design-1 Session! @beach66
, #VPofProduct at @UntetherAI
, will present his paper, "Energy-Centric #AI Acceleration with Chiplet-Based Design." Learn more about the event and register now at https://
bit.ly/3iStbHQ!
AI HARDWARE
-
Energy-Centric AI Acceleration with Chiplet-Based Design Summit
By
–
-
GroqChip enables real-time risk calculations for investment portfolios
By
–
Our performance enables real-time, low latency capabilities needed to manage complex risk calculations in investment portfolios. Watch this #SC22 demo by @MaxelerTech of a risk application based on Value at Risk (VaR) running on GroqChip™. https://
youtu.be/NE-5MW46nH4 -
ConvNets Remain Superior Due to Parallelism Advantages
By
–
Nope, convnets are still better due to parallelism. @ylecun probably just made his recommendation based on parallel computation and hardware.
-
AI Research Entry Barrier Rises to Training 50-70B Models
By
–
two years ago the entry point to participate in AI research was to have a couple of GPUs for training or finetuning now the entry level is to be able to regularly train a 50-70B params model on a couple hundred billion tokens
-
GPU Cold Boot Performance Concerns and Firecracker Equivalent Solutions
By
–
is there no "firecracker for GPUs" yet? multi-second cold boots sounds rough!!
-
A100 GPU Pricing at $18/Hour: Unit Economics and Utilization
By
–
banana pricing page makes an A100 $18 per hour, I feel like most unit econ can support that? i guess it's a question of utilization %?
-
GroqChip Enables Low Latency Speech Transcription Demo
By
–
Watch how GroqChip™ is used to run low latency, high throughput speech transcription in an enterprise setting in this #SC22 demo of Bifrost. https://
youtu.be/FjxkViqyFoI -
AI Development Beyond Moore’s Law: The Real Work Begins
By
–
When AI is done riding Moore’s law, the real work will begin.
-
Tenstorrent’s Ascalon RISC-V chip for data center AI
By
–
"future gens of Tenstorrent chips will use the company’s in-house #riscv design, Ascalon, which was presented at the RISC-V Summit… Tenstorrent’s data center inference and training chips, Grayskull and Wormhole, are in production now." Thanks @sallywf
-
Cerebras: New Architecture for Large-Scale Machine Learning
By
–
In their recent article, @BainCapVC mentions Cerebras as a "new architecture… that [is] fundamentally better suited for large-scale machine learning" Read their full article here: