- Neurophos develops the Tulkas T100 optical processor capable of calculating 470 petaFLOPS AI
- Optical transistors are today 10,000 times smaller than conventional photonic transistors on silicon
- Dual crosshair design incorporates 768GB of HBM for memory-intensive workloads
Austin-based startup Neurophos has revealed that it is hard at work developing an optical processing unit called the Tulkas T100, which promises huge advancements in computing.
Funded by Bill Gates’ Gates Frontier Fund, the company claims the chip can deliver 470 petaFLOPS of FP4 and INT4 compute while consuming between 1 and 2 kW under load.
Its optical tensor core measures approximately 1000 x 1000, approximately 15 times larger than the standard 256 x 256 dies used in current AI GPUs.
Optical transistors and extreme speeds
Neurophos’ optical transistors aim to overcome traditional semiconductor limitations by extending Moore’s Law through higher computational density without increasing power consumption or chip size.
Despite its size, the startup says it only requires a single core per chip, supported by numerous RAM and vector processing units to maintain throughput.
Its optical transistors are about 10,000 times smaller than current silicon photonics, allowing a high-density array to fit on a single chip the size of a reticle.
“The equivalent of the optical transistor that we get today in Silicon Photonics factories is huge. It’s about 2mm long,” said Patrick Bowen, CEO of Neurophos.
“You simply can’t fit enough on a chip to get a computing density that remotely rivals today’s digital CMOS. »
The Tulkas T100 runs at 56 GHz, far exceeding previous CPU and GPU clock speeds.
SRAM powers the tensor core to maintain efficiency, and SSD storage can make it easier to move large data sets during testing and simulation.
The chip uses a dual-crosshair design with 768GB of HBM to support memory-intensive AI workloads.
Neurophos says the first-generation Tulkas T100 will focus on the pre-fill stage of AI inference by handling the processing of input tokens for large language models.
Bowen plans to pair Tulkas chip racks with existing AI GPU racks to accelerate computing.
However, the company does not expect full production until mid-2028, with initial deliveries numbering in the thousands.
Engineers are currently testing a proof-of-concept chip to validate the claimed computational density and power consumption.
Competitors such as Nvidia and AMD are also investing heavily in silicon photonics, signaling growing competition in this area.
AI tools and memory bandwidth constraints remain central considerations as optical accelerators seek to complement conventional GPUs.
Although the Tulkas T100 shows the potential to advance AI computing, its practical impact remains uncertain until the company achieves reliable production.
The optical approach remains experimental and faces challenges related to SRAM requirements, vector processing, and CMOS manufacturing integration.
Optical transistors could accelerate matrix multiplication and reduce energy per operation, but their efficiency depends on memory, SSD storage, and AI integration.
Neurophos says its chips are compatible with standard semiconductor fabs, but mass production depends on solving these technical challenges.
Via The register
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds. Make sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp Also.
