Plan to keep advanced chips from China with tracking tech gains support in Congress Proposed legislation gaining steam in Congress this week would require high-end GPUs and AI chips to include location-tracking safeguards to ensure US-designed components don't end up in nations against Uncle Sam's wishes, with exporters on the hook for compliance. On Thursday, a group of bipartisan legislators introduced [PDF …
So chips in a data centre with high L-band attenuation through reinforced concrete have GPS? Nope, not going to work...
Chips have to "phone home" so you can check IP address? Hmm, just have open Internet access routed via a VPN to ${Country_of_choice} then.
In related news pigs have been seen flying!
And even if the chips can phone home, IP address localization is not foolproof...
Lots of US companies bought IPv4 ranges from LACNIC and AFNIC (and even from RIPE) for US usage due to the lack of available ranges in the US.
Which localize them... in Africa or the Pacific area with an US physical address.
Lots of US companies bought IPv4 ranges from LACNIC and AFNIC (and even from RIPE) for US usage due to the lack of available ranges in the US.
Rather a joke on the US if the only out was wider adoption of IPv6 in the US which I suspect is currently miniscule. The PRC and East Asia generally IPv6 is more prevalent and maybe the rule.
I don't know if or how geolocation works with IPv6 but I would think each /48 prefix would map directly to the entity assigned the prefix which I guess in the absence of NAT or proxy/gateways would be better than IPv4.
Another case of politicians proposing laws without seeking any technical guidance of how things actually work first. Just like the backdoors in E2E encryption only for the good guys they keep going on about.
They might as well just pass a law that says every GPU must come with a magic unicorn which can fly back to the US and report on its location for all the good the law will do if enacted.
"Pi=4"
It was Pi=3.2 and was an Indiana House Bill that never became law.
In good American style, the bill neither was an attempt to improve the law or lives of the people of Indiana, nor give a "good" value of Pi. It was some kind of patent scam.
Rather, the bill was aimed at benefiting its author, who claimed to have patented a new method for "squaring the circle", which he proposed to let the state of Indiana use free of charge if they would pass his bill! Its opening statement is clear:
PS: I too remembered a different value and story. But this is the consensus of history sites.
Only us older coders would get the reference in the headline "Every shipment you make, every FLOP you generate, Uncle Sam will be watching you" to the 1980's song Every Breath You Take by The Police aka Sting! Good job Register article writers and editors!
LOL!
That said, this total Bollocks with tracking of GPUs by the USA is going to be COMPLETELY SUBVERTED since we here in Canada are disclosing as Worldwide, Completely Free and Open Source under Multiple Licence Terms including but not just limited to GPL-3 licence terms, the full tape-out design and all documentation for our fully-ITAR-free 128-bits super-processor chip which is a combined-CPU/GPU/DSP/Vector processor! It WAAAAAAAAAAAAY OUT-PERFORMS any NVIDIA, AMD or Intel GPU or CPU and completely undercuts what is being proposed!
Since it's line traces and transistors are much larger-than-normal at 60 microns wide and 100 microns deep on 200 mm by 200 mm (20 cm by 20 cm) stacked borosilicate glass plates, you can make these superchips at home DIY (Do It Yourself) using copper powder deposition and laser sintering using our also-free downloadable multi-head 3D printer design and then run them at 5 Volts at 2 THz for 50 PetaFLOPS sustained at 128-bits for the Floating Point, Fixed Point and Integer operations!
This is what our superchip design contains:
128-bits-wide Combined-CPU-GPU-DSP-Vector Array Processor with 1024 cores at 8 threads per core for 8192 total threads running at 2 THz clock speed and 50 PetaFLOPS Sustained for 128-bits wide Floating Point, Fixed Point and Integer arithmetic. Has 200 mm by 200 mm multi-layer Borosilicate glass stacked dies with copper line traces of 60 microns wide at 100 microns thick running at 5 volts for optimal clock stability and mitigating RFI/EMI, crosstalk and inductive loads. Dies are interspersed with cooling plates that are 5 mm thick hollow-tube-etched aluminum plate for fast thermal transfer capability, and each cooling plate is filled with silicone oil rapidly circulated in and out of each plate for best cooling of the above and below CPU die layer.
The processor has the following built-in functionality:
a) Built-in SDR (Software Defined Radio) for 100 MHz to 120 GHz bandwidth transceiver capability.
b) On-board fractal antennae with 32-bits per sample ADC/DAC and DSP (Digital Signal Processing) with 240 Gigasamples per channel (integer 32-bits per sample) at 128 channels of input and output.
c) Hardware-accelerated ASCII and UNICODE character string handling and search functions and advanced multi-parameter multi-language string search functions.
d) Powerful multi-state Boolean Logic processor with 0.00% to 100.00% weighted-state Expert System/A.I. acceleration
e) Tiled display support for up 128-displays each going up to 16,384 by 16,384 pixels resolution at up to 1000 fps video I/O with support for 32-bits, 64-bits and 128-bits wide pixels in RGBA, YcbCrA, HSLA, CMYKA, CIE-XYZA, CIE-LABA and Greyscale formats with hardware-accelerated 2D-XY/3D-XYZ line/curve drawing, fill functions and built-in real-time 2D-XY/3D-XYZ convolution kernels of 3x3/3x3x3, 5x5/5x5x5, 7x7/7x7x7 and 9x9/9x9x9 matrix sizes and presets-based video filters/image enhancement such as 2D-XY/3D-XYZ SOBEL/CANNY edge detection and pixel-to-vector conversion and A.I.-assisted object recognition.
f) Audio 128 channels input and 128-channels output at integer 32-bits per sample at 10 MHz sample rates with built-in real-time 2D-XY/3D-XYZ convolution kernels of 3x3/3x3x3, 5x5/5x5x5, 7x7/7x7x7 and 9x9/9x9x9 matrix sizes and presets-based audio filters, 128 built-in MIDI/MADI Input/Output/Passthrough audio ports and 2048 instruments/voices audio synthesizer with built-in ROM-based high-resolution instrument/voice samples, specialized sound effects and advanced noise-reduction systems, and high-quality A.I.-assisted speech-to-text conversion and A.I.-assisted text-to-speech conversion.
g) MEMS-based millimetre accuracy 3D-XYZ global location services and 3D-XYZ spatial orientation with Decimal Degrees for Lat/Long and metres for altitude with up to 0.000000000 to 259.999999999 global position accuracy levels and 0.000000000 to 999,999.999999999 metres for altitude accuracy. (i.e. millimetre-levels of positional accuracy)
h) Encrypted XML/JSON/Web Metadata Processor with multi-format 128-channels input and 128-channels output for real-time Audio/Video/Peripheral-Device formatted XML/JSON text, pixel and vector graphics overlays and binary or simple text metadata streams.
I) 128-stream input/128-stream output Quantum Computing Resistant, anti-Shor's algorithm encryption processor with support for current, future and legacy encryption standards
j) 128-port full-duplex dense-wave multiplexing fibre-optic communications router at Terabit+ data transfer speeds per optical port. The entire chip uses only optical ports for all off-chip I/O!
k) 128 of 65,536 by 65,536 2D-arrays of SIMD and MIMD vector processors for separate processing of 8-bit, 16-bit, 32-bit, 64-bit and 128-bits wide Floating Point, Fixed Point and Signed and Unsigned Integer values and 8-bit and 16 bit weighted-state boolean values for Expert System and A.I. processing. Vector processor supports for accelerated local-SIMD/MIMD-core storage and processing of 2D-XY/3D-XYZ convolution kernels of 3x3/3x3x3, 5x5/5x5x5, 7x7/7x7x7 and 9x9/9x9x9 matrix values.
I think that should put a few deep Halloween Super-Scares into basically EVERY CPU and GPU manufacturer since it's all FREE for you to download, build and use!
North Canadian Aerospace has it and YOU can get it Summer 2025!
V
I guess by the time the US legislature and Americans generally finish playing with themselves† the PRC etc will have overtaken the US in this domain and in other domains which if anyone sufficiently sentient remains in the union to realise that then a rude awakening is guaranteed them.
† if ever. The nuns always maintained that blindness was a hazard.
Goto the North American Aerospace website in Summer 2025 and download the tape-out files, all documentation AND the 3D printer construction files. We made this 128-bits wide Superchip to specifically to be able to be built at home using Borosilicate glass plates that are $25 CAD ($20 USD) each and copper nano-powder that is 3D printed using laser sintering for a mere $55 CAD ($40 USD) per bag. The rest of the printer and chip etcher parts can easily be bought at Home Depot, Canadian Tire, Amazon, Temu or Aliexpress.
We put a LOT of thought into this and made sure even high-schoolers could make and built them! Everything was designed to be as error-correcting as possible which is WHY we chose the 60 micron line trace widths and 100 micron line trace depths AND Borosilicate Glass for the dies! We know what we are doing! There's a 100+ eggheads working on this for a VERY long time and they ain't dumb! Even the Operating System software is very forgiving and error-correcting because these are NOT clean-room conditions so there has to be a lot of "slop" be allowed during the 3D printing of such a device at home. Most everyone else is making things much much smaller in terms of chip-making processes while we went completely the opposite by making everything physically much larger and higher-powered (5 volts+) to ensure things work!
V
Rep. Baldrick (R): "I have cunning plan wherein we place an advanced location determining circuit in the silicon of every advanced chip. The circuit itself is so advanced that it requires its own such circuit."
Sen. Blackadder (D): "I think as usual I can see a microscopic flaw in your cunning plan but I am sure we get start up funding for a company to "bid" for government contracts to further develop the concept."
Richard Curtis even aided by Rowan Atkinson and Ben Elton couldn't have scripted contemporary US politics without having valid doubts concerning their own sanity.
One option might be embedding a small processor in the GPU etc that after a predetermined time it disables or limits the GPU's functionality.
Each chip in customer's batch would have a unique batch specific cryptographic key fused or masked into it.
The device driver or a user space application (daemon) would be required to periodically upload into the chip a cryptographically protected time limited enabling ticket renewable before expiry from a vendor service.
Obviously the chip would need to be able to keep some state internally to prevent replay attacks.
If any of the chips are believed to have been acquired contrary to US restrictions the vendor can stop issuing tickets for the batch. No ticket, no ride.
Offered in the clear understanding that domestic users will avail themselves fully of the self immolation opportunities presented†.
† if the timely renewal SSL certificates is often a bridge too far...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
