News Chinese tech giants boosted Nvidia GPU purchases by 4x to 6x during Q1

[Admin]

Chinese tech giants continue to buy Nvidia AI GPUs despite DeepSeek impact, demand slowdown.

Chinese tech giants boosted Nvidia GPU purchases by 4x to 6x during Q1 : Read more

 

[Constellar]

Wait a minute--I though the China fanboys have been telling us that China no longer needs the United States.
China fanboys, what gives? First, Chinese entities are smuggling like 100,000 USA-made Nvidia H100's through shell companies operating out of Singapore, and now this, with tencent and Alibaba's unquenchable thirst for USA chips.
Now I'm like really confused here... 😃

 

[The Historical Fidelity]

Wait a minute--I though the China fanboys have been telling us that China no longer needs the United States.
China fanboys, what gives? First, Chinese entities are smuggling like 100,000 USA-made Nvidia H100's through shell companies operating out of Singapore, and now this, with tencent and Alibaba's unquenchable thirst for USA chips.
Now I'm like really confused here... 😃

Nvidia should have had the foresight to know a complete ban was inevitable. Hopefully they have devised an un-hackable method (or as un-hackable as us mere mortals can muster) of a geographic challenge that prevents smuggled GPU’s from being utilized in banned markets. For example, If the challenge fails, then a master fuse within the GPU die will permanently render the chip useless.
Otherwise, this new total ban will be as in-effective as the previous attempts.

 

[JarredWaltonGPU]

Nvidia should have had the foresight to know a complete ban was inevitable. Hopefully they have devised an un-hackable method (or as un-hackable as us mere mortals can muster) of a geographic challenge that prevents smuggled GPU’s from being utilized in banned markets. For example, If the challenge fails, then a master fuse within the GPU die will permanently render the chip useless.
Otherwise, this new total ban will be as in-effective as the previous attempts.

I don't think Nvidia has any real desire to prevent China from getting GPUs; it's only doing what it needs to do to comply with US export restrictions. If the US weren't blocking sales, you can be sure Nvidia would be shipping truckloads of B200/B300 racks to China.

But the kill switch is something I've wondered about. It would absolutely be possible to create such a thing. At the same time... you don't just blow $50,000 worth of hardware (or however much it costs) because of a failed communication. Internet went down? Oops! There goes millions of dollars in server hardware!

And if you can't disable the hardware due to failed networking... well, then China / whoever just builds an appropriate firewall to prevent any communication with the system that would tell the fuse to blow.

 

[The Historical Fidelity]

I don't think Nvidia has any real desire to prevent China from getting GPUs; it's only doing what it needs to do to comply with US export restrictions. If the US weren't blocking sales, you can be sure Nvidia would be shipping truckloads of B200/B300 racks to China.

But the kill switch is something I've wondered about. It would absolutely be possible to create such a thing. At the same time... you don't just blow $50,000 worth of hardware (or however much it costs) because of a failed communication. Internet went down? Oops! There goes millions of dollars in server hardware!

And if you can't disable the hardware due to failed networking... well, then China / whoever just builds an appropriate firewall to prevent any communication with the system that would tell the fuse to blow.

I was actually thinking the opposite. All the logic would be onboard the chip, an authentic positive geographical confirmation, by a GPS chip built directly into the GPU die hidden amongst the billions of transistors, would cause the GPU to activate. If the GPS logic cannot confirm and authenticate what country it is within, then it will remain in pre-activation mode and re-attempt confirmation. Only when the GPS logic confirms and authenticates that it is within a banned country, then the on-die fuse gets blown.

It’s just what I would do if I ran Nvidia, but like everything in life, I’m sure there are unforeseen consequences with my idea.

 

[Eximo]

Only when the GPS logic confirms and authenticates that it is within a banned country, then the on-die fuse gets blown.

GPUs are typically underneath a big pile of metal, inside of a metal chassis, inside of a building. Not the best environment for GPS.

You could rig up a local antennae spoofing far off satellites to circumvent such a thing. Would just have to get the timing delay correct.

 

[The Historical Fidelity]

GPUs are typically underneath a big pile of metal, inside of a metal chassis, inside of a building. Not the best environment for GPS.

You could rig up a local antennae spoofing far off satellites to circumvent such a thing. Would just have to get the timing delay correct.

A spoofed satellite would fail authentication. The U.S. army does it all the time using both Navigation Message Authentication + Chimera Signal Level Authentication which renders spoofing useless since the only option is for the “spoofer” to relay the original unadulterated signal + navigation message, so the GPU GPS chip would still see the true location it is being used at.

The reason why spoofing is still a thing is because GPS receiver manufacturers fail to implement these official system protections in most civilian models.

 

[jp7189]

Nvidia should have had the foresight to know a complete ban was inevitable. Hopefully they have devised an un-hackable method (or as un-hackable as us mere mortals can muster) of a geographic challenge that prevents smuggled GPU’s from being utilized in banned markets. For example, If the challenge fails, then a master fuse within the GPU die will permanently render the chip useless.
Otherwise, this new total ban will be as in-effective as the previous attempts.

Nvidia doesn't seem to have any interest in complaince beyond avoiding fines. Case in point, the 5090D is a full fat 5090 with a special driver and firmware to detect AI workloads to artificially limit performance. That doesn't sound very hard to work around, and they certainly could have implemented a hardware solution if they truly cared to comply beyond a superficial surface level.