Open-Source AMD GPU Implementation Of CUDA "ZLUDA" Has Been Taken Down

[phoronix]

Phoronix: Open-Source AMD GPU Implementation Of CUDA "ZLUDA" Has Been Taken Down

Back in February of this year you may recall the interesting news that was announced on Phoronix that AMD Quietly Funded A Drop-In CUDA Implementation Built On ROCm: It's Now Open-Source. That open-source ZLUDA code for AMD GPUs has been available since AMD quit funding the developer earlier this year. But now the code has been retracted. It's not from NVIDIA legal challenges but rather AMD reversing course on allowing it to be open-source...

https://www.phoronix.com/news/AMD-ZLUDA-CUDA-Taken-Down

[Weasel]

Priceless.

"Open source friendly" company shills in shambles.

[ThomasD]

I don't even see how AMD profits from this. Bad news and also they don't benefit from less-capable cards.

[V1tol]

Priceless.

"Open source friendly" company shills in shambles.

Company asks to remove something semi-legal and what their users didn't have anyway. 0/10, never buying red again, jumping on fully proprietary green train now.

[sophisticles]

Let's be honest with ourselves, even if the request came from AMD's legal department, the reality is the request came from NVIDIA's legal department and AMD agreed to the back channel request to have this taken down.

I know people here will talk all sorts of trash about NVIDIA and this thread will likely hit 15 pages of insults, but the reality is that the open source community needs to come to grips with the concept of intellectual property rights,

NVIDIA spent billions of dollars to develop CUDA and the hardware to use it and they have a right to be the sole beneficiaries of their investment.

I encourage AMD to likewise invest money to develop their own proprietary CUDA competitor, their stock is tanking, it has gone below the $140 floor that many analysts had for them, despite a strong earnings report.

The tech industry is getting hammered, Intel has tanked to $20 a share and some analysts are saying they are facing an existential crisis.

Business can't be based on a "me too" mind set, you need something unique to get people to spend money.

At the moment only NVIDIA has that and they have a right to protect it.

[duby229]

Well, this was predictable. I was right here in February and I called it! ROCm is out., what other way is there to build a CUDA alternative? OneAPI?

[sophisticles]

I don't even see how AMD profits from this. Bad news and also they don't benefit from less-capable cards.

Their cards are not less capable because of this, they still have the same functionality.

And they benefit by not getting sued into oblivion by NVIDIA.

[varikonniemi]

Let's be honest with ourselves, even if the request came from AMD's legal department, the reality is the request came from NVIDIA's legal department and AMD agreed to the back channel request to have this taken down.

I know people here will talk all sorts of trash about NVIDIA and this thread will likely hit 15 pages of insults, but the reality is that the open source community needs to come to grips with the concept of intellectual property rights,

NVIDIA spent billions of dollars to develop CUDA and the hardware to use it and they have a right to be the sole beneficiaries of their investment.

I encourage AMD to likewise invest money to develop their own proprietary CUDA competitor, their stock is tanking, it has gone below the $140 floor that many analysts had for them, despite a strong earnings report.

The tech industry is getting hammered, Intel has tanked to $20 a share and some analysts are saying they are facing an existential crisis.

Business can't be based on a "me too" mind set, you need something unique to get people to spend money.

At the moment only NVIDIA has that and they have a right to protect it.

you are delusional if you think any law can prevent someone from making drivers that support another company's product.

In fact trying to prevent such an effort would be seen as anticompetitive measure and illegal in the sane part of the world.

[billyswong]

NVIDIA spent billions of dollars to develop CUDA and the hardware to use it and they have a right to be the sole beneficiaries of their investment.

I encourage AMD to likewise invest money to develop their own proprietary CUDA competitor, their stock is tanking, it has gone below the $140 floor that many analysts had for them, despite a strong earnings report.

The whole problem of GPU compute in AMD is, they never invest in a compute protocol that works for ALL GPUs from them, with the same support length of normal graphics API. Stupid segmentation means GPU compute in AMD is not practical without up-front heavy investment. They can let GPU compute in cheap consumer cards or APUs be slow. But it has to be still fully functional. Without such guarantee, no wonder all other 3rd party applications are reluctant and slow in implementing their compute API. The market of their API is too small. Being "open-source" cannot save the game.

[billyswong]

you are delusional if you think any law can prevent someone from making drivers that support another company's product.

In fact trying to prevent such an effort would be seen as anticompetitive measure and illegal in the sane part of the world.

I think the legal trick is likely done like this:

Everyone who read the CUDA API documentation must claim to agree with the NDA before proceed. The NDA said you shalt not create a rival product that implement the API. For clear legal CUDA implementation, one will then need to find someone who has never read the CUDA API documentation to do a clean room reverse engineering. Factor in the CUDA API is a moving target, the whole process become prohibitively hard.