AMD's graphics DNA splits into two

If this trend continues, I can see RDNA & CDNA ending up as two separate companies

https://www.pcworld.com/article/3531413/amd-cdna-compute-gpus-radeon-rdna-graphics-cards.html
I doubt that it would split into two companies. Two divisions probably, but there is a tremendous amount of overlap between them even if they are for two different targets. I can't imagine them wanting to spin off another company. It would be like them spinning another company off for their EPYC CPUs and desktop CPUs. It just doesn't make sense.
 
I doubt that it would split into two companies. Two divisions probably, but there is a tremendous amount of overlap between them even if they are for two different targets. I can't imagine them wanting to spin off another company. It would be like them spinning another company off for their EPYC CPUs and desktop CPUs. It just doesn't make sense.

I was thinking of this because, for AMD gaming is driven by semi-custom (consoles), whereas compute by HPC.

There are shared elements in designing/architecture. But when it comes to implementation I think split might happen due to the vastly different focus

Intel by contrast (under Raja) have said that they will re-use same architecture across IGP, discrete, & HPC. That's probably because they see Graphics hardware as something to be sold along with their CPUs in laptops/super-computers

In case of Radeon, the GPUs have been stand alone & use cases also vastly different between professional & consumer divisions
 
It's a distinction without a difference based on how the same work is divided and labeled on paper, and the titles of the offices created to direct that work.

We're going to need to hit a much harder wall for process node improvement before CPU cores and GPU cores start to materially diverge in design between different target markets.
 
It's a distinction without a difference based on how the same work is divided and labeled on paper, and the titles of the offices created to direct that work.

We're going to need to hit a much harder wall for process node improvement before CPU cores and GPU cores start to materially diverge in design between different target markets.
This. Yes, different target markets, so some differences (more compute vs more pixels), but the basics between the two will still be shared. It's not like they are doing vastly different things. I could possibly see AMD coming out with an AI card more specific (obviously) to businesses, which may differ enough, but I don't think it would be spun off quickly as they would have some really stuff competition and would need some time to mature. GPU wise, I don't see a split in the future (at least not the next 5-10 years).
 
This. Yes, different target markets, so some differences (more compute vs more pixels), but the basics between the two will still be shared. It's not like they are doing vastly different things. I could possibly see AMD coming out with an AI card more specific (obviously) to businesses, which may differ enough, but I don't think it would be spun off quickly as they would have some really stuff competition and would need some time to mature. GPU wise, I don't see a split in the future (at least not the next 5-10 years).

What makes an AI chip different to a GPU? Go into as much detail as you can. I will likely not understand most of it but I'll others might.
 
I doubt that it would split into two companies. Two divisions probably, but there is a tremendous amount of overlap between them even if they are for two different targets. I can't imagine them wanting to spin off another company. It would be like them spinning another company off for their EPYC CPUs and desktop CPUs. It just doesn't make sense.

So, 3 companies?
 
This. Yes, different target markets, so some differences (more compute vs more pixels), but the basics between the two will still be shared. It's not like they are doing vastly different things. I could possibly see AMD coming out with an AI card more specific (obviously) to businesses, which may differ enough, but I don't think it would be spun off quickly as they would have some really stuff competition and would need some time to mature. GPU wise, I don't see a split in the future (at least not the next 5-10 years).
Intel has Xeon Phi, which really is different enough -- Nvidia makes different enough products that it'd make sense.

Yet we're still waiting for AMD to really approach gaming GPUs a bit separate from their compute products.
 
What makes an AI chip different to a GPU? Go into as much detail as you can. I will likely not understand most of it but I'll others might.
https://www.intel.ai/nervana-nnp/
Just an example, if it was the same thing Intel would not be making these + gpus. The hardware can be specific to the underlying neural network rather than being more general purpose. It also doesn't need a frame buffer, texture management, hdmi support and a host of other bits and pieces. It can be much different than a GPU, although doesn't have to be (which was my point, this can be different enough that they could split, but I believe this to be extremely unlikely too).
 
So, 3 companies?
Lol, no.. more likely 1 company. Like I said, the only possibility I could imagine was if they had a very specific piece of hardware (not general purpose), that they may have a reason to spin off. But even that doesn't really make sense to split. They would rather sell the server CPU + GPU + any other specialized hardware as a package.
 
Lol, no.. more likely 1 company. Like I said, the only possibility I could imagine was if they had a very specific piece of hardware (not general purpose), that they may have a reason to spin off. But even that doesn't really make sense to split. They would rather sell the server CPU + GPU + any other specialized hardware as a package.
The cost / benefit of AMD producing an AI-specific part is what I find questionable, and why I brought up the example of them not splitting their gaming and compute efforts similarly to how Nvidia has done, with Nvidia removing compute capability from lower end products as well as building a compute-lite, gaming focused top-end product for the last two generations as well.

Unless AMD already has a customer lined up and has a defined advantage (extremely low cost + enough performance), like they did with the consoles, this is an unlikely avenue to explore. They have lower-hanging fruit to go after elsewhere.
 
The cost / benefit of AMD producing an AI-specific part is what I find questionable, and why I brought up the example of them not splitting their gaming and compute efforts similarly to how Nvidia has done, with Nvidia removing compute capability from lower end products as well as building a compute-lite, gaming focused top-end product for the last two generations as well.

Unless AMD already has a customer lined up and has a defined advantage (extremely low cost + enough performance), like they did with the consoles, this is an unlikely avenue to explore. They have lower-hanging fruit to go after elsewhere.
I didn't mean it was likely, I was just saying that is the only far fetched idea I could imagine. Sorry if I came across like it was something that I thought likely to take place.
 
I didn't mean it was likely, I was just saying that is the only far fetched idea I could imagine. Sorry if I came across like it was something that I thought likely to take place.
Not at all!

I'm just adding opinion to yours, not trying to counter or detract :)
 
Back
Top