Something that is specifically designed for it can always eventually replace the current. It may take time but GPGPU is limited to certain performance and power efficiencies. ASICs came out and destroyed GPUs in every way possible. Give it time. I am sure we will see a shift from GPGPU in AI tasks.
You didn't say ASICs, you said FPGAs.
Something to know about AI is that it's different than crypto currencies, where there's a set algorithm that you can just implement in hardware. For AI, people are still developing new layer types, and programmability is key for that.
However, it's still the case that the most intensive parts of deep learning tend to be fairly consistent, and those parts have been virtually hardwired in Nvidia GPUs, not unlike how they have cast key portions of the 3D graphics pipeline in hardware blocks.
So, the story around AI really looks a lot like graphics, in that both programmability and hardware acceleration are key. Memory bandwidth, as well. Really, the key deficit of GPUs relative to purpose-built AI chips is that they tend to have less on-die memory.
I' m not saying purpose-built AI chips won't eventually
win the day, but it's important to understand why
GPUs have have staying power.
Also, I really don't buy the idea that FPGAs are going to displace GPUs, in AI. They don't have the same level of flexibility as GPUs, nor do they have the level of efficiency and compute-density of purpose-built AI chips.
the only reason why it may benefit AMD is they do not have the funds to compete with Intel and Nvidia at the same time. Intel has the money to start GPUs and it wont affect their CPUs. AMD does not.
Did you ever hear of a CPU called Zen 2, or a GPU called Navi? Those were both made while building both CPUs and GPUs. Not only that, but AMD was in weaker financial shape, at the time. So, please tell me again how they don't have money to do both.
The only thing I agree with is that they haven't had resources to properly focus on AI. But, GPUs are obviously about much more than that. A large amount of GPU compute still doesn't involve AI, even if AI is starting to work its way into HPC.
If Intel is successful with their GPUs and is able to push into second place then AMD will be a third place product.
There are a lot of assumptions in that. And yet, you expect AMD to just shut down their successful and strategically vital GPU division and run for the hills? Because Intel is coming, and they might
be more successful than they were with Larrabee, Xeon Phi, or cellphone SoCs, or than they currently are with their 10 nm manufacturing process? With that kind of attitude, we would never have gotten Zen!
Not saying it will happen but Intel has the money to push harder than either one.
Not saying it will happen, but you are
saying that AMD should kill of f their GPU business, just in case? As I showed above, the mere fact of it being Intel doesn't mean they will automatically
be successful. And even if they are, there's still a lot of room for AMD to play, not to mention their partnership with Samsung to bring RDNA to mobile.
AI is a good market to try to get into. Emerging markets tend to be good margin markets.
a good market like 5 years ago. Now, the competition is so fierce that it's like the GPU business back in 1998, where there were like 20 companies all jockeying for position. AMD has no real presence, in this market. I wish it weren't the case, but they're hardly a footnote.
AMD could spin it off with an agreement to continue to work together. It doesn't have to be a complete spin off.
How is a smaller GPU player going to be more
competitive? If AMD wanted to raise more money for it, they'd have a much easier time selling shares than a newly-spun-off, standalone GPU maker. And you didn't answer the questions I posed about all of AMD's products and business deals that involve both their CPU cores and iGPUs or dGPUs.
I was just thinking in terms of financials trying to compete with Intel and Nvidia at the same time is very hard when you are just finally turning the company around from near disaster.
The actual turnaround came in like 2017. AMD is now in fine
shape. Companies don't need to have zero debt. And the debt is not a problem, as long as AMD's gross margins are growing, while its debt is shrinking.
Last year, AMD had GPU inventory overhang from the crypto-bust, their GPUs were years out-of-date (prior to the Navi launch) and most of their CPU sales were on Zen+, which was still struggling to be competitive. In both CPUs and GPUs, their product stack is now much more competitive, and they're getting lots more OEMs on board, especially in the laptop and server segments - both of the highest growth PC markets. 2020 looks like it's going to be a really good year, for AMD. Maybe some storm clouds towards the end, but I'm glad they have someone steering their ship with more guts than what you showed in that comment.