4 ms·

I admit to the extent I thought about nvidia I just kind of assumed they would fade away as companies like intel and apple integrated graphics processing into t

I admit to the extent I thought about nvidia I just kind of assumed they would fade away as companies like intel and apple integrated graphics processing into their own SOCs in phones tablets laptops etc. And I also figured the crypto boom was a temporary unsustainable surge in the market for discrete gpus. Now LLMs have taken up the crypto slack in terms of gpu demand. (I think? Is this what’s going on?) The question I have is whether there is any reason nvidia won’t be commoditized out of this market. Maybe someone who is smart at hardware can explain.

mensetmanusman 3y ago
If China can learn to get along with everyone, and they are allowed to purchase high-tech manufacturing equipment needed to train a workforce over the next two decades, then yes, it will eventually be commoditized.
bee_rider 3y ago
Ultimately CPU, GPU, and memory will be in the same package because, who wants unnecessary latency? But it isn’t clear to me if Intel or Nvidia are further ahead when it comes to figuring that out.
- twelve40 3y ago
  it's kind of a mystery why they haven't yet. The G part of the trillion dollar GPU almost sounds like a misnomer by now.
  - bee_rider 3y ago
    I think they basically have, it is just a really, really difficult problem to make a CPU that is competitive with Intel and a GPU that is competitive with NVIDIA, so putting it all together is not for the faint of heart.
- solarkraft 3y ago
  On consumer devices this is increasingly happening. In data centers probably not so much.
- Duplo 3y ago
  [dead]
- gateorade 3y ago
  I'm not sure this is obvious. For the consumer market, maybe. But at the end of the day, there are physical limits on how much compute you can squeeze into any given area. If your job requires more parallel compute than can reasonably be squeezed into a package for power/thermal reasons, then distributing the workload off-package is a requirement. Maybe hardware will morph into a more heterogenous mix than we have now, with many single-package cpu-gpu nodes working in parallel, instead of a few cpus orchestrating a gigantic sea of GPUs, but maybe not. I'm more convinced that this is just one stage of the cycle. GPUs in consumer hardware weren't really a thing before the 90s. From the 90s to like 2013 graphics software (and it's ability to load hardware) improved rapidly. Since then it's kind of stagnated and a lot more of the focus has shifted to doing the same amount of rendering with less power/less heat/less space. Even if we do see a shift toward SoCs/APUs/whatever in consumer hardware over the next few years, I'd bet on some sort of paradigm shift to come along (fully raytraced rendering?) and swing the pendulum pack toward big discrete GPUs at some point again.
xur17 3y ago
> And I also figured the crypto boom was a temporary unsustainable surge in the market for discrete gpus. It's kind of amazing how perfect the timing was for LLM popularity (right around the time Ethereum moved to PoS, and the crypto market cooled down).
- scotty79 3y ago
  I read that a lot of gpu miners went smoothly from mining to AI training.

mensetmanusman 3y ago

If China can learn to get along with everyone, and they are allowed to purchase high-tech manufacturing equipment needed to train a workforce over the next two decades, then yes, it will eventually be commoditized.

bee_rider 3y ago

Ultimately CPU, GPU, and memory will be in the same package because, who wants unnecessary latency? But it isn’t clear to me if Intel or Nvidia are further ahead when it comes to figuring that out.

twelve40 3y ago
it's kind of a mystery why they haven't yet. The G part of the trillion dollar GPU almost sounds like a misnomer by now.
- bee_rider 3y ago
  I think they basically have, it is just a really, really difficult problem to make a CPU that is competitive with Intel and a GPU that is competitive with NVIDIA, so putting it all together is not for the faint of heart.
solarkraft 3y ago
On consumer devices this is increasingly happening. In data centers probably not so much.
Duplo 3y ago
[dead]
gateorade 3y ago
I'm not sure this is obvious. For the consumer market, maybe. But at the end of the day, there are physical limits on how much compute you can squeeze into any given area. If your job requires more parallel compute than can reasonably be squeezed into a package for power/thermal reasons, then distributing the workload off-package is a requirement. Maybe hardware will morph into a more heterogenous mix than we have now, with many single-package cpu-gpu nodes working in parallel, instead of a few cpus orchestrating a gigantic sea of GPUs, but maybe not. I'm more convinced that this is just one stage of the cycle. GPUs in consumer hardware weren't really a thing before the 90s. From the 90s to like 2013 graphics software (and it's ability to load hardware) improved rapidly. Since then it's kind of stagnated and a lot more of the focus has shifted to doing the same amount of rendering with less power/less heat/less space. Even if we do see a shift toward SoCs/APUs/whatever in consumer hardware over the next few years, I'd bet on some sort of paradigm shift to come along (fully raytraced rendering?) and swing the pendulum pack toward big discrete GPUs at some point again.

xur17 3y ago

> And I also figured the crypto boom was a temporary unsustainable surge in the market for discrete gpus. It's kind of amazing how perfect the timing was for LLM popularity (right around the time Ethereum moved to PoS, and the crypto market cooled down).

scotty79 3y ago
I read that a lot of gpu miners went smoothly from mining to AI training.