4 ms·

The world market is bigger than 5 systems. I'd argue it looks more like relational databases in ~1980. Most people to this day do not know what a relational dat

The world market is bigger than 5 systems. I'd argue it looks more like relational databases in ~1980. Most people to this day do not know what a relational database is, but they probably use one multiple times a day. Over the years, databases have gotten bigger, faster, more complex, more powerful, easier to use, and more open. Deep learning--and machine learning more generally, because that's what we're really talking about here--is going to go through the same process. NVIDIA is positioning themselves to be the vertically integrated supplier in this market. The really interesting part is that no one is sure yet what kinds of applications these tools will open up -- it is likely to be an enabling platform/toolset in much the same way relational databases, app stores, or IaaSs were. AMD and Intel are woefully behind. Intel because it's not yet a big enough market ($137B market cap compared to NVIDIA's $12B, Intel needs an opportunity to be literally 10x as interesting to be interested). Why AMD is not all over this I'm not sure... If it's interesting for NVIDIA, it should be very interesting to a $1.6B operation like AMD.

ris 11y ago
"Why AMD is not all over this I'm not sure..." NVidia have been very good at encouraging the use of their proprietary CUDA. They also have a slight FLOPS advantage I'm told (where AMD have an integer advantage)
- pjmlp 11y ago
  It helps when CUDA allows for C++, Fortran and any other language with a compiler that spits PTX, while OpenCL stayed for so long as stone age C.
_delirium 11y ago
> Deep learning--and machine learning more generally, because that's what we're really talking about here For Nvidia's market, imo we have to be careful to separate out which kind of ML specifically. Certain kinds of training of deep networks get large speedups from GPUs, which isn't true of all ML techniques. So if Nvidia wants to get a big win in this space by being a vertically integrated tool supplier leveraging GPU hardware, the strategy is tied to a bet on a specific subset of ML techniques, those where the ratio of GPU:CPU performance is especially large.
redacted 11y ago
> Why AMD is not all over this AMD are barely treading water in the markets they already compete in. Both Intel (CPU) and NVIDIA (GPU) have stolen back enormous mind and market share [0]. I doubt they could afford the outlay of capital and research time to build something competitive. Maybe if they were in better shape financially a braver board/CEO could pivot them away from the CPU/GPU markets they've lost to something different. Their current road-map takes AMD through to 2016, after that who knows. [0] NVIDIA's 9XX series GPUs have been incredibly aggressively priced and marketed. AMD effectively ceded the high end consumer (gaming etc) market since September last year. Gaming parts are big PR for both companies, so it's quite a bad sign IMO.
- ris 11y ago
  "Maybe if they were in better shape financially a braver board/CEO could pivot them away from the CPU/GPU markets they've lost to something different." Yes, wouldn't it be great to let the market leaders in these sectors have a total uncontested monopoly...
dmd 11y ago
> The world market is bigger than 5 systems. He's referencing this: https://en.wikipedia.org/wiki/Thomas_J._Watson#Famous_misquote https://en.wikipedia.org/wiki/Thomas_J._Watson#Famous_misquo...
creshal 11y ago
> Why AMD is not all over this I'm not sure... They're barely staying alive, and have been axing branches left and right to "focus". Like selling off their smartphone graphics branch a year after the iPhone was announced… it's now rather successful under Qualcomm (Adreno is an anagram of Radeon).

ris 11y ago

"Why AMD is not all over this I'm not sure..." NVidia have been very good at encouraging the use of their proprietary CUDA. They also have a slight FLOPS advantage I'm told (where AMD have an integer advantage)

pjmlp 11y ago
It helps when CUDA allows for C++, Fortran and any other language with a compiler that spits PTX, while OpenCL stayed for so long as stone age C.

_delirium 11y ago

> Deep learning--and machine learning more generally, because that's what we're really talking about here For Nvidia's market, imo we have to be careful to separate out which kind of ML specifically. Certain kinds of training of deep networks get large speedups from GPUs, which isn't true of all ML techniques. So if Nvidia wants to get a big win in this space by being a vertically integrated tool supplier leveraging GPU hardware, the strategy is tied to a bet on a specific subset of ML techniques, those where the ratio of GPU:CPU performance is especially large.

redacted 11y ago

> Why AMD is not all over this AMD are barely treading water in the markets they already compete in. Both Intel (CPU) and NVIDIA (GPU) have stolen back enormous mind and market share [0]. I doubt they could afford the outlay of capital and research time to build something competitive. Maybe if they were in better shape financially a braver board/CEO could pivot them away from the CPU/GPU markets they've lost to something different. Their current road-map takes AMD through to 2016, after that who knows. [0] NVIDIA's 9XX series GPUs have been incredibly aggressively priced and marketed. AMD effectively ceded the high end consumer (gaming etc) market since September last year. Gaming parts are big PR for both companies, so it's quite a bad sign IMO.

ris 11y ago
"Maybe if they were in better shape financially a braver board/CEO could pivot them away from the CPU/GPU markets they've lost to something different." Yes, wouldn't it be great to let the market leaders in these sectors have a total uncontested monopoly...

dmd 11y ago

> The world market is bigger than 5 systems. He's referencing this: https://en.wikipedia.org/wiki/Thomas_J._Watson#Famous_misquote https://en.wikipedia.org/wiki/Thomas_J._Watson#Famous_misquo...

creshal 11y ago

> Why AMD is not all over this I'm not sure... They're barely staying alive, and have been axing branches left and right to "focus". Like selling off their smartphone graphics branch a year after the iPhone was announced… it's now rather successful under Qualcomm (Adreno is an anagram of Radeon).