Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

blogs.nvidia.com

Post reply on HN
Print view View original

261–270 of 347 posts

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#261

Post by omgJustTest »

dijit wrote: post #18
drewg123 wrote: post #10

I'm confused. Is there any relationship between the recent Ampere Arm64 servers ( https://news.ycombinator.com/item?id=22475036 ) and Nvidia's "Ampere Architecture", or is it just a case of them using the same name?

I don't like that people downvoted you for asking a question. If someone thinks the question is stupid or not doesn't mean that a downvote is warranted. (nor an upvote, answer the question and move on.) To answer though; it's just a coincidence, as you might already know Nvidia uses famous scientists (especially in the field of electricity) as the names of their microarchitectures. * Volta (Alessandro Volta, inventor…

Likely this is more confusion sowing by NVIDIA. While yes, they are names of scientists, in general they are confusing as similar products usually exist within their microelectronic subdomain.

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#262

Post by trsohmers »

person_of_color wrote: post #257
trsohmers wrote: post #119

Earlier quoted context omitted.

Well considering they are filing for bankruptcy you are going to be waiting a while...

Source?

A very simple google search would suffice, but here: https://www.eetimes.com/wave-computing-files-for-chapter-11-...

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#263

Post by BeetleB »

adev_ wrote: post #121
fluffything wrote: post #94

Earlier quoted context omitted.

> They make good hardware, but there's a lot of lock-in, and not a lot of transparency. This sounds like you'd like NVIDIA to open-source all their software. I see this type of request a lot, but I don't see it happening. NVIDIA's main competitive advantage over AMD and Intel is its software stack. AMD could release a 2x powerful GPGPU tomorrow for half the price and most current NVIDIA users wouldn't care because wh…

> AMD could release a 2x powerful GPGPU tomorrow for half the price and most current NVIDIA users wouldn't care because what good is that if you can't program it?. Correction: Nobody will be able to use the AMD hardware (outside of computer graphics) because everybody has been locked-in with CUDA on Nvidia. They can not even change even if they want to: it is pure madness to reprogram an entire GPGPU software stack e…

>> ARM and Intel make great software [..] doesn't open-source any of that either for the same reasons as NVIDIA.

> That's propaganda and it's wrong.

Very convenient of you to have omitted what was in the square brackets:

> Intel MKL, Intel SVML, ... libraries, icc, ifort, ... compiler

Show me the open source MKL, Intel SVML, icc and ifort.

Some (all?) of it may be free, but it's not open source.

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#264

Post by catblast »

ska wrote: post #158
blattimwind wrote: post #39

Earlier quoted context omitted.

Tesla is famous in the US for inventing polyphase AC and induction motors, but this is really one of these stories were a bunch of people invented the same thing very closely to each other due to a precipitating reaching of understanding. Note that Tesla's designs were IIRC two-phase which is largely inferior to three-phase. The push for three-phase and associated designs and inventions (three phase transformers on a…

> but this is really one of these stories were a bunch of people invented the same thing very closely to each other due to a precipitating reaching of understanding. This is by far the dominant case of invention. Truly independent work is incredibly rare.

The thing that's a little different about Tesla in the US at least, is he is so incredibly fetishized by eg high-profile idiots: https://theoatmeal.com/comics/tesla, or conspiracy nutjobs like the International Tesla Institute (http://teslatech.info/ttevents/prgframe.htm http://tesla.org/tesla_fair_abq.htm) publishing and promoting Tesla related conspiracies and hawking investments in snake-oil technology like Rand Cam engines, VMSK, and all manner of "over-unity" machines

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#265

Post by tbenst »

XCSme wrote: post #259
tbenst wrote: post #157

Important to remember the half-precision tensorcore misrepresentations where the 8x improvement over fp32 claimed on Imagenet with tensorcores (V100) was actually only 1.2-2x [1,2]. Furthermore, there are major precision issues with network architectures like variational autoencoders and many others. We use V100s for Richardson-Lucy like deconvolutions for example, where we have near-exact photon counts up to 10,000…

I am not that much into ML, just fiddled with it a bit, is tf32=fp16?

Not quite, but close. “tf32” is 18 bits, but with the same 10 bits of exponent that fp32 has. It’s the range of fp32 with the precision of fp16. It’s a shame to see such unoriginality in new number representations. I’d much rather see Posit hardware acceleration: https://web.stanford.edu/class/ee380/Abstracts/170201-slides...

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#266

Post by slaymaker1907 »

XCSme wrote: post #259
tbenst wrote: post #157

Important to remember the half-precision tensorcore misrepresentations where the 8x improvement over fp32 claimed on Imagenet with tensorcores (V100) was actually only 1.2-2x [1,2]. Furthermore, there are major precision issues with network architectures like variational autoencoders and many others. We use V100s for Richardson-Lucy like deconvolutions for example, where we have near-exact photon counts up to 10,000…

I am not that much into ML, just fiddled with it a bit, is tf32=fp16?

No, tf32 has the same size exponent field as f32 but the mantissa size of f16, 10 bits.

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#268

Post by nostrademons »

m463 wrote: post #248
chrischen wrote: post #193

Earlier quoted context omitted.

And yet we grant patents so liberally, giving a windfall to the first person who files.

Remember the long view - patents cause people to hurry to publish and share their ideas publically. Why shouldn't they be granted liberally. In a few years, the temporary monopoly falls away and the benefit passes to everyone. I think they should work to make them even cheaper and easier to file.

There's a strong adverse selection effect, though. Because you need to publish to be granted a patent but can sue whenever anyone infringes (whether willful or not), the incentive is to patent obvious approaches that don't work well and hold the best approach that you're actually using as a trade secret. That way, anyone attempting to replicate you likely ends up in a patent minefield, yet you don't give away the keys to the castle in a patent where you have to detect infringement yourself.

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#269

Post by bigdict »

wegs wrote: post #122
bigdict wrote: post #52

Earlier quoted context omitted.

Have you heard about CuPy?

Yes. I won't bet my business on a one-vendor solution with a medium-sized community which might disappear at some point. If CuPy supported NVidia and AMD, and was folded into Numpy, I'd buy the biggest, beefiest GPU I could find overnight.

CuPy has experimental support for Rocm.

Re: Nvidia CEO Introduces Nvidia Ampere Architecture, Nvidia A100 GPU

#270

Post by p1esk »

tbenst wrote: post #157

Important to remember the half-precision tensorcore misrepresentations where the 8x improvement over fp32 claimed on Imagenet with tensorcores (V100) was actually only 1.2-2x [1,2]. Furthermore, there are major precision issues with network architectures like variational autoencoders and many others. We use V100s for Richardson-Lucy like deconvolutions for example, where we have near-exact photon counts up to 10,000…

Why is fp32 sufficient but not tf32 for that task?
Post reply on HN