Ampere, Nvidia's latest GPU architecture is finally here – spanking-new acceleration for AI across the board

Nvidia has lifted the lid on a fresh line of products based on its latest Ampere architecture, revealing its latest A100 GPU - which promises to be 20X more powerful than its predecessor and capable of powering AI supercomputers – as well as a smaller chip for running machine learning workloads on IoT devices. CEO Jensen Huang …

[Rameses Niblick the Third Kerplunk Kerplunk Whoops Where's My Thribble?]

That's all very impressive...

...but can it run Crysis?

[Anonymous Coward]

Re: That's all very impressive...

Yes but it can't run Doom... Original Doom... Dammit, that joke used to be funny.

[Mr Sceptical]

Re: That's all very impressive...

I was going to come in with the Crysis question, with a tangent:

Can it give you a realistic, learning, Turing test (of behaviour) passing, AI foe in Crysis?

And once it's beaten the best humanity can offer, time to power up the cyborgs and ROTM! --->

[Anonymous Coward]

Re: That's all very impressive...

Crysis....?

I'm not holding out much hope of it being able to run a modern desktop.

Still waiting on being able to run Wayland with Nvidia hardware....

Not very impressive at all if they can't cover the basics.

[Gonzo wizard]

OK ok so it's fast

But I wouldn't want the electricity bill. 400W per GPU at full utilisation. And the SuperPod - 1120 GPUs or 450Kw per hour at full utilisation... granted it is a more efficient platform than your regular CPU for ML but the heat... the running cost... :-

[Stephen 1]

Re: OK ok so it's fast

Yeah that's a lot of energy, 1.62e+9 Joules in fact. Then of course you need to run it for more than one hour.

[Boothy]

Re: OK ok so it's fast

Just as a comparison, some current GeForce RTX 2080 Ti cards pull over 300W at full load (reference cards are around 280W).

That's with a Turing card, so on the TSMC 12nm process. Ampere uses 7nm instead, and is still pulling ~33% more power!

[MiguelC]

Re: OK ok so it's fast

I bet NVidia will soon start marketing their brand new "Personal nuclear power plants (TM)"

[Ian Johnston]

Does anyone use these for graphics?

[Boothy]

Do you mean directly? As in a workstation/PC?

If so, then no, at least not as far as I know. The A100s' are specifically designed for data-centre usage. They don't even have video output on them.

But Ampere, the microarcitecture the new A100 is built on, is coming to workstation and mainstream cards at some point, we just don't know when yet.

For ref, nVidia have stated the Ampere based chips will replace all current mainstream consumer (i.e. regular GTX/RTX), prosumer (Titan) and professional (Quadro) cards. With the expectation being that the Titan and Quadro cards will be similar, if not the same as the chips being used in the A100, and the other cards being a cut down version (i.e. less CUDA cores etc).

[Mike 137]

Impressive, but ...

This is hardly just a "GPU" any more, is it? It may use the architectures developed for graphics processing, but it's more likely destined for deep data and AI rather than merely putting up animated screenfulls.

[steelpillow]

Re: Impressive, but ...

I believe that the G nowadays stand for General rather than niche Graphics (which it presumably also does rather well).

[Anonymous Coward]

No love for AMD's new Radeon VII Pro then?

[Nursing A Semi]

But

As far as shareholders are aware, all these A100s are going to be sold to gamers to power their rigs for the next generation of mincraft surely?

[Marcelo Rodrigues]

Re: But

"As far as shareholders are aware, all these A100s are going to be sold to gamers to power their rigs for the next generation of mincraft surely?"

That's where the profits come from, you know. They will only sue if someone starts mining cryptocoins with it...