bit-tech.net

Nvidia launches Quadro K6000 as world's fastest GPU

Nvidia launches Quadro K6000 as world's fastest GPU

Nvidia's latest Quadro K6000 is claimed to be the world's fastest GPU, and packs a whopping 12GB of GDDR5 to back up its throughput performance.

Nvidia has announced what it claims is the world's most powerful graphics card - but before you get excited, it's talking about a Quadro board aimed at workstation users rather than a GeForce part for gamers.

Unveiled at the SIGGRAPH conference in Anaheim, California late last night, the Nvidia Quadro K6000 is designed to take over from the existing Quadro 6000 - and, in doing so, offers a significant upgrade with five times higher compute performance than the card it replaces.

Based around Nvidia's Kepler architecture, the Quadro K6000 includes 2,880 CUDA cores connected to 12GB of GDDR5 memory over a 384-bit bus offering a claimed 288Gb/s of bandwidth. All this power is contained on a double-slot PCI Express 3.0 board, drawing a claimed 225W over two six-pin power connectors. Support for four simultaneous displays is included, over DVI-I, DVI-D, and dual DisplayPort 1.2 connections with support for 4K resolutions.

In the release of the Quadro K6000, Nvidia is claiming a couple of records: the most powerful GPU, and the largest graphics memory - and it's making sure everyone knows it. 'The Nvidia Quadro K6000 GPU is the highest performance, most capable GPU ever created for the professional graphics market,' crowed Ed Ellett, senior vice president of the company's Professional Solutions Group, at the unveiling. 'It will significantly change the game for animators, digital designers and engineers, enabling them to make the impossible possible.'

The company certainly had support for those claims: staffers from Pixar, Nissan, and Apache were all trotted out to offer soundbites about how access to pre-release Quadro K6000 boards had revolutionised their workflows - from Pixar's ability to preview more of a given scene in real time to Apache's claim of a three-fold boost to the performance of drilling simulation package InsightEarth compared to a Quadro K5000 board.

At the same event, Nvidia also announced new Quadro chips for laptop workstations starting with the headlining K5100M and working down through the K4100M, K3100M, K1100M, K610M and K510M parts.

Pricing for the hardware has yet to be confirmed, with Nvidia promising that its various customers will start offering boards in the autumn with prebuilt workstations confirmed from HP, Dell and Lenovo and add-in boards from PNY Technologies, ELSA, Leadtek and Ryoyo.

14 Comments

Discuss in the forums Reply
Maki role 24th July 2013, 12:16 Quote
Hmm is this what people originally labelled as the Titan Ultra? Certainly it shares the idea of having all the clusters activated on the GK110 chip. Wonder if they're planning to release a gaming version with all the clusters available? Can't imagine why they're need to at this stage though, not to mention the price would be eye watering.

Still, 12GB of memory is insane, that would be so incredibly useful. The difference between even 6 and 12GB is colossal when dealing with the kind of 3D workloads this card is intended for. Scenes will regularly break 6GB without a sweat, 12 will add an awful lot more headroom.
tozsam 24th July 2013, 14:52 Quote
I work in cg, but mainly stills. I have hardly ever seen a scene use less than 8-9Gb of ram! We have not been able to try realtime rendering (which uses the graphics card for rendering) up until this point because of that.
RedFlames 24th July 2013, 15:00 Quote
No word on price...

though, based on how much the high-end Quadros currently cost, I'll wager if you have to ask, you can't afford one...
r3loaded 24th July 2013, 15:16 Quote
Quote:
Originally Posted by Maki role
Hmm is this what people originally labelled as the Titan Ultra? Certainly it shares the idea of having all the clusters activated on the GK110 chip. Wonder if they're planning to release a gaming version with all the clusters available? Can't imagine why they're need to at this stage though, not to mention the price would be eye watering.
Yep, 2880 cores seems to indicate a fully unlocked GK110, making it even more powerful than the Titan or Tesla K20 (I wonder if they'll add a new Tesla equivalent to the K6000 though). You'll never see this chip appearing in a consumer card though as chip yields and binning will make it crazy expensive, and that's before they whack on the premium for being an ultra luxury product!
schmidtbag 24th July 2013, 15:18 Quote
Seems like nvidia finally cares a little more about workstation graphics. To me, it seems like they put all their time into desktop graphics, tablets, and CUDA servers, but not so much of workstations. AMD for a while has been a great choice for workstation graphics due to the often better performance for a significantly lower price (though this does depend on what programs you use).

I feel like a couple W7000s in crossfire would offer nearly as good of performance as this quadro but for a lower price, though, it would have 4GB less VRAM. However, I find it pretty hard to believe that more than 8GB of VRAM is *actively* in use. The quadro's performance likely comes from its cores. Workstations IMO are the best for multi-GPU setups, since stuff like micro-stuttering isn't much of a problem when it comes to things like rendering.
dream1 24th July 2013, 16:31 Quote
Just a titian with 12gb memory
Maki role 24th July 2013, 17:17 Quote
Quote:
Originally Posted by dream1
Just a titian with 12gb memory

Except that 12GB of memory is a pretty big deal and is much more useful. Not to mention the decreased TDP, increased core clock and increased CUDA core count...
greigaitken 24th July 2013, 21:51 Quote
"The company certainly had support for those claims: staffers from Pixar, Nissan, and Apache were all trotted out to offer soundbites about how access to pre-release Quadro K6000 boards had revolutionised their workflows"

Should have used multiples of previous model and had that revolution early.
dream1 25th July 2013, 00:38 Quote
Quote:
Originally Posted by Maki role
Quote:
Originally Posted by dream1
Just a titian with 12gb memory

Except that 12GB of memory is a pretty big deal and is much more useful. Not to mention the decreased TDP, increased core clock and increased CUDA core count...

yeah i know i work with 3D every day. But the price whats coming on that not for me. Probably over 1000$. I'll wait and see.
RedFlames 25th July 2013, 01:03 Quote
Quote:
Originally Posted by dream1
yeah i know i work with 3D every day. But the price whats coming on that not for me. Probably over 1000$. I'll wait and see.

Well considering the current range-topping quadro, the 6000 [no K], without any of the add-on gubbinz like Frame Lock is a smidge over £3000 [~$4800], I can't see it's replacement being any cheaper.


... I always keep that link handy for whenever a customer whines about the price of a GPU ...
SchizoFrog 25th July 2013, 02:51 Quote
Just out of curiosity, no need to get too technical but what is the difference/intended use between a Quadro and a Tesla based GPU?
RedFlames 25th July 2013, 03:06 Quote
Quote:
Originally Posted by SchizoFrog
Just out of curiosity, no need to get too technical but what is the difference/intended use between a Quadro and a Tesla based GPU?

Tesla cards can't/don't output video, they're purely for CUDA/OpenCL GPGPU based number crunching...

Quadros are to Geforce [FirePro/FireGL to Radeon if you're AMD inclined], what Xeons are to the Core series, they're tweaked/optimised for desktop 2d/3d-based grunt work suck as CAD and 3d Modelling...
They're generally under-clocked compared to their geforce counterparts, primarily to increase reliability, but the programs they're partnered with eat VRAM like it's going out of fashion [such 3d scenes with poly counts comfortably in the millions], hence the Quadro cards having 2x-3x the frame buffer of the consumer-grade Geforces... Quadros tend to be better at/with OpenGL rendering than Geforce Cards too... Finally they're geared towards driving hi-res, pro grade monitors, hence displayport being standard on pretty much every Quadro card of the past few years, long before DP started to appear on consumer cards...

They also support add-on cards like SDI input/output and Frame Lock most of which I have no clue what they do...
SchizoFrog 25th July 2013, 05:04 Quote
That is along the lines of what i was thinking but something you mentioned does raise another question for me. If Tesla cards can't/don't output video, how come the Titan cards and now the GTX780 and GTX770 cards are based on the Tesla K20 GPUs and not a Quadro part?
xaser04 25th July 2013, 09:58 Quote
Quote:
Originally Posted by SchizoFrog
That is along the lines of what i was thinking but something you mentioned does raise another question for me. If Tesla cards can't/don't output video, how come the Titan cards and now the GTX780 and GTX770 cards are based on the Tesla K20 GPUs and not a Quadro part?

This is the wrong way round to look at it.

All of the cards (Geforce, Quadro, Tesla) are based on the Kepler architecture. GK110 (in various bin states)being used for the Tesla K20 / Quadro K6000 and Titan/780 , whilst GK104 (again in various bin states) is used for the Tesla K10 (IIRC) / Quadro K5000 (again IIRC) and 770 and downwards.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums