Hardware Nvidia's Hybrid SLI technology

[Tim S]

http://www.bit-tech.net/hardware/200..._sli_preview/1

On the morning of CES, Nvidia announced its Hybrid SLI technology. Rather than rushing out a quick article, we decided to spend time doing some real analysis of what was discussed. Read on to find out how this exciting technology is shaping up.

[notatoad]

the single DVI port is going to kill this thing. judging by the post your workstation thread, most of the target market for this has two or more monitors hooked up.

i can really see this being a killer feature for gaming laptops though. drop into a low-power graphics chip to stretch the battery on the road, and then fire up the powerful card to play some games when you plug it in.

[Woodstock]

i was thinking at the start an article, great i know will have to pay for a chip to save power, but towards i could start to see the bennifits, but i think ill wait till i see some numbers

[Tim S]

Nvidia said that costs shouldn't increase massively with the IGP on every motherboard.... but if you think about it the other way: how much are you spending on electricity (assuming you contribute towards the bill) for a high end GPU to sit idle?

[naokaji]

i think the idea is good, less power consmption and less noise for pretty much the same price, shame about it though that it wont work with any nvidia graphicscard.

[[USRF]Obiwan]

I also do not understand why they want to include a DVI and a VGA connector. It is totaly useless. Why not 2 DVI or maybe 2 new Displayport type connectors with both dpp to dvi/vga converters.

And another thought: why not build a lowpower extra 'litle sister' GPU on the graphicscard itself. So when raw power is needed the 'bigdaddy' GPU kicks in on the graphicscard. This also is solution for the "the benefit of those extra 16 stream processors is going to be very small and it could actually act as a decelerator" part, because its all sitting on the same card and uses the same memory.

nontheless it is a very good article Tim

[Tim S]

Originally Posted by naokaji

marketing folks in action...

no single graphicscard uses that much power, and same goes for sli, but if they took complete system power usage then the 0W is wrong...

anyway, i think the idea is good, less power consmption and less noise for pretty much the same price, shame about it though that it wont work with any nvidia graphicscard.

Actually, you're forgetting 3-way SLI - 1200W PSU minimum 3x 185W+ 8800 Ultras = pretty close to 600W. Of course there's a bit of marketing involved, but taking GPU power draw in 3-way SLI from well over 500W to next to nothing actually made me thought for a second time about putting that kind of insanity in my own system.

[Woodstock]

Originally Posted by Tim S

Nvidia said that costs shouldn't increase massively with the IGP on every motherboard.... but if you think about it the other way: how much are you spending on electricity (assuming you contribute towards the bill) for a high end GPU to sit idle?

thats good then, and not owning a high end gpu i have no idea what they cost to sit idle

[naokaji]

nvm

[Tim S]

it says SLI load and SLI idle in the picture you posted.

[Woodstock]

wouldnt it be a better idea to make high end cards use less power

[naokaji]

Originally Posted by Tim S

it says SLI load and SLI idle in the picture you posted.

too true my bad... somehow i'm seeing things that arent there

[Omnituens]

Im just looking at that board - 3 SLi cards = no free PCI slots.

There goes my X-Fi card

[Cupboard]

There was something similar to this a little while ago - it was a laptop (I think Alienware) that had both integrated and discrete graphics - when you wanted to game you switched on the graphics card, otherwise you got a better battery life with the integrated. I can't find it anywhere though.

[Paradigm Shifter]

Originally Posted by Cupboard

There was something similar to this a little while ago - it was a laptop (I think Alienware) that had both integrated and discrete graphics - when you wanted to game you switched on the graphics card, otherwise you got a better battery life with the integrated. I can't find it anywhere though.

Various laptop ODM's have done this in the past. Rock Direct had their Pegasus line using that idea of discrete/integrated - choose between the two with a switch a while ago. Not needing a new laptop, though, I don't know whether that's still an option.

[djDEATH]

what i see as good in this is the ability to buy a decent motherboard with onboard graphics. Currently, they're all sh*te, and forces you to buy a discrete GPU just to get your machine working. In my case, it forced me to buy a 6600GT when really all i wanted was onboard until i could afford a decent card.

At least now, people wanting onboard graphics can also enjoy the benefits of a fully fledged chipset, and also drop in a cheapo 8500GT to more than likely double its performance.

[Aterius Gmork]

this is cheesecake!

[Tyinsar]

Originally Posted by [USRF]Obiwan

...
And another thought: why not build a lowpower extra 'litle sister' GPU on the graphicscard itself. So when raw power is needed the 'bigdaddy' GPU kicks in on the graphicscard. This also is solution for the "the benefit of those extra 16 stream processors is going to be very small and it could actually act as a decelerator" part, because its all sitting on the same card and uses the same memory.

nontheless it is a very good article Tim

Quoted For Truth. Your suggestion also eliminates complex interfaces between the motherboard & graphics card and even the need for an nvidia motherboard - maybe that's why not. !@#/$%& marketing

[ryanjleng]

Oh, this is insanely difficult to get things right for the BIOSes and Drivers. All the combinations of things that can go wrong.

Love the power efficiency intention.... but adding mGPU on every chipset to promote Hybrid-SLi? Why not step down the discrete GPU instead? and build all the power efficiency features into a PCI-X 2.0 card. Hybrid between PCI-X slots... are we there yet?

As to the argument of using the mGPU for a centralized fallback GPU and power control... won't it be easier to do it on a discrete cards.

Could there be fundamental GPU die design contraints at play? Probably they can't step down on power without get hit on the best performance due to additional control logics consuming on the GPU die real estate.

What cost most on a motherboard is ICs and connectors. This i think is a case of bundling for new solutions.

There are so many questions and angles.... i'm nauseated.

Is NVidia reacting to the coming GPU integration into the CPU die by INTEL and AMD?

Thanks for the article. Fantastic read.

[willyolio]

wouldn't passing the GPU's processing through the motherboard degrade performance when you really need it?

i was hoping there would be the other way around, like having some kind of direct pass-through the main GPU, which would still let you turn it almost completely off.