New Hydra Technology to Eclipse SLI and CrossFire

[gigabit]

http://news.softpedia.com/news/-90222.shtml

[IWasHungry]

Hmmm I'll believe it when I see it.

[Shiggity]

Perfect scaling eh?

I'll believe it when I see it +1 ;)

[CGR]

Ehhhhhhhhhhhhhhhhhhhhh

+1 for the believe when see....

[White_Pawn]

Hmmm I'll believe it when I see it.

yeah. +1. Sounds like sketch.

[Mr.Guvernment]

It is uncertain whether the new technology will actually work, but if it does, computer users will be faced with an alternative to today's high performance rigs.


+ infinty when i see it :) but sounds very interesting!

[Mpegger]

Wasnt something like this tried many years ago when SLI was first introduced with the 3DFX cards? I could have sworn some company did the software SLI thing but it actually hurt performance and never worked correctly (only supported a few models).

Even with a hardware/software solution, I cant really see it being that effective, and being limited in availability unless its cheap to incorporate into the motherboards without massive re-engineering.

[4GHZ_or_bust]

Still early. And still plausible but I need to see some real stuff, not some half baked theories that looked good on paper.

In the meanwhile, I hear there's a new dual CPU mobo that supported both AM2+ and C2D CPUs and you can have 2 of AMD< 2 of Intel, or 1 of each running at the same time.

[ratbuddy]

I think I saw something about this a few days ago on the Inquirer or Fudzilla. It's an addon card that somehow coordinates your graphics cards and spits the output to your monitor. I wouldn't be too quick to call it farfetched, it's possible this thing does something like takes the render instructions from your games, splits them off into a parallel friendly workload for the cards, snarfs back up the output, and reassembles it for display..

Not saying I think it'll be awesome or anything, just saying it's possible...

[Hardin]

Is it ok to call bs on this?

[gigabit]

Is it ok to call bs on this?
NO its not..lol

[Hardin]

NO its not..lol

I guess we'll find out.

[gigabit]

I guess we'll find out.
Seriously i doubt if we will ever see this

[Neural Net]

It's not BS seeing as Intel owns LucidLogix.

[ratbuddy]

It's not BS seeing as Intel owns LucidLogix.

No they don't. Intel Capital just has a stake in them. Big difference.

[Neural Net]

No they don't. Intel Capital just has a stake in them. Big difference.

Well anyway, Intel wouldn't invest in this technology if it were pie in the sky stuff, so no, it's not really different.

[ratbuddy]

Well anyway, Intel wouldn't invest in this technology if it were pie in the sky stuff, so no, it's not really different.

Owned by Intel would mean Intel fully controlled the company and that really is 'really different' from Intel making an investment in a potentially cool technology company..

Quote from http://www.lucidlogix.com/about/investors.html:

---
Intel Capital
With an overall strategy to enable innovation, Intel Capital seeks out and invests in promising technology companies worldwide. We focus on both established and new technologies that help to develop industry standard solutions, drive global Internet growth, facilitate new usage models, and advance the computing and communications platforms.

As part of Intel Corporation, Intel Capital calls on some of its best and brightest to evaluate prospective investments, offer business and technology guidance to our portfolio companies, and provide unrivaled access to the latest developments in the industry. We are among the largest venture capital entities in the world with offices in established and emerging markets around the world.

Since 1991, Intel Capital has invested more than US$7.5 billion in approximately 1,000 companies in 45 countries. In that timeframe, 168 portfolio companies have gone public on various exchanges around the world and 212 were acquired or participated in a merger. In 2007, Intel Capital invested about US$639 million in 166 deals with approximately 37 percent of funds invested outside the United States.

Intel Capital has made a number of well known investments around the globe. These include Actions Semiconductor, Bellrock Media, Broadcom, CNET, CitrixSystems, Clearwire, Elpida Memory, FPT, India Infoline.com, Inktomi, Insyde Software, Integrant Technologies, Marvell, MySQL, NIIT, PCCW, Red Hat, Rediff.com, and Research in Motion, Sasken, SiRF, Smart Technologies, Sohu.com, Techfaith, VA Linux, and WebMD.
---

Yes, it's encouraging that Intel Capital thinks this company might be up to something worthwhile, but it's by no means a sure thing :)

[Neural Net]

Well yes that's what I was getting at, Intel wouldn't put any money into something that doesn't have actual potential. Their research departments will know a lot more than you or I so it's pretty safe to say that if Intel has invested in this technology, there's a high probability that this works, but it may not be cost effective.

[reclaimer122]

If this were an add-on card, wouldn't we see massive latencies?

[ratbuddy]

If this were an add-on card, wouldn't we see massive latencies?

Perhaps not. A single frame of 1920x1200x32bit is at most 73728000 bits of information. In total, at (say) 100 FPS, you're passing 7372800000 bits to or from this card. That works out to 921 megabytes or so (I'm using hard drive maker math) per second. That's well within the range of a normal PCIe x4 slot.

My guess is this card is going to just accept rendered pieces of the screen from each GPU and throw them back together. The math does get a little ugly at 2560x1600, needs almost twice the bandwidth, or a PCIe x8 slot, still doable though..

[Shiggity]

Whereareu shared frame buffer for multi-gpu cards.

[deathman20]

Seems odd, but I'd like to know how it could use both an nVidia card and an ATI card at the same time expecially with driver conflicts.

Yeah I saw this a few days ago as well. Saying same 100%+ scaling or something like that, which doesn't make sence how you could scale more than 100%.

[Shiggity]

Seems odd, but I'd like to know how it could use both an nVidia card and an ATI card at the same time expecially with driver conflicts.

Yeah I saw this a few days ago as well. Saying same 100%+ scaling or something like that, which doesn't make sence how you could scale more than 100%.

It's possible to have over 100% scaling, but that would mean the Hydra engine is actually accelerating how fast the cards are talking to one another, which seems very doubtful.

Like how the 4870x2 can beat 2 4780's in crossfire because of the slightly better communication between the GPU's via the 4870x2. Or how IBM's cell processors accelerate how fast the Opterons work in the Roadrunner Super Computer.

I agree with you though on the driver issue, how could it be possible to accelerate how well the cards can exchange data with different drivers, different types of ram, different clocked cores, etc. If it works it'd be pretty amazing.

[Mpegger]

I'm gonna guess, that it works by acting as the primary video device, with the actual video cards working as slaves. There sould be a way to render graphics in a interlaced fashion built into most modern video cards, so it may force 1 video card to render only odd lines, and the 2nd to render only the even lines. This way, even though you may be running at a high resolution, the horizontal resolution would be half, in effect allowing each video card to work faster since its processing less data.

Would be interesting to see if it actually works, but I dont think it would be any better (or cost effective) compared to the current Cross-Fire and SLI.

Maybe Intel is more interested in being able to support multiple video cards without having to license anything from Nvidia or ATI on thier chipsets.

[ratbuddy]

I'm gonna guess, that it works by acting as the primary video device, with the actual video cards working as slaves. There sould be a way to render graphics in a interlaced fashion built into most modern video cards, so it may force 1 video card to render only odd lines, and the 2nd to render only the even lines. This way, even though you may be running at a high resolution, the horizontal resolution would be half, in effect allowing each video card to work faster since its processing less data.

Would be interesting to see if it actually works, but I dont think it would be any better (or cost effective) compared to the current Cross-Fire and SLI.

Maybe Intel is more interested in being able to support multiple video cards without having to license anything from Nvidia or ATI on thier chipsets.


I think Intel is probably more likely to be interested because of the many-core nature of Larabee. They might have a way to offer legacy support here without having to write new drivers for every single game.

[Shiggity]

Bump, more news about it on Anandtech

http://www.anandtech.com/tradeshows/showdoc.aspx?i=3379

http://www.dailytech.com/article.aspx?newsid=12719

100% scaling might become reality! :D

(Only drawback is that Windows OS's don't support multiple display drivers = no ATI / Nvidia solutions until Microsoft changes that). But if this solution actually does come out and work, it'll just destroy current x-fire and SLI solutions.

Not to mention Larabee could come out and already be perfectly scalable across multiple GPUs.

[deathman20]

Not to mention Larabee could come out and already be perfectly scalable across multiple GPUs.

Its a key reason why Intel is investing some major dollars into that company. So they can get the benifits of multicard solutions without screwing with drivers or controllers to do such.

[Shiggity]

Its a key reason why Intel is investing some major dollars into that company. So they can get the benifits of multicard solutions without screwing with drivers or controllers to do such.

Yup, really smart plan ;) Got to love having the cash flow of Intel

[Kuroimaho]

Really hope that this delivers looks like the real Multigpu solution.

[gigabit]

anyone see this
http://www.dailytech.com/Chipmaker+Hydras+Stunning+Work+May+Render+CrossFir e+SLI+Obsolete/article12719.htm

[ratbuddy]

Its a key reason why Intel is investing some major dollars into that company. So they can get the benifits of multicard solutions without screwing with drivers or controllers to do such.

Damnit I said THAT a month ago too. Why are we doing a new thread on this :p

[mjw21a]

Lol, I hope it works. I wouldn't mind a mainboard with this chip, integrated graphics + my 4850..... Would be a nice upgrade. Will wait to see what happens with this before updating my CPU, RAM & mobo......

[Neural Net]

If this works, it means Nvidia and ATI have been extremely lazy with their multi GPU solutions, perhaps deliberately so. :p

[mjw21a]

If this works, it means Nvidia and ATI have been extremely lazy with their multi GPU solutions, perhaps deliberately so. :p

Hmmmm, reminds me of a few articles stating that the graphics industry has been relatively sheltered without any particularly fierce competition..... Articles relating to Larrabee.....

If it works then I'd welcome the demise of both SLI and Crossfire. A single solution for all would be much preferreable to two separate solutions.... Especially if it performs better.

[Shiggity]

If this works, it means Nvidia and ATI have been extremely lazy with their multi GPU solutions, perhaps deliberately so. :p

I don't think they dliberately had something better and refused to release it. They just didn't want to spend more R&D money on something that already worked (semi-decently well heh).

Now we'll be able to have Havok physics and PhysX in one machine, ATI and Nvidia :D

This is kind of off-topic, but did anyone see the new MATRIX GPU's from Asus? They say they will allow you to change voltages on the memory and core in the OS, (no need to do hard / soft mods anymore). Speculation is that the overclocking is insane on them.

[deathman20]

This is kind of off-topic, but did anyone see the new MATRIX GPU's from Asus? They say they will allow you to change voltages on the memory and core in the OS, (no need to do hard / soft mods anymore). Speculation is that the overclocking is insane on them.

Thats news to me. Is that ATI or nVidia cards? I know the newer ATI ones everyones basically saying stay away due to well they suck for OCing.

[Rinne]

The guys at extreme tech "tested" it at IDF and were quite impressed..

from the articles it seems this could really be a good thing, especially if it's built into the graphics cards or mainboards, so you don't have to buy a seperate extension card..

[EDIT] sorry, forgot to post the link :P

http://www.extremetech.com/article2/0,2845,2328495,00.asp

[CGR]

The guys at extreme tech "tested" it at IDF and were quite impressed..

from the articles it seems this could really be a good thing, especially if it's built into the graphics cards or mainboards, so you don't have to buy a seperate extension card..

It would be built into motherboards from what I read.

[deathman20]

It would be built into motherboards from what I read.

Or GPU's or Add-in card. Perfer mobo or add-in card or messing up ideally then we wouldn't have to worry about more stuff on the GPU messing with our clocks :)

[CGR]

Are you sure it could be built into cards? I thought the point was to have it done on the motherboard, where it could then split everything and send it to the cards.

[ajrettke]

Well I guess it could be built into a card....but it would be smarter to have it on the mobo. This prevents having 4 GPU's with 4 lucid chips, which would make 3 of them useless (since one chip is gonna be the master).

What I think would be interesting to see if this takes off is utilizing ATI and nVidia cards in the same platform. In theory (this is just my base knowledge level theory) driver optimizations on the lucid could allow each card to shine at what it's better at...i.e. physics calcs get sent to nvidia, poly's to ATI, and whatever card has better shaders goes to that card or whatever works best.

Intel probably wants this to work because if they want to use SLi they have to pay for the right or a chip which allows SLI to work...I can imagine intel is not impressed with nvidia charging for that especially when ATI offers Xfire for free.

[Hardin]

I guess this will also mean that intel's nehalem motherboards wont have or need the nforce 200 chip.

[Shiggity]

I guess this will also mean that intel's nehalem motherboards wont have or need the nforce 200 chip.

Unfortunately the early boards already have them. At least Asus's new X58 has it, said it could do x-fire and SLI in the specs.

If this Hydra thing really takes off with mobo makers, Nvidia is going to lose a lot of money not being able to sell their nforce 200 anymore. Or will you need the Lucid chip AND the nforce chip to make SLI work?

[Hazaro]

Tech Report article as well.

http://techreport.com/articles.x/15367

[Mpegger]

This is starting to look good. First quarter 2009 is starting to look like a good time to upgrade again, without having to relegate the GTX to a 2nd system, and instead put it to good use. Hopefully, they will release mobos for Core2Duos with this chip, as I'd like to only have to purchase a new video card and mobo if at all possible to save money yet maximize performance.

[Shiggity]

Anandtech also did a good article on it - http://www.anandtech.com/video/showdoc.aspx?i=3385&p=1

http://images.anandtech.com/reviews/video/lucid/chip.jpg

I want one ;)

[ratbuddy]

If this Hydra thing really takes off with mobo makers, Nvidia is going to lose a lot of money not being able to sell their nforce 200 anymore. Or will you need the Lucid chip AND the nforce chip to make SLI work?

This thing renders SLI obsolete.

No pun intended.

[gigabit]

This thing renders SLI obsolete.

No pun intended.


Sure looks that way if this ever comes to market

[mjw21a]

Hmmmm, can't say I'd ever bother with a second video card, though I WOULD like to try running my current card + a little fps boost from integrated ATI/nVidia graphics. Very nice :santa2:

[gigabit]

Hmmmm, can't say I'd ever bother with a second video card, though I WOULD like to try running my current card + a little fps boost from integrated ATI/nVidia graphics. Very nice :santa2:

Yeah but when you go to buy a new video card and say you dont sell your old card atleast you could still use it .And it doesnt matter if your old card is ATI or nvidia.Thats pretty awesome in my opinion

[Rinne]

Regarding that ATI and NVIDIA go the way of Multi GPU cards, I think this could even become necessary...

[Mpegger]

Could, but if I'm reading all those articles correctly, you cant mix different company (Nvidia & ATI) cards together because of Windows inability to have more then 1 Direct3D driver active simultaneously. The same goes for cards from the same company with different driver versions.

This means with Nvidias unified drivers, you can mix and match different generation cards easily, so long as the one driver installed supports both (or more) of the GPUs installed.

Theoretically, it looks like the Hydra could support cards from any manufacturer together, but because of Windows, you'll be forced to use just one manufacturer because of the need to use just one driver for all the video cards in 3D mode. So no onboard video would be usuable with a PCIe video card unless thier from the same manufacturer.

[>HyperlogiK<]

Didn't Alienware try something like this in 2001, I think they called it 'Video Array'

[mjw21a]

Yeah but when you go to buy a new video card and say you dont sell your old card atleast you could still use it .And it doesnt matter if your old card is ATI or nvidia.Thats pretty awesome in my opinion

I agree entirely. It means with the next update you're not restricted to one manufacturer. Great for me as I've no loyalty to any manufacturer. They're in it for profits, I'm in it for best bang for buck. :beer:

[Mpegger]

Ahem... try reading the articles and my post again (or for the first time). You currently cannot mix different manufacturers cards (Nvidia with ATI) because of Windows inability to have more then 1 Direct3D driver active at any one time. This includes different driver versions from the same manufacturer.

[mjw21a]

I hope this changes soon..... Looking at moving to OSX soon anyways.....

[stasdm]

Excellent idea! :clap:

But NO ATX/eATX mobo will take 4 graphic cards!
They will have to go into longer m/b to fully utilise x48 or 5400 connectivity (first PCIe v.2 - 4 graphic cards - 8 slots, second PCIe v.2 - two x8 connections, and a couple of south bridge expantions slots for additional USB's or what else) - with standard ATX on-board connections it will fit only into the 20-slots long mobo. :argue:

[Hardin]

Excellent idea! :clap:

But NO ATX/eATX mobo will take 4 graphic cards!
They will have to go into longer m/b to fully utilise x48 or 5400 connectivity (first PCIe v.2 - 4 graphic cards - 8 slots, second PCIe v.2 - two x8 connections, and a couple of south bridge expantions slots for additional USB's or what else) - with standard ATX on-board connections it will fit only into the 20-slots long mobo. :argue:

These are all atx motherboards. http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=1494436126&Description=790fx&name=4

[Mpegger]

ASUS and MCI (not to mention AMD cpu bound)..... not exactly something I'd want to use or even have.

[jobrien2001]

In the short term I would buy this new tech sitting in between a 4870x2 and not have to worry about a new mobo... until I change my setup, and the tech gets better. Perfect scaling on a 4870x2 would rock.

[korruptedONE]

So, say you COULD run an ATI and an Nvidia card at the same time with this chip.

Won't you get different splotches of image quality based on what part of the scene is rendered by which GPU? I'm not trying to start an image quality debate but with different drivers drawing different parts of the scene, wouldn't you see differences, even if they are subtle?

Of course all we can do is speculate but to me it seems that you would see noticable differences in the rendered scene.

[stasdm]

These are all atx motherboards. http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=1494436126&Description=790fx&name=4

Ya, but they are using both PCIe native links, Hydra will need inly one. Are you going to pay for "theoretically available" second PCIe x16?

And can you use even south-bridge connections with 4 graphic cards installed?

Might be good for gaming machine, but what about serious workstation?

[stasdm]

So, say you COULD run an ATI and an Nvidia card at the same time with this chip.


Not to-morrow (might be never)

[korruptedONE]

Not to-morrow (might be never)

That wasn't the point...

I think I answered my own question though: http://www.anandtech.com/video/showdoc.aspx?i=3385&p=3

"After the workload is distributed, the buffers are read back to the Hydra chip and composited before the final scene is sent to the proper graphics card for display. Looking a bit deeper, here is a block diagram of the process itself from Lucid's whitepaper."

Monday brain fart.....of course the output is coming from one card and not two.