Intresting read
http://www.realworldtech.com/page.cfm?ArticleID=RWT093009110932&p=1

interesting read. nvidia has been pushing this for a while. i am not quiet sure why not more companies take that route. i really blame it most on compatibility. much of the coding and command structure would have to be redone. if they could find a way to make it more "windows" friendly and at the same time keep performance i think it be much better welcomed.

double post wtf

Yeah i agree BoT.I just hope they can do it soon.

I really dont want to go with a 58xx/59xx for my next video card.But i do want a DX11 video card.Even though i dont need it.I gues ill have to wait to see whats best.More than likely it will be ATI for gamers

well, the new ati's are monsters, no doubt but really it's just double of what was before.
which is good but not really impressive to me. the gt300 will be similar, just doubling what was already physically there but i think nv will take it a step further and improve on internal coding and processing
rumor is anyways.
i was just about to get a 285 but i will wait and see what the new gt300 brings

I think more and more we are gonna have to really study what card is for you. I think its alot easier to buy PC components if you are PURE GAMER like me. Although it costs more, but if you do alot more with your PC then maybe nvidia cards are for you. With CUDA, Physx and much better FOLDING capabilities.

On the other hand ATi have this EYEFINITY lark which alot of peeps are going mad over.......

I dont care what does what, as long as it does as its meant to do...I have to say nvidia have taken their eye of the ball a hell of alot so I am determined to give the underdog a shot this time.....they deserve it. ATi has kept its head down and worked like a motherf**ker to get this GPU out and it deserves KUDOS for that. I hope they get the drivers sorted now.

I only wish they had a CPU to compliment the GPU too.

CUDA/GPGPU = irrelevant for hardcore gaming.

Gief 75fps in all games forever, SM9.0, DX15 and make games look like RL.

Plug & Play said:

i absolutely agree and kudos is given. they did a fine job with the card and gen. before card too. i wish they had more support from vendors pushing there product and they would be much better off

sovietdoc said:

As I said Doc....If you do more with your PC....I know its nothing to do with hardcore gaming. Apparently nvidia are pulling out chipset market so no more motherboards with nvidia chipsets.

Never had one so cannot really comment on performance....

TBH it looks like nvidia is going to hit FAR wide of the mark with this one.

WHile yes, this sort of technology, and the direction they're headed is undoubdedtly advantageous for compute applications / render farms / super computing etc, they are straying FAR from their bread and butter audience - gamers & modellers.

I don't see much of an advantage any of this tech will offer graphics in gaming, or indeed in the workstation graphics area - where pixel & vertex power are still king.

They're almost going TOO parallel. I think their 'GPU' (and I use that term loosely in this instance) is becoming more and more like a bespoke CPU. And not a very good one either - as it really doiesn;t support any of the major languages in use today. Even CUDA is beginning to look like it will lag far behind the open-source arena of OpenCL.

AMD seem to have gone the right direction for their market, and really kept it aimed at gaming / graphics, while allowing for parallel computing within the architecture, rather than the other way around as it seems nvidia is going.

I think you can already see the downfall of nvidia's current strategy - compare a G80 with a GT200. Gt200 is double the size, double the cost (per GPU), massive losses in relation to yields, yet does not perform that order of magnitude better than G80 in gaming (certainly not on a price : performance scale).

While GT200 is indeed impressive in its graphical ability's - if you look at the architecture, you would imagine it would even be far more powerful.

I don't know where this is going to end, but I can see an nvidia on an island shouting 'really, this is a good idea, don't listen to Intel & AMD'.

Ferma is going to be overpriced, underperform (in relation to architecture design, cost, energy consumption etc) for what the majority of users will want it for, and will really show us how nvidia want to push parallel computing.

Nvidia are trying to develop outside of the box. The only problem is, they're outside a different box to everyone else.

I'm with dublin on this one...

I'm kinda wondering why they don't run two product lines... a GPU aimed at dominance in the PC graphics market and a PPU (parallel processing unit) aimed at the scientific computing community.

Seems like every time a product is aimed at several different applications it does all of them acceptably well but doesn't really excell at anything, while upping the cost because you are forced to pay for functionality you'll never use.

It's a bit sad, that when someone steps off the reservation and does something different that it's the ones in a DIFFERENT reservation keep saying how bad it is!
CUDA, is a good thing from a technology point, and should be respected as such.
99% of what you folks seem to assume are necessary for a good gaming system, came down from mainframes, mini's, and graphical workstations in the past. Having an application development platform/environment for the HW is just as, if not more important than the HW itself.
KUDOs to NVidia.

both nvidia and ati fail to realize the big picture and the big picture is intel.
nvidia is trying to do there own thing, some vendors go for some don't. ati has amd as it's in house product line and vendor support and doesn't mind to play nice with intel.
intel however sees amd behind ati and is not much interested to build bridges with nvidia.
intel already, pretty much pushed nvidia out of the chipset market and if it wasn't for amd then ati would be sitting on the edge on this one too.
intel plans to take gpu's down next and larabee, if it finally comes will just be a start

Gerritt said:

Gerritt, I most certainly am not criticising nvidia for attempting something different - they've been heading down this road since G80.

The criticism I have for them is half assed support for current programming models, and full support for a proprietry language / API.

Parallel computing is going to be the norm, definitely. With multicore CPU's, massively parallel GPU etc its something we as consumers won't have a choice about really.

Its the execution. While Larabee will apparently be essentially a number of simple x86 CPU's thrown together, making it highly programmable, Fermi seems to be stuck down the proprietry road.

As a developer, would you rather use an open source, highly advance compute language like OpenCL, or even use C/C++ (aka Larabee), or would you rather pay massive licenses to use a proprietry API such as CUDA to develop for only specific platforms?

I think the choice is obvious here - it cuts down on development costs, platform restrictions etc to do it through OpenCL, not CUDA.

When nvidia first announced CUDA's programming model a number of years ago, and showcased what could be done, I thought it was a huge leap forward. Unfortunately, the rest of the world has caught up with nvidia now, and there is no need for a proprietry API, utilising an essentially proprietry platform.

I think nvidia have firmly set their sights on the supercomputer and scientific modelling arena, while providing support for the actual graphics (being workstation or gaming etc), while everyone has decided not just to tack that element on, but add support with opposing priorities.

I fear for nvidia if they head down that lonely road.

Good point Dubs, but I cant help feeling a smack of DESPERATION out of the green camp....all the talk of dropping mobo chipsets to dedicated anti virus running on GPU's....I mean come on. I paid very good money for a GTX295 and it never really got a fair crack at the whip....I was just reading the latest BIT TECH review on the Asus 5870 voltage Tweak card and in CRYSIS the GTX295 "HAMMERS" the 5870...I mean HAMMERS it....can you imagine the card with good drivers. I know its a dual card against a single card but its also nearly a year old.

I am looking for the Sapphire 5870 Vapour X.....or the 5870x2/5900 Vapour X...hmmmm nice.

Look, I know they're heading to this whole "gpu computing" thing and it does have awesome applications. Being able to transcode my pr0n vids on the GPU will take a lot less time. Using GPGPU in Fermi will potentially make it a lot faster cracking my friend's passwords and hacking their Wow accounts. It will accelerate my awesome windows FX more like flip 3d and all the things I never use. This new tech is like ZE AMAZING.

On the serious note, tough: Yes I see potential for the technology.

No, I don't care about it because it won't impact what I do with the computer.

If I could have a videocard with scalable DX interface that could support 11,12 and etc without having to swap the videocard, and if the card itself would be so damn fast that it would last many many years running all games full gfx at 75+ fps, I would gladly take that instead of videocard that does gpgpu computes.

I like everyone's opinion about the topic. But even though I agree with most that the amount of features the card has are pretty unnecessary for the average customer I still think that the card will be a performance king. Nvidia won't forget the enthusiast market by delivering a lousy product, I do not think so.

I'm not really sure where you are going with that argument.
CUDA is reasonably OPEN. Yes it is HW centric, or at least vendor centric, but over all, a good operation system for the specific platform.
Granted, it doesn't have any specific application to present gaming platforms, but in the longterm it does provide for a platform that supports realtime 3d environments......just as soon as the HW catches up.

Gerritt said:


Where I'm going is that nvidia is putting all its eggs in the one basket.

Unless of course they release versions of CUDA that are not hardware specific, or they concentrate on supporting fully, more open platform APi's such as OpenCL, and fully support languages such as C++ for development, I can see this attitude backfiring.

CUDA I know, is loosely based upon C++, but when there are other API's available that do the same job, can take advantage of the same hardware, but are NOT hardware specific (OpenCL) - I really fail to see the point, especially from a developer point of view.