Carl Jones of Crytek, Ashu Rege of NVIDIA, Harish Sivaramakrishnan of Adobe, Keita Iida of NVIDIA, Philippe Vachey of DSK Supinfocom, Robin Alter of Kreeda Games, Tridib Chowdhury of Adobe, Rev Lebaredian of NVIDIA, Sumit Gupta of BitRhymes, Hemanth Sharma of Adobe, Simon Green of NVIDIA, Varun Nair of Blue Frog, Krishna Pediredla of Drona Labs, Jithin Rao of Ubisoft and Imran Khan of FXLabs is coming this february to India's first and independent annual summit for the game development ecosystem - India Game Developer Summit (gamedevelopersummit dot com) to speak on various topics. For more info please log on to gamedevelopersummit dot com

but at least it will be better than Larabee. If you want to get a good laugh, think about how Intel's effort into making GPUs turns out. The sad part is what they do to the poor people who buy laptops with their graphics inside because they don't know any better.

ATI rules right now, and probably will over Fermi, but I'd still bet on NVidia having a decent refresh even if Fermi doesn't set the world on fire.

ATI is golden right now with the 5000 series.

Ummm semi accurate...Charlie Demerjian is an absolute nvidia hater lol. this is coming from a guy who runs ATi, I will never bother with this guys articles... ever.

Ive been wondering what happened to Charlie, looks like he was let go from this site huh, and yes the site has been sucking ever since, time to delete you from my bookmarks!!!!

This site sucks without Charlie. I don't read his site very much but I read it as much as I read this one now.

And to Carrie. A GTX 280 won't overclock into that TPD. Your numbers are wrong. TPD is not the problem the problem is normal usage and as I understand it getting the ion doping process perfect enough to handle things while simultaneously getting the interconnect layers to actually connect without a billion screw ups. This is what happens when people won't make compromises. They push too far and too hard and blow up.

Um, Nvidia doesn't make the chips, TSMC did and they should have used better stuff. This is akin to firing a coach when the quarterback throws 5 interceptions losing the superbowl. Did the coach throw those balls? On top of that manufacturers (HP who took partial blame at least, paying millions, Dell etc) didn't listen to thermal suggestions from Nvidia. I'm not a fanboi of either (I have a 5850 on backorder at amazon for 2.5 months now...jeez...I won't let go of that $260 price with free shipping despite amazon's efforts...LOL) but if AMD takes any longer to make their chips mainstream instead of making all available chips 5870's I'm going to be seriously looking at Fermi I guess. No I'm not blaming AMD, just wondering why they're aren't fabbing from Global Foundries yet (as TSMC seems to be making everybody's stuff late, or scrapping the hell out of a LOT of chips lately). I don't have my card because of TSMC, not because of AMD. AMD doesn't make SQUAT. TSMC is holding up my card. When your chip sucks it's usually because of who PRODUCED it not the designer. Granted you can have a bad design too, but NVIDIA-GATE was really TSMC-GATE and OEM-GATE (should have used better cooling designs). Blame the correct people next time.

On a last note I do believe the sources that say FERMI will rock. They're certainly more reliable than CD and don't come with the bias. CD is rumors, the others are REAL DATA.

From what I'm reading, nVidia might as well skip the original Fermi for all its problems and call this new version Fermi 2. See:

http://en.wikipedia.org/wiki/Enrico_Fermi_Nuclear_Generating_Station

I'm not. Very exciting article. Nvidia's arrogance will seemingly finally backfire.

Having to sign up to a forum is a pita when all I want to do is join in on the comments.

The inquirer acolite said:

"But really, a single GPU chip that draws 225W is simply unheard-of. Let's just put it this way.... blablabauuu (I don´t have any idea about it)... blablablauu"

WTF?

"It is simply unheard-of"?

Are you a journalist?

Facts:

GTX 280 had 236W of Max TDP.

GTX 275 has 219W of Max TDP.

HD4890 has 190W of Max TDP.

HD2900XT had 215W of Max TDP.

So, THIS max TDP isn´t a new and top record, your arguments are treacherous, the specs that you take are for Tesla products ONLY (your pdf is about this, NOT about ALL Fermi products), and finally, you are taking the garbage from one source that your site fired from here some months ago... xD

You are ridiculous...

Do you know that Tesla products are more constraint about TDP that gaming hard?

Something about computing racks, little space, many cards in the same case... etc.

The Tesla products specs are different to the gaming products specs.

Do you catch this?

Congratulations:

The Inquirer was going to be the garbage pump that it was some time ago.

Noone here, including CM, really knows what Fermi is going to bring to the table and how it will perform. It may suck or it may dominate.

So hold your words lest you sound like a fool.

I say it's the yield which causes the drop in cores, the heat would lower GHz.
Remember: a new process, a new design...

history has shown Charlie to be right far more often than he is wrong. And when he is wrong someone usually feeds him bad info.

Considering the early less than 10% yields for A1 this development should come as no surprise to anyone. Nor the reduction in clocks.

If you miss Charlie head on over to Semiaccurate, The forums are pretty lively.

'' The Inq got very boring once charlie left, all the articles now are just boring or either linked to other articles, like this one.''

I TOTALY AGREE WITH YOU !!!

Nothing is proven yet. and going from charlies work over the last 6 months, saying this card wont be out before xmas, a good chance he is right.

and LOL and the Inq taking a past employee "Charlie" and taking his work to show an article and then adding, ridiculous statements like it may need phase change cooling.

@Mike

The Inq got very boring once charlie left, all the articles now are just boring or either linked to other articles, like this one.

Maybe Nvidia should have went to IBM or GloFo to produce chip then they wouldn't be having such a hard time as they are now. There's probably a reason why ATI went to GloFo since TSMC couldn't produce.

OK, kids, time to do some research!

Tesla is a suite of GPGPU products built around on Nvidia's G80 and later GPUs. Fermi is an upcoming new GPU from Nvidia.

So, children, do you thing that Nvidia will sell a Tesla product containing a Fermi processor?

And, for bonus marks, if A contains a B, does that make A a B?

225W? No way, how about I stick to this 9600GT instead. At 225W for the video card alone I will be sweating bullets while gaming.

semi-accurate actually inaccurate

http://www.legitreviews.com/news/6999/

So that's one for one against. Oh wait that is Charlie D Spreading info against Nvidia .. enuff said

Oh please, I hate nVidia as much as the next guy (got ripped off on broken NV40 circa 2003), but to make the claim that Fermi will require phase change cooling is laughable. If you are going to copy pasta Charlie's hardwork and reporting, why would you add such nonsense?

NVidia may be a disaster, but they aren't stupid enough to release a graphics solution that requires phase change cooling. Hell, they aren't even stupid enough to release something that requires water cooling.

Due to condensation/insulation concerns, it would never ever ever ever happen. Ever. Never.

Realistically, all that nVidia can do is cripple the design further and further until it will work ok with a beefy air cooler, something that they appear to be doing as we speak.

It is possible that nVidia will make some speciality boards with non-crippled silicon for speciality purposes, that could use extreme cooling, but not for a mainstream consumer part.

But when you consider that Oak Ridge has told nVidia to piss off, even extreme examples with full 512 SPs seems unlikely now. The power requirements for cooling alone would be outrageous let alone the actual consumption of the boards.

I'm bored by this article

Did you notice the question mark there? The one asking, is this PDF irrelevant?
I'd not class myself as a noob but will defer to your slander. With a name like The Watcher, you obviously have some ego issues.
There's no need to call folks noobs when they have missed a point, it in itself makes you appear like the typical cyber warrior, full of casual insults and bravery when in fact on line, anyone can be 'big and tough'.
So, back to the point, thanks for pointing out Tesla C20 series is in fact 'Fermi'. I've followed it closely and hadn't realised they were sticking with the Tesla 'monicker' which seems absurd when it's a whole new architecture*. I've read so much about Tesla HPC's recently I'd assumed these were parts already out.
*considering how quickly they renamed desktop parts with a minor tech revision, it's contradictory to keep a new techno wonder with the same name as its predecessor (oh but of course, it's for scientists who wouldn't understand that the $xxxx part they're buying isn't the newest HPC CUDA board from NV.)
Pah!!

With a chip as big and complicated as Fermi it is surely normal practice to design it so that it is made of several independant sections so that a small flaw in one of them does not make the whole chip useless and improves the effective yield. Perfect chips will have 512 cores, ones with a small localised flaw, can have part of the chip disabled to give say 448 cores or less however I see they still clock at 1.25-1.4 Ghz which gives 600-700 Gflops double precision floating point which is a world beating spec with 225 W of power (3.1 Gflops per Watt). Maybe the yield of perfect chips is low, so they are saving the perfect ones for "special customers" and they are setting expectations for a more general release with fewer cores. Nothing to worry about - sounds just like good engineering.

Love To See Nvidia Survive, Lived down block, in basement company. AMD tracked down thru telco to eastern europe & those relatives where toothless nuts.

Talking About AMD & time/Charges would last 30 Secs. Talk About SlaughterHouse 5 & sacred dressden Whimpers last all Night. Nividas' Relative Newness Has taken Lacking Host to Core.

When Vista Ultimate Finally went Plausible by June '7 with 690 Virtulization, Finding suitable Gamer Wasn't Easy. FORGET Nvidia Entirely. Finally ATI Fire Series Broke thru, Simpler, More Likely it'd work.

ALL Game Designers Started ALL over Falling Back to 200, then 150, finally at 1XX level, Workable Cards Started to appear. You could still get xp, so what was huff, only Ultee' & those brave Enough To Storm Ramparts of HELL, Did Improvements testee.

Forget 160, 2900 3200 & on & on And all inbetween. Finally in Pci-e 2.0 Vista started to pop Reliably on All Models. Nvidia Never Gotten that Much Figured Out. ITS' TOO COMPLEX FOR HOME STARTUP NOW. Charlies' loss of Grace Was Called significant Economic downturn. Starting with NBC selling Off Up & Coming theINQ, theINQ haveing to let go & let Go SomeMore.

Its Unlikely without Miracle cure orBefriending by Knowledgeable sources, Nvidia Can Make It. Yet DIGITAL Design IS PUBLIC Property after two Years & Nvidia May Be Waiting Out its' Time.Cramming XP Tech into Entirely Revamped system.

drashek

I could heat my house with a few of those.

more woodscrews!!!

Serves them right for releasing hundreds of thousands of faulty GPUs to be soldered into people's $1000+ laptops. Nvidia = Fail.

Charlie actually gave good arguments as to why he speculated what he did. And guess what, he turned out to be right. I don't know why he lost his old job...I guess someone doesn't like controversial. Or maybe the green goblin payed to have him taken out.

charlie might know his stuff (might) but its hard to take someone seriously when ALL they do is bash nvidia. im an ati boy but i still dont think continuous bashing of one company is very smart. i never once remember him saying anything bad about ati. im very happy he left he so i dont have to read his overly bias crap

Those limits mean that the card will draw less power than my two Radeon 4890's. The O'Ced HIS has a 8+6 and both have to be plugged in to run. The standard MSI has a pair of 6 pin connectors. The pdf says 1-8 pin or 2-6 pin. That is less power for the Fermi than the 4890. About on par with the 5870. There may well be thermal transfer problems but they shouldn't be more than what have existed on cards previously available. I value Dave Patterson's opinion more than any of the usual sources since it was his people who blew the whistle on Intel screwing around with the compilers in benchmarks as a result of the Franklin procurement process.

"That PDF links to Tesla C2xxx boards. I thought Fermi was NOT Tesla? Tesla boards are out there currently, Fermi isn't - so is this PDF irrelevant?"

Alternatively - read about the cards;

"Based on the new NVIDIA CUDA™ GPU architecture codenamed “Fermi”, the Tesla™ C2050 / C2070 computing processor is designed from the ground up for high performance computing."

Weak

Inq could have just cut out the middleman and kept Charlie on as the man sure knows his chips.

Not a fanboi, but I do enjoy pissing em off and lately most my geek arguments center on 5xxx vs GT300. Most refuse to believe that Fermi is just so very wrong regardless of how powerful it will be.

this whole thing about Fermi remember me the R600 launch !

dont remember ?

ATi pushed back the thing about 4 times, while saying that it will beat G80 ! The final product was a BIG , HEATING, POWER HUNGRY chip, that didnt perform well....

Looks like nVidia is facing the exact same situation ....

Fermi : BIG, POWER HUNGRY chip , that will not perform as expected !!!

That PDF links to Tesla C2xxx boards. I thought Fermi was NOT Tesla? Tesla boards are out there currently, Fermi isn't - so is this PDF irrelevant?

It's from CD? Pinch of salt required. There hasn't been much recently about Fermi (past 3-4 weeks) so maybe it's not quite going right. I'm an ATI owner now but i'm still thinking Fermi is lurking and it's big ugly and powerful.
But to be fair, it'll surely need to cost a bomb to make it a plausible product (recoup of R&D plus manufacturing costs)??