Nvidia unveils Fermi-based Tesla GPUs

Nvidia has unveiled new Fermi-based Tesla products, the Tesla 20 series GPUs, which are aimed at the high-performance computing market. Designed for parallel computing, the company said its Tesla 20 series graphics chips reduce the cost of computing by delivering the same performance of a traditional CPU-based cluster at one-tenth the expense and one-twentieth the power consumption.

[Nihilism]

The day is not over, we may yet hear info about the consumer models. The card still looks sexy though, it has obviously gone on a diet since the gt200 cards.

[toaster]

Any of these cards are way overkill for most games out on the market right now. Plus most people aren't going to drop 12grand on a card. You can get two in SLI and it will still run any game, especially if you go the 295GTX route. One 295 is only around $500-$550 and you can get two and still have amazing performance.

I don't think many programs will be fully compatible with these Tesla cards right off the bat anyways so better to wait until its consumer friendly.

[Nihilism]

the tesla cards are for science research etc.....they are not available to buy to average consumers. The same as workstation gpu's costing thousands...they are not for gaming.

[toaster]

Look at my comment 6.1

[Jellzy]

Check 6.2

These cards don't have any form of display output, so in no way could they be used for Gaming.

:)

[MNicholas]

DP performance is decent and price competitive with the PS3's Cell processor. However, the revised Cell processor available from IBM leaves Fermi in the dust in every possible way.

More performance/$ (32 and 64 bit)
Broader instruction set
More programming paradigms (serial, parallel)
Fully programmable data/task management
Lower operating cost
etc...

[Nihilism]

no it doesn't, if you are going off the amount of GFLOPS, that doesn't translate into real world performance, this card is a beast, no cell processor can crunch like one of these

[MNicholas]

because they are limited by what can fit in their cache (resulting in about 40% utilization while processing large data sets) but with Cell, due to it's unique memory architecture, you can actually achieve very close to theoretical floating point performance.

The real magic of the Cell processor isn't just the high density of processing muscle but the effectiveness with which it transports data in and out of the execution units.

[Nihilism]

"As for Fermi vs Cell, transistor for transistor, watt for watt, "

do you hear yourself, your somehow trying to say that the number of transistors and power usage = performance.

Why don't you tell me clock speed = performance.

or that a 3.2ghz quad AMD = a 3.2ghz i7

you have no idea what your talking about, the hardware stats themselves mean nothing....the benchmarks, mean everything. How about you take your backwards 10 year old logic and preach to the ps3 fanboys who are still convinced that a cell processor is better than current desktop quad's at non gpu accelerated processing.

You also realise that the high end server based cell's go for over $12000???, because you could buy a cluster of these cards for the same price.

[MNicholas]

There is a world outside of x86.

You wrote:

<<you have no idea what your talking about, the hardware stats themselves mean nothing....the benchmarks, mean everything.>>

You couldn't be more wrong. Benchmarks are meaningless when you're talking about custom developed applications.

For custom applications like software rendering and super-computing (where Cell and Fermi will be utilized) peak performance and data management are everything. Cell and just about any modern GPU will humiliate any x86 chip in both those areas.

<<How about you take your backwards 10 year old logic and preach to the ps3 fanboys who are still convinced that a cell processor is better than current desktop quad's at non gpu accelerated processing.>>

What the hell are you babbling about? The Cell is vastly better than any desktop CPU at a variety of non-GPU accelerated processing.

There's a reason the Cell is being used in hundreds of super-computing applications around the world including in the massively powerful super-computer called "Roadrunner."

Having said that, the i7 will almost certainly outperform a Cell at running Microsoft Word.

<<You also realise that the high end server based cell's go for over $12000???, because you could buy a cluster of these cards for the same price.>>

They cost that much because that's how much clients (like universities, businesses and govt) are willing to pay. One day when you grow up you'll understand that in niche market segments, such as super-computing, manufacturing cost has no impact on retail pricing.

The Cell is a tiny little chip and costs very little to manufacture. If the market conditions warranted it, IBM can drop it's price by an order of magnitude and still take home a healthy profit.

In contrast, Fermi is a gigantic chip with 3 billion transistors. It's an order of magnitude or few more costly to manufacture than the itsy-bitsy Cell.

[O2_Addict]

Your obviously outgunned dude.

mnichols is famous on N4G for his tech knowledge. I also know for a fact that people have even copied his comments from N4G on other websites. Fanboys on both sides hate him for telling the truth about PS3 and 360.

If i have one complaint about him is that he writes too much and also gets too technical. He should try to make shorter comments. Hear that mnichols??? Short and sweet!!!

[Pandamobile]

GT 300 pl0x.

[Nihilism]

i bet you'll beat me to it when they release the info you bastard! /jk

but seriously they are supposed to release the cards soon, they are doing a press release with working demo's....if ever there was a time to swing some benchmarks my way...now is it

[primesuspect]

We'd LOVE to get a sample on the bench. We're thinking this thing is gonna absolutely tear through F@H work units...

[Ju]

I'm just reading through OpenCL (and CUDA) and some of those specs really make this a "custom computer on a board" than anything else. I'd like to see how this baby crunches through OpenCL code. System calls ? Recursive functions ? (C++ support - well, that was there before on a compiler level, no ?). So, now you really can port C code to the GPU, huh? Awesome.

[StarScream4Ever]

Good thing the price is for computer developers other wise I would have a heart attack right now. X_X

[DevastationEve]

I wanna say this now...back in july I tried to run F@H on my laptop 8600m gs.

And that was the last I ever heard of it. The gpu failed and now that laptop is useless.

And I hate nVidia ever since because of their faulty laptop gpus. I appreciate the work they do, but it doesn't get me a new laptop.

[primesuspect]

You gotta do the cookin' by the book to fix that GPU! (and your laptop)

http://icrontic.com/article...

[primesuspect]

This is a supercomputer-on-a-card, designed for heavy math, Toaster.. It's definitely not a "gaming" card.

[toaster]

I said it is overkill for any game. A PC just for gaming doesn't even need this card. And then I gave an example of what would be good for gaming. It's a no brainer that these cards aren't targeted at gamers.

The companies who are going to buy these cards have big wallets and they'll use it for themselves for whatever reason, I don't know.

[Jellzy]

"The lack of ability to output images to a display is the main difference between Tesla products and ordinary video cards"

http://en.wikipedia.org/wik...

Basically this renders them useless for gaming...

[DeadlyAnGeL91792]

and this is how i know that humanity is always advancing i mean jesus christ that thing is a true beast and uses LESS power wtf where was this 5 years ago? lol

[Nihilism]

it's official, no gaming fermi till next year:

http://www.nvidia.com/objec...

"Editors’ note: As previously announced, the first Fermi-based consumer (GeForce®) products are expected to be available first quarter 2010."