NVIDIA Fermi: Architecture discussion

Yeah, they like to wast half a year just for fun. Next we might hear that since Fermi also has to compete with LRB, it won't matter if it doesn't show up before 2011.

 

Intel is awefully late on releasing a LRB related whitepaper. There's nothing like whitepaper/factsheet/powerpoint slides catfights

 

Upcoming Fermi demo on Nov 17th.

http://www.nvidia.com/object/sc09.html

 

I wonder if this will be an actual functional GT300 board or if it will be just a simulation.

chavvdarrr, they need the card back.

 

"...Fermi-based Tesla GPUs...."

I am confused, isnt Telsa different than the actual Fermi/GT300.
And at the cost comparison, 1/10=20$ of a 200$ I7, or a 100$ of a I7EX?

I really cant see any A2 cards going for less than 5000$ each to very select customers, and the first batches of A3 for 3500$ Quadros... then maybe a trickle of uber enthusiast cards at 800$ but will have a supply constrained premium of at least 25% more for 1000$ desktop card... Then into July for the 600$ everybody card?

 

poll time:

With wood screws [ ]
Without wood screws [ ]
Pay no attention to the man behind the curtain [ ]

 

http://channel.hexus.net/content/item.php?item=21142&page=4

 

The Compute line-up will logically still be called Tesla.

 

Perhaos they will rename G200b into Tesla, so that have a owrking card.

 

There are GT200(b) based Tesla products out there, Tesla is just a productfamilyname, just like GeForce and Quadro are

 

EVGA: It's coming ?

 

Lol, that EVGA hybrid GTX275/GTS250 monster is being released for black friday (since that is what the link is, a BF announcement, they're gonna cut something like $50 off of the price of the board)

 

Compatible "CUDA" Hardware with additional Graphic capability.

 

Fermi Tesla is out:


520-630 GFLOPs DP
typical consumption 190W
availability May 2010

http://www.heise.de/newsticker/meldung/SC09-Tesla-Rechenkarten-mit-Fermi-Chips-860474.html

@Nvidia: http://www.nvidia.com/object/io_1258360868914.html

 

That German site also states:

""For single-precision calculations, the cards double the performance to reach approximately 1260 GFlops at a clock speed of the 512 cores of 1230 MHz. In comparison, a traditional rival of the AMD Radeon HD 5870 achieves a processing power of 2.7 TFlops (SP) or 544 GFlops (DP)"

Also interesting, from the nvidia page:

"Up to 3GB and 6GB (respectively) on-board GDDR5 memory *"
* With ECC enabled, user available memory will be 2.625GB for a C2050 and to 5.25GB for a C2070

So why would you 'loose' 447 MB of RAM on the 3GB and 894 MB of RAM on the 6GB version when ECC is enabled?

 

Because that much amount is kept aside to store the ECC codes.

 

This one board doesn't look sawed off, but the dark backplate still blocks the SLI connector.

 

What for does a Tesla card need this anyway?

 

Also, since when do tesla cards have a display connector?

 

Should we expect higher clock-rate for the GeForce SKUs? This 1230MHz thing is disappointingly slow.