NVIDIA GT300 ''Fermi'' Detailed

[btarunr]

NVIDIA's upcoming flagship graphics processor is going by a lot of codenames. While some call it the GF100, others GT300 (based on the present nomenclature), what is certain that the NVIDIA has given the architecture an internal name of "Fermi", after the Italian physicist Enrico Fermi, the inventor of the nuclear reactor. It doesn't come as a surprise, that the codename of the board itself is going to be called "reactor", according to some sources.

Based on information gathered so far about GT300/Fermi, here's what's packed into it:

• Transistor count of over 3 billion
• Built on the 40 nm TSMC process
• 512 shader processors (which NVIDIA may refer to as "CUDA cores")
• 32 cores per core cluster
• 384-bit GDDR5 memory interface
• 1 MB L1 cache memory, 768 KB L2 unified cache memory
• Up to 6 GB of total memory, 1.5 GB can be expected for the consumer graphics variant
• Half Speed IEEE 754 Double Precision floating point
• Native support for execution of C (CUDA), C++, Fortran, support for DirectCompute 11, DirectX 11, OpenGL 3.1, and OpenCL

Update: Here's an image added from the ongoing public webcast of the GPU Technology Conference, of a graphics card based on the Fermi architecture.



View at TechPowerUp Main Site

 

[laszlo]

i see a 5870 killer

now all depend on pricing

 

[DaedalusHelios]

i see a 5870 killer

now all depend on pricing

Pricing will be a huge factor. Lets hope they wise up to a lower MSRP than the current trend.

 

[pr0n Inspector]

Awaits for nerdgasm and nerdrage.

 

[human_error]

a billion more transistors than 5870 - this thing is going to be huge! (size wise, although i'd wager performance will be very nice too).

Now we just need a decent price war between ati and nvidia on the dx11 cards and i'll be happy.

 

[Howard]

ohhhh, crown~
ready to wear it!!!

 

[$ReaPeR$]

this looks extremely promising

 

[FatForester]

This looks really interesting, hopefully the yields are good enough so they won't charge an arm and a leg for it. I just have to wonder if "reactor" relates to anything to do with its thermal output? 3 billion transistors on a 40nm process has to get pretty warm.

 

[happita]

[*]32 cores per core cluster
[*]Up to 6 GB of total memory, 1.5 GB can be expected for the consumer graphics variant

What the?!

The price might be decently competitive, I don't see a 512-bit memory interface like the previous cards they had before jacked up on price.

 

[[I.R.A]_FBi]

rassclaat

 

[FatForester]

What the?!

The price might be decently competitive, I don't see a 512-bit memory interface like the previous cards they had before jacked up on price.

I'd say the 1.5 GB variant is for regular end-users like us, while the 6 GB is for workstations that could use the extra memory for CAD or other really heavy tasks.

 

[buggalugs]

i see a 5870 killer

now all depend on pricing

Only for idiots who think 10 extra fps is worth $300 more.


The 4870/4890 have done very well while there are faster Nvidia cards.

 

[qubit]

Shame that bus isn't a full 512 bits wide. They've increased the bandwidth with GDDR5, yet taken some back with a narrower bus. Also, 384 bits has the consequence that the amount of RAM is that odd size like on the 8800 GTX, when it would really be best at a power of two.

 

[DaedalusHelios]

Only for idiots who think 10 extra fps is worth $300 more.


The 4870/4890 have done very well while there are faster Nvidia cards.

"Only idiots" is reserved for people insulting a possibly better product before seeing the official pricing.

I am sorry, but thats just what came to mind.

Don't get pissed with a paper launch. The GPU isn't on shelves or priced yet so its still useless to everybody.

Shame that bus isn't a full 512 bits wide. They've increased the bandwidth with GDDR5, yet taken some back with a narrower bus. Also, 384 bitys has the consequence that the amount of RAM is that odd size like on the 8800 GTX, when it would really be best at a power of two.

Well since its a different GPU architecture we do not know how well it would scale to higher bandwidth memory yet. No benches mean we are still in the dark.

 

[gumpty]

Only for idiots who think 10 extra fps is worth $300 more.


The 4870/4890 have done very well while there are faster Nvidia cards.

QFT.

Not going to notice much difference going from 100FPS to 200FPS either. The pricing is the key. If it beats the 5870 and they can get the price close enough to ATI's offerings, then we all win with price warz. If it beats the 5870 but is too expensive ... it will mean nothing.

 

[Atom_Anti]

QFT.

Not going to notice much difference going from 100FPS to 200FPS either. The pricing is the key. If it beats the 5870 and they can get the price close enough to ATI's offerings, then we all win with price warz. If it beats the 5870 but is too expensive ... it will mean nothing.

I think you are right!

 

[csendesmark]

i see a 5870 killer

now all depend on pricing

Bigger transistor count mean lower yield + less chip/wafer = more expensive card
AMD can drop prices

 

[laszlo]

Only for idiots who think 10 extra fps is worth $300 more.


The 4870/4890 have done very well while there are faster Nvidia cards.

as i estimate it'll be more than 10 fps,i say between 50-100 over 5870

so the "idiots" who want a future proof card(better than 5870) will buy it if the price won't be much higher than 5870;i expect price around 500 or less;we're not anymore in the dark ages when nvidia&ati by common agreement has overpriced the high-end cards just to charge as they want

 

[btarunr]

Bigger transistor count mean lower yield + less chip/wafer = more expensive card
AMD can drop prices

Uh no, the equation will be similar to that between RV790/RV770 and GT200. Despite everything, GTX 200 series cards have been affordable.

 

[Benetanegia]

Half Speed IEEE 754 Double Precision floating point is just sick!! (I had to say it here too.)

Bigger transistor count mean lower yield + less chip/wafer = more expensive card
AMD can drop prices

Yields don't matter in this case because Nvidia is paying for chip and not per wafer and yields are good, it's been cofirmed by Nvidia. All the news about bad yields were false and spread by AMD's competitive analysis team, whatever that is.

Anyway although that formula is true in the technical aspect, it doesn't take a very important thing into account, we don't know how much evey company pays per waffer. Nvidia makes twice as many chips (they've been selling twice as much) so as it happens with every other inter-company volume deal in the world:

more products = less $ per product
more wafers = less $ per waffer

 

[ZoneDymo]

Yeah nice, dont care though, with there way of doing business, im going ATI.

 

[laszlo]

Yeah nice, dont care though, with there way of doing business, im going ATI.

you should care because this will force ati to drop prices and i bet you like lower prices no?

 

[DaedalusHelios]

Yeah nice, dont care though, with there way of doing business, im going ATI.

I have heard their PCB wafers are made of ground up babies. Only a sicko would buy Nvidia cards.

I buy both so I guess the jury is out on me.

 

[naram-sin]

"Only idiots" is reserved for people insulting a possibly better product before seeing the official pricing.

I am sorry, but thats just what came to mind.

Don't get pissed with a paper launch. The GPU isn't on shelves or priced yet so its still useless to everybody.

Well since its a different GPU architecture we do not know how well it would scale to higher bandwidth memory yet. No benches mean we are still in the dark.

Sorry for the long quote, regarding paper-launch, I think you're right. It seems to me that, considering our paper-launch experience in a last couple of years, that this one is extremely narrow, stll in kind of alpha stage and I think it is yet to be largely revised at least 2-3 times, before we can do some actual scaling with ATI solution and get first realistically expected performance figures. Because all of stated is kind of sci-fi to me. As I said, for now.

Btw, it could all be a piece of... hype to try and halt a couple of percentages of extreme enthusiasts in buying of 58xx and motivate them in waiting for GF100/GT300/Fermi/Reactor or whatever... (btw, doesn't Reactor makes U automatically think of high temperatures?)

 

[gumpty]

Yeah nice, dont care though, with there way of doing business, im going ATI.

Seriously man, whether you like it or not, you making a purchase of anything is part of how you run your business. Excluding half your potential business partners because of rumours about bad behaviour is not the smartest move. Cut off your nose to spite your face?