The GT300/Fermi Thread

[[XC] gomeler]

Just as the title says, this is the Fermi thread. You find something you want to share about Fermi, stuff it in here. We don't need a new thread every day til the fictitious launch day.

Play nice or you'll get the hammer.



Edit: Not to spoil your threat Gom, but this thread has 65 pages ATM so I thought I would take a posters advice and toss the only official knowledge on this first post:

Nvidias white paper on the arch: http://www.nvidia.com/attach/2957021...rt&primitive=0 [Jan 19, 2010]
More info: http://www.nvidia.com/object/gf100.html

If anyone has links to nVidia they feel are credible / official news feel free to send me a PM and I'll review and toss it up.

- Serra

[Ashraf]

Gonzalo-Gonads removed from Xtreme News/NVIDIA/ATI.

[Vapor]

Let's reboot this thread....

No threadcrapping this time

EDIT: a banhammer has made its first rounds...multiple people were sent on vacation and had News access indefinitely revoked....they should have been less derisive in their posting (after all, this is a stickied thread in the News section with a very explicit warning in the first post).

[BenchZowner]

Finally!

Now let's hope some people realize that this is a thread about Fermi's "REAL" news and not a thread to post silly pictures and or a battleground for AMD and nVIDIA fanboys.

If you've got nothing interesting to post, don't post.
It's that simple.

[[XC] gomeler]

Rofl, just politely told Gonzalo-Gonads to pipe down. I clearly am too soft.

[LordEC911]

~550-580mm2 die size on 40nm, about the same size as 65nm G200.
512 "CUDA cores"
~1.5TFlops of SP and ~768GFlops of DP, depending on final clocks.
48ROPs

10.5" board for GF100.
Core clocks <750mhz and shader <1700mhz, expect similar clocks to GTX280/GTX285. <5ghz GDDR5 on a 384bit bus, giving <240GBps.
TDP of ~225-250w.

A lower binned, cutdown and down clocked variant with a TDP of ~170w is expected if they are planning a dual GPU board before an optical shrink.

Early performance estimates put it at a 20-40% advantage over HD5870.

[SamHughe]

~550-580mm2 die size on 40nm, about the same size as 65nm G200.
512 "CUDA cores"
~1.5TFlops of SP and ~768GFlops of DP, depending on final clocks.
48ROPs

10.5" board for GF100.
Core clocks <750mhz and shader <1700mhz, expect similar clocks to GTX280/GTX285. <5ghz GDDR5 on a 384bit bus, giving <240GBps.
TDP of ~225-250w.

A lower binned, cutdown and down clocked variant with a TDP of ~170w is expected if they are planning a dual GPU board before an optical shrink.

Early performance estimates put it at a 20-40% advantage over HD5870.

Source?

[Sadasius]

How ironic this is called the 'Fermi' thread. Basically it entails that it will have a meltdown or it will produce a sizable amount of energy. But seriously...any sources for anything concrete?

[XSAlliN]

What is Fermi (never heard of it)? Searched google and end up with this:

http://ro.wikipedia.org/wiki/Enrico_Fermi

http://www.fnal.gov/

...which is really strange, cause I always thought that smart people, as in intellectual individuals are a minority.... That's why this quote is confusing:

We don't need a new thread every day til the fictitious launch day.

[Mad1723]

What is Fermi (never heard of it)? Searched google and end up with this:

http://ro.wikipedia.org/wiki/Enrico_Fermi

http://www.fnal.gov/

...which is really strange, cause I always thought that smart people, as in intellectual individuals are a minority.... That's why this quote is confusing:

It's the codename for the new generation of Nvidia GPU, more often called GT300 series

[XSAlliN]

Oooooooohhhhhhh, now that makes sense - the annual nVidia/ATi GPU parade. I won't even bother with this one, by 2011 a new model will be 20x faster compared to GT300 (less or more)... Anyway a Mod/Admin should add GT300 in title, so even the ones that are sick and bored of ATi/nViudia annual pared, know what is it about...

[zalbard]

Anyway a Mod/Admin should add GT300 in title, so even the ones that are sick and bored of ATi/nViudia annual pared, know what is it about...

Good point!

[LinkinParkBoy]

GTX 360 = 320bus x 4 = 1280MB
GTX 380 = 384bus x 4 = 1536Mb

?

[Serpentarius]

What is Fermi (never heard of it)? Searched google and end up with this:

http://ro.wikipedia.org/wiki/Enrico_Fermi

http://www.fnal.gov/

...which is really strange, cause I always thought that smart people, as in intellectual individuals are a minority.... That's why this quote is confusing:

here's another one ... i came across it while reading the Particle Accelerator
http://en.wikipedia.org/wiki/Fermilab

[Sly Fox]

I just hope Fermi is everything Nvidia and pals have hyped it up to be.

It's hard not to be really skeptical of Nvidia lately, but I know I for one am waiting for hard numbers before buying anything.

With any luck it'll be 8800 GTX version 2.0 and be a must-buy.

Otherwise I'm gonna be pretty sad I wasn't playing games on a 5870 this whole time.

[Chumbucket843]

if you guys were wondering why its named fermi it is because it was announced on enrico fermi's b-day.

i would expect it to be a very fast chip if they can tackle some bottlenecks like triangle setup and get it to clock at 1400+ MHz.

[ajaidev]

GTX 360 = 320bus x 4 = 1280MB
GTX 380 = 384bus x 4 = 1536Mb

?

I am a bit confused the white paper said "i think" fermi had 6 memory controllers so that means 256MB*6 memory controllers = 1.5gb in total 6 memory chips.

The bus should only show the bandwidth and not the capacity !!

GTX 360 "if made with a 320bus" can have 5 memory chips and that means 5* 256MB = 1.2gb in total of 5 memory chips one memory controller will be unused.

[trinibwoy]

^ Correct. It'll probably be 10 1024-Mbit chips though for a 320-bit bus. I don't think the 2048-Mbit GDDR5 chips are out yet.

[madcho]

No proof that any EE sample are out for now.

And nobody should post informations without any source.

[jaredpace]

same as an 8800gtx, just lay-out is different

[firas]

i'm i missing something?
what happened to this sticky in Nvidia section?
http://www.xtremesystems.org/forums/...d.php?t=235658

[Piotrsama]

i'm i missing something?
what happened to this sticky in Nvidia section?
http://www.xtremesystems.org/forums/...d.php?t=235658

I think it once was on the news section, but got moved there...
Why? I don't know.

Maybe they should bring it back here and delete this thread.

[ajaidev]

Reg now for the tesla showcase

https://www2.gotomeeting.com/register/606754650

I have started this thread instead of posting it here because its relatively short notice and the ex-poser needed could only be given in a separate thread IMO.

[LordEC911]

No proof that any EE sample are out for now.

And nobody should post informations without any source.

Ummm... early samples have been out for at least a month or so, A2.
They will have A3 back at the end of the month/beginning of Jan.

[WeeMaan]

I think it once was on the news section, but got moved there...
Why? I don't know.

Maybe they should bring it back here and delete this thread.

Yeah that's right, I first posted it in the news forum and then it was not official.
Later a mod renamed it to the official Fermi thread, and after a week or so it got moved to the nvidia forum for some reason.