The GT300/Fermi Thread - Part 2!

[Otis11]

I condone the re-branding of products method, in business signed: Decami 2/2/2010

You and everyone else can complain all day about how much it "annoys" you, but the fact of the matter is, its an insanely profitable market strategy and used by more companies than you can possibly know, not just Nvidia, this is nothing new, at all.

whats it do? make money on the weakest links of the consumer, we fortunately are not those links, be happy that you aren't and move on. Nvidia rebrands, oh god, lets freak out every time they do, and every time every other company does. Not gonna change a thing. I figured people would be over this by now.

back to fermi talk....

OT note - Car industry anyone? Extreme example - difference between a '68 and '69 Camaro. Anyone?

Back on topic. So by the time this card comes out we'll have been waiting how long again?

This better be freakin good...

[Solus Corvus]

Pictures of the production Tesla card with 448 SPs show a single 8-pin which bodes well for power consumption IMO.

There is also a 6-pin on top.

[ElSel10]

Really? With less than half the texturing power of a 5870?

I read somewhere that the GF100 TMUs are 4-way while RV870 is 2-way or something like that. I have no idea what that means though.

[ajaidev]

Pictures of the production Tesla card with 448 SPs show a single 8-pin which bodes well for power consumption IMO.

Seems like beavers dont have a good memory lol



can you see the 8 + 6 pin

[tajoh111]

I think Intel is a good example of what happen when you don't do what nvidia is doing by jumping to the 4xx instead of the 3xx with fermi. Intel has probably the most confusing naming system in history. The core i7, i5,i3 and pentium processors need an extreme overhaul. There are too many processors that are named in the i7 range that shouldn't be there. Similarly the i7 980x is a hella confusing name, it should have been the i9 975 or something to show it has six cores.

I missed the days when processors frequency denoted speed.

I see way too many pc builds now a days that throw in a computer with a 1366 processor in a 1156 socket build on forums or add a triple channel memory kit with a dual channel memory system. These are real suggestion too, not imaginary anecdotal situations where people upgrade from 8800gt to 9800gt to 240gts, that could happen but don't.

[NapalmV5]

People seem talking out of their rear these days.

totally agree great example right here

That's nice, a "would've" here, could've there, should've there..

Fact is, this did not come out - period.

As for the rest of your info, seeing as how you've claimed in thread's past that you don't really know what's going on, but it's based on your conjecture, I won't be holding my breath. Whatever happened to the rules about people claiming things without being willing to back it up? It's gotten people in trouble before...

yeh if i dont l*** ati arse.. im a total retard

go on have more kool aid.. take it easy

[NapalmV5]

Really? With less than half the texturing power of a 5870?
65% of the Polymorph engine of GTX480?
I'm sorry... if GTX470 beats a stock 5870 by more than 20% I'll be impressed.


We will see... I honestly doubt it.

neither the gods nor nvidia shall beat the texturing power of a 5870!!

seriously just whats so bad about nvidia beating ati offerings ?? dont you guys want 6000 from ati ??

[trinibwoy]

I read somewhere that the GF100 TMUs are 4-way while RV870 is 2-way or something like that. I have no idea what that means though.

Don't know where you read that but no. They both have quad-TMUs and GF100's theoretical peak is lower than Cypress and waay lower than Hemlock. If it has any chance of competing in texture limited scenarios it would need much higher efficiency.

[Biker]

Nnnnnnnnnoooooooooooooooo....!

Off topic train wreck in 3....2....1....



Am I taking the Fermi?

We'll see soon after release

[kgk]

Pictures of the production Tesla card with 448 SPs show a single 8-pin which bodes well for power consumption IMO.

What?

There's a 6 and 8 pin.

[570091D]

OT note - Car industry anyone? Extreme example - difference between a '68 and '69 Camaro. Anyone?

the only shared piece of sheet metal between the '68 and '69 camaro is the roof.

[BenchZowner]

What?

There's a 6 and 8 pin.

I think he means just 1 8pin & 1 6 pin instead of dual 8 pins as the first rumors regarding the card's power req.

[saaya]

Nnnnnnnnnoooooooooooooooo....!

Off topic train wreck in 3....2....1....



Am I taking the Fermi?

We'll see soon after release

nice pic hahah

The HiNT chip is blown, but I've got a spare HiNT chip and an appointment with a BGA IR rework station.

The board powers up, Bios is accesible, so most likely it's the HiNT chip, so it should be working after the bridge chip is replaced. Fun stuff, looking forward to it.

ps, you're right about the bridge chip being sensitive, motherboard incompatibility can be a problem, the chip will simply overheat due to it, and blow.

Better be on topic again, otherwise we will get into trouble.

ahhhh so if its a certain agp standard itll use higher signaling voltages and then that fries the bridge? makes sense... good to hear you figured it out before running into the problem!

btw, any news about GF104?
geforce 5xx?
seriously, how many sps do you think itll have? it should be as fast as a 285... and it will probably 128bit memory... so it will have about the same bw... 285 has 240sps... so maybe 256?

[N19h7m4r3]

That looks like a really odd place to have the power connectors.

One on top and one at the back.

Just odd.

[saaya]

That looks like a really odd place to have the power connectors.

One on top and one at the back.

Just odd.

the current tesla cards have the same layout already...
has to do with how the cases they plug in are designed i think...

[N19h7m4r3]

the current tesla cards have the same layout already...
has to do with how the cases they plug in are designed i think...

Ah alright.

I've only ever dealt with Quadro, and not Tesla.

So it seems really weird.

[saaya]

Ah alright.

I've only ever dealt with Quadro, and not Tesla.

So it seems really weird.

heres a pic
http://www.geeks3d.com/public/jegx/2..._by_colfax.jpg

[N19h7m4r3]

heres a pic
http://www.geeks3d.com/public/jegx/2..._by_colfax.jpg

That looks amazing.

It's a pity Tesla and Quadro cards cost so much.

Wouldn't mind trying a pair

[NapalmV5]

Don't know where you read that but no. They both have quad-TMUs and GF100's theoretical peak is lower than Cypress and waay lower than Hemlock. If it has any chance of competing in texture limited scenarios it would need much higher efficiency.

fermi: x*y=z
x = texture units
y = shader clock
z = texture fillrate


fermi:
higher pixel+texture fillrates
higher shader power
higher memory bandwidth/size
higher efficiency across the board

and more but according to ati brilliants here it barely matches 5870 lol

[Macadamia]

God, get rid of theiz ATI fanbois in teh Fermi thread! Their ruining my premature cum experienzzzzzz!


Sigh, I wish ChrisRay was still alive so there was actually somebody on the green side that I could truly respect.

[mindfury]

fermi: x*y=z
x = texture units
y = shader clock
z = texture fillrate

Feimi tmu is running at half of the shader clock.

[Florinmocanu]

fermi: x*y=z
x = texture units
y = shader clock
z = texture fillrate


fermi:
higher pixel+texture fillrates
higher shader power
higher memory bandwidth/size
higher efficiency across the board

and more but according to ati brilliants here it barely matches 5870 lol

Huh? according to this table released by one of the french previews of fermi, your wrong.

[kgk]

Huh? according to this table released by one of the french previews of fermi, your wrong.

Doesn't the 5970 have a 2x256 bit memory interface, not a unified 512?

[SKYMTL]

I think he means just 1 8pin & 1 6 pin instead of dual 8 pins as the first rumors regarding the card's power req.

This.

[firas]

God, get rid of theiz ATI fanbois in teh Fermi thread! Their ruining my premature cum experienzzzzzz!

haha nice one