The GT300/Fermi Thread - Part 2!

[Musho]

They scrapped Fermi altogether it seems. They are going to fight against ATI with their oldgen cards with tweaked drivers! No seriously, they've managed to get a pretty nice gain in a few games. Hopefully that gain will be seen in games that aren't listed as well. This does proof driver development never really stops. A better Fermi driver might actually bring the performance way higher, even if Charlie's initial numbers are correct.

[Andrea deluxe]

GTX 480

[FischOderAal]

I'm sceptical on how they achieved that, just like saaya. How can you pull such an improvement out of nowhere without decreasing IQ? NVIDIA had plenty of time to improve the drivers, why suddenly now?

€dith: Thanks @ Andrea deluxe!

[Olivon]

Thanks Andrea, nice pics

[saaya]

http://en.wikipedia.org/wiki/Infinite_monkey_theorem

Basically, if someone writes enough FUD (like a thousand-word rant about NVIDIA, their chips, etc.), some of it may turn out being right but that doesn't equal KNOWLEDGE about the topic.

wait... let me get this straight... your saying that everything charlie was right about was a pure coincidence of him just writing up so many rumors that one of them had to be true...

of course hes not always right, who is?
and with nvidia and others purposely spreading wrong information to their partners and contacts and even within the company, how can you blame him?
so what, if 2 people you know for a long time both contact you and give you very similar hints about an upcoming product, you wouldnt write it up because you cant confirm it and it might be wrong?

my guess is that either he speculated that there wont be 512parts at launch, or whoever fed him infos speculated that there wont be 512core parts at launch, because the yields of fully functional parts was that bad and they couldnt ship a notable amount of cards. and maybe they initially didnt plan it and then they did, maybe because they have to, for pr reasons, to have the fastest part in reviews as they cant clock the gpus as high as they had hoped... maybe their yields improved...

first he said yields were so bad that there wouldnt be 512 parts at launch. then he said there would be some, but very few... those statements are very close, if there are very few press sample 512core cards then you might as well say there arent any 512cards. just like people claimed the 5870 was a paper launch because availability in some regions was terrible...

even if jensen himself would be your best buddy and tell you what would happen when, he would be wrong over and over because plans change... unexpected things happen...

GTX 480 sale price UK £449.99
GTX 470 sale price UK £379.99

dont ask where i got these numbers cause i aint telling, but this is the recommended RRP for these parts come end of march

670$ 494E
565$ 417E

so basically
699$ 499E
599$ 449E

?

according to pcgh it should be:
749$ 549E (+80$/50E vs your price)
475$ 349E (-100$/-67E vs your price)

why the confusing prices?
so roughly 650-750$ for the 480 (daym!) and 450-550$ for the 470 (wow!)

at that price the 480 better beats the 5970 across the board and the 470 should whipp the 5870s 4ss!

name - cost - %perf - %price - %/%
GF100-U 3gb 700$ 117% 117% 0
HD5970 2gb 600$ 100% 100% 0
GTX295 2gb 525$ 81% 87% +6
GF100 1.5gb 500$ 83% 83% 0
HD5870 1gb 400$ 71% 67% -4
HD5850 1gb 325$ 62% 54% -8
GTX285 1gb 375$ 58% 62% +4
GTX260 1gb 225$ 47% 37% -10
HD5830 1gb 250$ 45%*1 42% -3
HD5770 1gb 175$ 43% 29% -14
HD4870 1gb 175$ 44% 29% -15
HD5750 1gb 150$ 37% 25% -12
GTS250 1gb 125$ 36% 21% -15
HD4850 1gb 125$ 35% 21% -14

4870x2 vs 5870
1600 vs 1600 shaders - same
750 vs 850 core mhz - 13% up
230gb/s vs 150gb/s - 34% down
100% vs 97% performance

gtx295 vs gtx470
480 vs 480 shaders - same
576 vs 625 core mhz - 8% up
1242 vs 1250 shader mhz - same
224gb/s vs 160gb/s - 29% down (guessing 4ghz 320bit)
100% vs 95% performance

lets be optimistic here and say the gtx470 is as fast as the 295 for "some" reason... maybe even slightly faster... this would mean its 15% faster than a 5870 on average. then 500$ is a reasonable price... in that case id expect the price to be 550$+ because its dx11 and its the latest greatest etc, and the 295 is available for 525$ right now...
and even if its slightly slower than a 295, id still expect the same price or higher than a 295, because.... its dx11, and its FERMIIII!!!!

gtx470 vs gtx480
480 vs 512 shaders - 7% up
625 vs 650 core mhz - 4% up (guessing)
1250vs 1300 shader mhz - 4% up (guessing)
160gb/s vs 230gb/s - 29% up (guessing 4.8ghz 512bit)
100% vs 115% performance

this should make it as fast as a 5970, maybe sliiightly faster...
a 40% higher price than the 470 for a 15% performance boost and a 15% higher price than the 5970 for roughly the same performance, even if its a single gpu... thats a bit much...
but wouldnt surprise me as i know there are enough people out there willing to pay that much

nice pics andrea!!!
looks like a 480, 12 memory chips, 6 channels, 512bit... 8+6 power... yepp, definately a 480
6 power phases... but not digital?
oh! no more seperate NVio chip! didnt know that!

[-=DVS=-]

If its 512 bit bus it should have either 1GB or 2GB of ram yes ?

[CadESin]

I really like the PCB layout, its nice and clean, I am starting to fear it will run really hot, only reason I could see to add the cut out in the PCB is if the cooler was unable to work well without it.

oh! no more seperate NVio chip! didnt know that!

Its really about time Nvidia did that, will also save them some money I bet.

I am starting to think I will really get to see Fermi in person at PAX later this month, It will be nice to have a two player field again in PC graphics.

[-Sweeper_]

If its 512 bit bus it should have either 1GB or 2GB of ram yes ?

it's 384-bit GDDR5

[BenchZowner]

It's definitely 384bit for the GTX 480, and 320bit for the GTX 470.

[SKYMTL]

[highoctane]

I am starting to fear it will run really hot, only reason I could see to add the cut out in the PCB is if the cooler was unable to work well without it.

Odds are the holes are tailored for SLI configs running nut to butt, probably doesn't affect single cards very much.

[zalbard]

I really like the PCB layout, its nice and clean, I am starting to fear it will run really hot, only reason I could see to add the cut out in the PCB is if the cooler was unable to work well without it.

I'm with you here, it looks really good. Hope it performs just as well!

[Olivon]

Encore DVHardware

And some enlightments :

The reference cooling solution for the GeForce GTX 480 is pictured below, but it's said to be quite expensive so some vendors may use cheaper alternatives. The launch of Fermi is expected to be quite small in size, the anonymous company that showed off this card said NVIDIA is expected to ship only 5,000 units at launch, and confirmed the story that second-tier graphics card makers will have to wait a bit longer for Fermi chips.

More supply on GTX 470, maybe ...

[SKYMTL]

wait... let me get this straight... your saying that everything charlie was right about was a pure coincidence of him just writing up so many rumors that one of them had to be true...

No. I'm saying even a blind man with a machine guy will hit a target if he knows which way to point and is given some pointers.

[Piotrsama]

So where's the knowledge about the topic? On this thread?

No knowledge here, the real deal were the tags on the part 1 thread

[Nedjo]

Wow. If FUDO writes something like this the performance must be bad. I mean, "faster than HD 5870 in some cases". Ouch...

I can't find "Fuad Abazovic" nowhere in that piece of writing?!

[LiquidReactor]

Wow that IHS is smaller then that on GTX 285 die.

Holy smokes its almoust the size of the G80 ihs.

Hahahaha....snap.

[jaredpace]

Wow that IHS is smaller then that on GTX 285 die.

Holy smokes its almoust the size of the G80 ihs.

Hahahaha....snap.

Yeah i noticed that too, doesn't mean the die coincides with ihs size through generations. Still betting it's between 490 and 560mm. ~60% bigger than a 5870 die.

[FischOderAal]

I can't find "Fuad Abazovic" nowhere in that piece of writing?!

You're right. Meant Fudzilla not Fudo. 'doh.

[LiquidReactor]

Yeah i noticed that too, doesn't mean the die coincides with ihs size through generations. Still betting it's between 490 and 560mm. ~60% bigger than a 5870 die.

According to who? I seriously doubt nVidia specifically changed the IHS mounting design for this chip. Actually wouldnt it be best to have a larger IHS if the chip itself is big to maximize contact area with the heatsink?

Basically all the 470-500mm^2 die size rumors are false.

[H2O]

IHS shot:

[Mk]

IHS shot:

A2??? i thought they will be on A3 or even B1 by now

[jaredpace]

Basically all the 470-500mm^2 die size rumors are false.

Oh, what are the die dimensions then?

[trinibwoy]

He said in 20\02 there was going to be a 512sp, someone said today he was going start to back peddling because in fact there is going to be a 512sp, and you agreed, do i need to make a drawing or it's this enough? And i don't know how why correcting in 20\02 something he said in 17\02 is a bad thing.

You don't see anything wrong with stating something as a fact and then changing your facts three days later? No wonder he has so many apologists...the standard he's held to is dirt low

Btw, what will you say if GTX 480 ends up having neither 512 or 448 shaders? There's also a 480 shader rumour floating about as well.

[LiquidReactor]

Oh, what are the die dimensions then?

yawn

Im just going to respond with this

IHS shot:

oh heres a link to semiaccurate. Im sure your friend charlie can put another spin on this

Oh god, he already got the latest story out. Whats that? Oh, nVidias A3 revision silicon was so bad nVidia had to use underclocked shader crippled A2 chips in GTX 470/480?


latest charlie story

Oh and I love your sig. Please dont take that down