The official GT300/Fermi Thread

[nnunn]

its giving me a Fermi.

Know whatcha mean. This thread's been giving me fermis for days

[ajaidev]

Yes it can reach 750Mhz* (conditions apply) nvidia will provide a water cooling kit along with fermi

XFX GTX380 H2O XXX 755Mhz WTF

[LordEC911]

I remember G80 surpassing the hype and blowing everybody through the roof. Those were glory days!

There wasn't much hype for G80, especially 6-9months before launch...
If anything G80 was under-hyped up until right before the launch, if I recall correctly it was about a month.

I don't see them charging anymore than the 5970, so pricing isn't really the issue here. NVIDIA knows what their products are worth in the market (remember the 295 launched at $500 the lowest price ever for the highest end part available, look back at the 8800 Ultra price) and they will adjust according to ATI.

So... selective memory or you just weren't around the GPU scene ~7 years ago?
R300, ~$399... ring any bells?

[andres1]

LordEC911 didn't you heard about 7 years of inflation?

[LordEC911]

LordEC911 didn't you heard about 7 years of inflation?

Ok... so now we are taking into account inflation?
$399 in 2002 is $479.80 in 2009.
http://www.bls.gov/data/inflation_calculator.htm

Good try... but unfortunately I am still right.

[LedHed]

I didn't realize you wanted me to talk about ancient hardware prices, what is the relevance? This economy now is nothing like it was 7 years ago, that is why I mentioned 295 price vs the 8800 Ultra price (launch prices).

[LordEC911]

I didn't realize you wanted me to talk about ancient hardware prices, what is the relevance? This economy now is nothing like it was 7 years ago, that is why I mentioned 295 price vs the 8800 Ultra price (launch prices).

"Llowest price ever for the highest end part available..." you didn't specify a timeline. So your comment was incorrect.

Yes, the economy is not what it use to be, so what does that say about the fastest GPU being priced at $399 in 2002 when the world economy wasn't in a recession?

[d50man]

$299 and i will buy it all day and all night just keep it cool enough not to melt

[weston]

So if the new revision works, will we see a release by february? I hope so, it's been too much waiting and rumors. They need to come through with something real.

[***Deimos***]

MSRP of nVidia cards

See if you can find a pattern here (you can win a cookie!)?
High-end, Date, MSRP/Street at launch
GeForce Oct99 $300
GeForce2 Apr00 $350
GeForce3 Feb01 $375
GeForce4 Feb02 $320-$450
GeForceFX Jan03 $400
6800 Apr04 $540
7800 June05 $600
8800GTX Nov06 $600 (Ultra debut at $830)
GTX280 June08 $650

G80 spec rumours early spring 2006 before launch of 7900GTX.

- G80 specs

* Unified Shader Approach
* SM4.0 support
* DirectX 10 complaint
* maybe up to 48 shader units (unified)? or 32 pixel pipelines with 24 ROPs (considering we have 512-bit bus) & 10 vertex units.
* 1ghz core ? no i dont think so .. i'd say 700-800Mhz
* GDDR3/4 memory interface up to 2GHz and 512bit controller.
* 512MB
* SLI ready
* Porduct name : GeForce 8800GTX
* release date : Q3 2006

48 and 72 unified shader units cost way too many transistors, i'm thinking more along the lines of 32 unified shaders. That'll be work enough as it is, since 32 geometry processors + 32 shader processors tend to take a lot of space.

Early 2000-ish. "BitBoys" 3dfx. video cards with 2-4 chips. And ofcourse EVERYBODY was sure "the next one" would have 32MB eDRAM - the fx, the 6800, the 7900, the 8800... After being dead wrong over and over I hope the horse if finally beaten dead.

[***Deimos***]

So if the new revision works, will we see a release by february? I hope so, it's been too much waiting and rumors. They need to come through with something real.

Something real? As opposed to... imaginary? So you believe 'Fermi' is a big joke - nVidia pretending to have a huge 40nm DX11 chip?

What we know.
* Its almost certain its real.
* It will launch, whether good or bad (ie R600, FX)
* Almost certain that it "works" (its unheard of to require A3 silicon to boot Windows to desktop).
* rough performance #FLOPS etc

What nVidia knows and isn't sharing
* image quality Fermi is "supposed" to have. ie angle independent AF? 64xAA?
* all the bugs they found so far. quantity (likely hundreds)? severity?
* initial chip yields and clocks. if respin was to improve this, then the early numbers are meaningless
* special exclusive features (nobody in forums expected 'eyeInfinity' from AMD)

What nobody, not even CEO knows
* final performance. I'm sure nVidia is benching games around the clock, especially for DX11, as they debug issues and optimize.
* final yields. The design was probably done nearly a year ago. Rumoured already at A3. Its now up to TSMC.. gotta pray and hope for the best.
* final clocks.
* extent of future optimization possible.


Let me know if I missed anything important or something should be in different category.

[YukonTrooper]

Let me know if I missed anything important or something should be in different category.

Three more renames of preexisting products.

[weston]

Something real? As opposed to... imaginary? So you believe 'Fermi' is a big joke - nVidia pretending to have a huge 40nm DX11 chip?

I'm not saying that at all. They have faked a lot of stuff already, like the first card which had screws holding the cooler on, the recent benchmarks which were photoshopped, etc...These are not real, but fermi obviously is real. I just want to see legit benchmarks, but if they don't have a working card or finalized clock speeds, that would be impossible at this time. This is so similar to ATIs R600 - lots of delays, lots of hype, lots of fake benchmarks...

[***Deimos***]

I'm not saying that at all. They have faked a lot of stuff already, like the first card which had screws holding the cooler on, the recent benchmarks which were photoshopped, etc...These are not real, but fermi obviously is real. I just want to see legit benchmarks, but if they don't have a working card or finalized clock speeds, that would be impossible at this time. This is so similar to ATIs R600 - lots of delays, lots of hype, lots of fake benchmarks...

No NASDAQ company (let alone Fortune 500) would EVER do what you are implying.

Making up benchmark numbers and deceiving your bond holders and investors WILL SURELY GET YOU IN A LOT OF TROUBLE.

100% sure nVidia said it WAS A PROTOTYPE and would ABSOLUTELY NEVER say its a retail board ready for sale.

Whatever benchmarks you find on the internet ARE DEFINITLY NOT made by nVidia. If they were real, they would be on nVidia's website, and there would be coverage by reuters and wallstreet etc.

Like R600? Maybe. That pretty much assumes you worked on R600 and Fermi and thus have the knowledge to make valid comparison. Delays? Did ATI in 2006 or nVidia in 2009 announce the release date? Hype? Which TV channel was that Fermi commercial on? Where are those big posters and ads on buses?

Virtually everything you're talking about (ie benchmarks, hype, whining about delays), is made up by folks in forums. Just want to make sure everybody is 100% clear - until nVidia actually posts it on their website, its all just rumours.

[pingpong1455]

Well I didn't feel like surfing 41 pages so do you guys have predictions as to fermis release date

[LordEC911]

100% sure nVidia said it WAS A PROTOTYPE and would ABSOLUTELY NEVER say its a retail board ready for sale.

Like R600? Maybe. That pretty much assumes you worked on R600 and Fermi and thus have the knowledge to make valid comparison. Delays? Did ATI in 2006 or nVidia in 2009 announce the release date? Hype? Which TV channel was that Fermi commercial on? Where are those big posters and ads on buses?

Virtually everything you're talking about (ie benchmarks, hype, whining about delays), is made up by folks in forums. Just want to make sure everybody is 100% clear - until nVidia actually posts it on their website, its all just rumours.

BS. Jensen said, "This puppy here is Fermi." It was a mockup, no silicon, the board was sawed down to size. When first questioned, many Nvidia employees informed the questioners that it was a "real" card and not a mockup. Then after it was out in the open, then they officially stated it was a mockup.

BS... this info is NOT made up by people on the forums. Some people have contacts and info about internal roadmaps/timelines. Hell, just look at the tape-out dates to see for yourself that both R600 and GF100 is late, AMD/ATi has even admitted that R600 was late. You expect to launch on A2/A12 silicon, not A3/A13. Things could have been a lot worse, in regards to GF100 delays, if Nvidia hadn't taken the risk of doing hotlots, would have added at least another month if not two.

[Weevil]

Well I didn't feel like surfing 41 pages so do you guys have predictions as to fermis release date

No, but I'm predicting there will be over 2500 posts in this thread by time it is.

[weston]

Please remove your head from your ass (or put back in).. whichever.

No NASDAQ company (let alone Fortune 500) would EVER do what you are implying.

Making up benchmark numbers and deceiving your bond holders and investors WILL SURELY GET YOU IN A LOT OF TROUBLE.

100% sure nVidia said it WAS A PROTOTYPE and would ABSOLUTELY NEVER say its a retail board ready for sale.

Whatever benchmarks you find on the internet ARE DEFINITLY NOT made by nVidia. If they were real, they would be on nVidia's website, and there would be coverage by reuters and wallstreet etc.

Like R600? Maybe. That pretty much assumes you worked on R600 and Fermi and thus have the knowledge to make valid comparison. Delays? Did ATI in 2006 or nVidia in 2009 announce the release date? Hype? Which TV channel was that Fermi commercial on? Where are those big posters and ads on buses?

Virtually everything you're talking about (ie benchmarks, hype, whining about delays), is made up by folks in forums. Just want to make sure everybody is 100% clear - until nVidia actually posts it on their website, its all just rumours.

lol, way to take everything i said out of context, twist my words around to something entirely different, and at the same time, be a complete .

*adds to blocked list

[Mescalamba]

Interesting flamewar.. or whatever it is. I´m pretty curios how good it will be and when it come out. I hope that it will be sooner than ATi finishes HD 6XXX series. We need nVidia, more than we can imagine.

[Andrea deluxe]

[***Deimos***]

nvidia does handle AF more efficiently but if you compare the 5870 to the 4890 the hit is much worse. its double the speed but gtexels are the same as 4890? so much for them telling us it didnt come with a performance hit. ATi has historically been known for doing graphics efficiently. this card completely forgets that. SSAA and the new AF are there to prove it. brute force methods like those might be inefficient but the image quality is great. i am expecting fermi to do the same. the cards are so powerful it doesnt really matter.

Just because currently HD58XX use very high quality AF and AA, doesn't mean that in future drivers cant disable/optimize. It just means that HD58XX have small "performance reserves", probably intended for Fermi launch.

Tesla pamphlet.. confirms what I've said all along - mostly less 512SP parts until yields improve.

As for Fermi
I feel their pain. I'm sure LOTS of overtime at N hq. I hope '09 ended well with major problems fixed. Also hope 'Dear Leader' was nice enough to unlock chains and release workers for holidays

Well rested and with clear minds, I'm sure nVidia will make make good progress after holidays.

[Chumbucket843]

AF is undersampled in 5000 series so it is a little better than 4000 series when it comes to blurriness/shimmering. sure it is angle independent but thats just a circle, not an in-game moving texture and who cares about mip levels? the point of that was to show scaling and the fact that higher quality comes at a cost but performance is still quite fast.

if you did not read the interview with ATi catalyst team they get high performance driver out the door and debug and improve xfire support in later drivers. its a little wacko to cripple your own card suspecting that the competition will be faster.

i have a suspicion that nvidia will take a similar route but have better sampling. they have 128 tmu's compared to 80 tmu's on rv870 so they have headroom.

[NickF]

I love how after all the delays they still couldn't get all 512 cores enabled, and running at a decent speed. Looks like nVidia really flopped it hard.

[H2O]

Correct me if I am wrong, but aren't the workstation level cards going to use A2 while the desktop cards will use A3? If so, wouldn't that mean that there's still hope for the GeForce cards shipping with 512 cores?

[ulticool]

Been holding off purchasing anything, but perhaps it's time to go ahead with an HD5870...

Ahem** cough**

Wait, trust me