NVIDIA Fermi: Architecture discussion

[Creig]

2 cards before christmas?
I guess one for you and one for me

So where are the other five functional GPUs going?

 

[no-X]

Four. Cause the third one sits in the famous terminator board

 

[trinibwoy]

2 cards before christmas?
I guess one for you and one for me

Lol too easy

 

[nanometer]

It kinda depends on the respin needed.

Please explain how nvidia can launch with "considerable quantities" given that 2 respins are needed from now till January...

 

[neliz]

Please explain how nvidia can launch with "considerable quantities" given that 2 respins are needed from now till January...

Also depends on the market they might target.
Remember that Tesla was launched before the GT200b3 showed up in consumer cards, thus effectively "hard" launching in their target market with minimal quantities (2 months before the first system was delivered) on an inferior spin (b2) with hand-picked GPU's.

You might see a Tesla announcement first and a GeForce launch date mid-Q1.

 

[Arty]

Nothing's really changed. It always was January for the first considerable quantities, no? As for will they or won't they announce it in December -- who knows.

+

Really now?

Feel free to search the database for something definite or even a "bet" from my behalf.

Ok, I wouldnt call it a 'bet' in your case as you were more liberal on time lines and never really committed like Deg.

 

[Groo The Wanderer]

Uhm... there is no way Fermi ships in January if it needs an A3. Absolutely no way, unless they tape-out A3 only very shortly after having taped-out A2 and much before having received the latter back from the fab. Which is extremely ridiculous and would indicate a massive screw-up. Most likely Fudo is just (once again) clueless and trying to rationalize how Fermi could have launched in late November/early December when that was never on the cards. Late December probably was, but I can easily believe that slipped to January (and of course they might still need an A3, just Fudo isn't the right person to determine that).

My personal suspicion is that the NV PR people who are feeding him are no longer able to keep a straight face, and are starting to tell him bits of the truth. Now he has to rationalize that with their past lies. Hard job.

-Charlie

 

[Kaotik]

Also depends on the market they might target.
Remember that Tesla was launched before the GT200b3 showed up in consumer cards, thus effectively "hard" launching in their target market with minimal quantities (2 months before the first system was delivered) on an inferior spin (b2) with hand-picked GPU's.

You might see a Tesla announcement first and a GeForce launch date mid-Q1.

b2 was good enough iirc for all but GTX295, and was used from day 1 on GTX275/285

 

[LordEC911]

b2 was good enough iirc for all but GTX295, and was used from day 1 on GTX275/285

Umm... all retail cards I have seen, GTX275/GTX285 have been B3.

 

[Kaotik]

Umm... all retail cards I have seen, GTX275/GTX285 have been B3.

After a quick check up, you're right it seems; GTX260 216's with 55nm cores however have been spotted with B2

 

[LordEC911]

After a quick check up, you're right it seems; GTX260 216's with 55nm cores however have been spotted with B2

Correct, that is what I found as well.
G200-103-B2 = GTX260+ and is the only desktop B2 and the main pics I found were from early reviews right around launch time.
G200-103-B3 = GTX260+ for sure in the lightning series.
G200-105-B3 = GTX275
G200-350-B3 = GTX285
G200-400-B3 = GTX295

 

[ChrisRay]

So where are the other five functional GPUs going?

To me of course.

 

[neliz]

After a quick check up, you're right it seems; GTX260 216's with 55nm cores however have been spotted with B2

Yeah, only the first review samples and some minor production runs where B2, it was replaced as soon as they could.
I'm pretty sure some people here can show you reviews of cards with b2 samples but you won't find those in retail.

 

[zildjian]

Correct, that is what I found as well.
G200-103-B2 = GTX260+ and is the only desktop B2 and the main pics I found were from early reviews right around launch time.
G200-103-B3 = GTX260+ for sure in the lightning series.
G200-105-B3 = GTX275
G200-350-B3 = GTX285
G200-400-B3 = GTX295

hi,

gpu-z and Riva tuner states that i have a b1 here's the screenshot:

 

[Tchock]

To me of course.

So you're FAHMan

 

[Vincent]

2 cards before christmas?
I guess one for you and one for me

2 cards without NVIOs

 

[LordEC911]

hi,

gpu-z and Riva tuner states that i have a b1 here's the screenshot:

Have you taken the cooler off and looked at the IHS?
B1 didn't see production, so either that is an ES or some sort of glitch/old software.

 

[neliz]

gpu-z and Riva tuner states that i have a b1 here's the screenshot:

GPU-Z 0.31 is bugged as far as I know, it also displays an incorrect texture fillrate.

GPU-Z 0.3.1 & GTX285 -> incorrect texture fillrates
I bumped into an apparent bug in GPU-Z 0.3.1.
It seems like the app calculates GTX285's texture fillrates based on a weird TMU amount of ~73.2 units. In reality, GTX285, just like GTX280 has 80TMUs.

For example, in this screenshot GPU-Z displays 56.6GTexels/s
56600MTexels/s / 773MHz = ~73.2Texels/clk

Correct fillrate figure would be:
773MHz * 80Texels/clk = 61840MTexels/s = 61.8GTexels/s

Taking your 648Mhz * 80T/c = 51.8GT/s instead of your 47.7

 

[DegustatoR]

Please explain how nvidia can launch with "considerable quantities" given that 2 respins are needed from now till January...

They aren't "needed".

 

[LordEC911]

They aren't "needed".