NVIDIA Fermi: Architecture discussion

[Groo The Wanderer]

I'm american, but one of the cool ones so I am too.


I'm not disagreeing on timeframe/slideleaks, I'm disagreeing that the 5970 is a response to anything. It's been planned for a while.

The response to Fermi is still waiting in the wings at AMD, waiting for Fermi to actually launch before they spring it.

Interesting, I didn't know they told anyone about that part yet.

-Charlie

 

[trinibwoy]

Also, since when do tesla cards have a display connector?

Why should we expect that to be a real card? It's not like they'll have cards ready to go now just sitting around till May waiting to be sold

 

[digitalwanderer]

Interesting, I didn't know they told anyone about that part yet.

They didn't, you don't really consider me just "anyone"...do you?

 

[Florin]

Also, since when do tesla cards have a display connector?

genüber heise online erklärte Nvidia-Mitarbeiter Andrew Humber, dass die Karten, die nun erstmals auch einen DVI-Ausgang besitzen, nicht vor Mai 2010 verfügbar sein werde

In other words, Tesla cards will have a DVI output.

 

[mapel110]

The response to Fermi is still waiting in the wings at AMD, waiting for Fermi to actually launch before they spring it.

Is AMD really so dumb to hold back a completed product?

Somebody should tell AMD that time is money. They should release anything when it's ready. When there is no opponent, they can get more money for it.

 

[trinibwoy]

Well given how hard it is for TSMC to make the "easy" RV870 it's fair to assume that this uber Fermi-killer won't fare much better aye?

 

[DegustatoR]

This one board doesn't look sawed off, but the dark backplate still blocks the SLI connector.

The PCB is unified but there is no need for SLI connector on Tesla. Strange that noone figured this out yet.

 

[Arty]

Well given how hard it is for TSMC to make the "easy" RV870 it's fair to assume that this uber Fermi-killer won't fare much better aye?

Real anything wins over paper dragons.

 

[trinibwoy]

So now yields don't matter when there's no competition?

Did anyone post this nice summary from vr-zone yet?

http://vr-zone.com/articles/fermi-in-trouble-/8054.html?doc=8054

 

[mapel110]

So now yields don't matter when there's no competition?

Did anyone post this nice summary from vr-zone yet?

http://vr-zone.com/articles/fermi-in-trouble-/8054.html?doc=8054

Guess not, but I like this one more.
http://www.semiaccurate.com/2009/11/16/fermi-massively-misses-clock-targets/
Nice rant from Charlie

 

[digitalwanderer]

Double ouch!

 

[trinibwoy]

Guess not, but I like this one more.
http://www.semiaccurate.com/2009/11/16/fermi-massively-misses-clock-targets/
Nice rant from Charlie

9500GT owners live in basements? I lol'd

But no thanks, I like my articles heavy on the info but light on self-praise and other niceties that compensate for real life insecurities. Lots of points for greater than semi-accuracy though.

 

[Arty]

So now yields don't matter when there's no competition?

Did anyone post this nice summary from vr-zone yet?

http://vr-zone.com/articles/fermi-in-trouble-/8054.html?doc=8054

Yields matter when there is competition, you cannot simply take context out of the equation.

 

[seahawk]

9500GT owners live in basements? I lol'd

But no thanks, I like my articles heavy on the info but light on self-praise and other niceties that compensate for real life insecurities. Lots of points for greater than semi-accuracy though.

But from all gossip sides, he seems to be right on track when it comes to Fermi.

Late, slow and expensive was what he said and it seems to be exactly that.

I can´t wait to write "EPIC FAIL for NV" when that thing launches in .... well May?

 

[MfA]

So now yields don't matter when there's no competition?

It constrains volume, but it didn't matter much for cost in the past because TSMC almost certainly provided discounts to offset them ... TSMC needs to fill the capacity anyway and get the runs necessary to optimize the process.

At this point though ATI almost certainly can fill the capacity (if I am to believe the reports of shortages). So are they going to give NVIDIA twice or more the discount per mm2 or will NVIDIA pay twice or more per mm2 of functional die than ATI?

 

[Ailuros]

It constrains volume, but it didn't matter much for cost in the past because TSMC almost certainly provided discounts to offset them ... TSMC needs to fill the capacity anyway and get the runs necessary to optimize the process.

At this point though ATI almost certainly can fill the capacity (if I am to believe the reports of shortages). So are they going to give NVIDIA twice or more the discount per mm2 or will NVIDIA pay twice or more per mm2 of functional die than ATI?

Or you can thicken the plot even more after reading this part: http://www.semiaccurate.com/2009/11/16/ati-58xx-parts-delayed-bit-more/

I'm daring to ask myself if TSMC truly has as many problems with 40nm as reported so far or if it's rather convenient for both IHVs to believe that they're as massive as reported. If I am to go into any awkward conspiracy theories the latter sounds more reasonable but at this point I doubt we'll find out.

 

[MfA]

I don't see a motive for them to keep both IHV's back ... it's not like TSMC is without competition. As for keeping back ATI specifically, I don't see that happening without 100s of millions changing hands if it happened at the corporate level.

 

[trinibwoy]

Late, slow and expensive was what he said and it seems to be exactly that.

It's certainly late. Don't recall seeing price or performance numbers anywhere. You'll have to wait a few months to tally his score.

It constrains volume, but it didn't matter much for cost in the past because TSMC almost certainly provided discounts to offset them ... TSMC needs to fill the capacity anyway and get the runs necessary to optimize the process.

Sure, but TSMC swallowing costs doesn't increase volumes. Remember the original question is why AMD doesn't launch its Fermi killer today.

At this point though ATI almost certainly can fill the capacity (if I am to believe the reports of shortages). So are they going to give NVIDIA twice or more the discount per mm2 or will NVIDIA pay twice or more per mm2 of functional die than ATI?

Hmmm don't follow. What does ATIs allocation have to do with Nvidia's discounts? The terms of those arrangements are negotiated independently no?

 

[Ailuros]

I don't see a motive for them to keep both IHV's back ... it's not like TSMC is without competition. As for keeping back ATI specifically, I don't see that happening without 100s of millions changing hands if it happened at the corporate level.

Well I'm obviously answering that sci-fi theory in my second paragraph of my former post.

 

[MfA]

Of course they are negotiated independently ... I'm just asking what the results of those negotiations are. If they are at capacity and they give NVIDIA a larger discount per mm2 then any allocation to NVIDIA will loose them money in the short run. In the long run they want NVIDIA&AMD to stay competitive of course, but still it looses them money in the short run.

So, how charitable are we to assume TSMC is?