NVIDIA GF100 & Friends speculation

[Bouncing Zabaglione Bros.]

Considering the battering they have been receiving since the rv770 bomb, I'd guess they'll launch mainstream stuff before B1.

Surely you mean they'll "rebrand old stuff again"? We've heard nothing about mainstream Fermi products.

 

[rpg.314]

Surely you mean they'll "rebrand old stuff again"? We've heard nothing about mainstream Fermi products.

No, launch != rebrand in my book.

May be we haven't heard anything all this while because all the attention to GTX4x0 is helping keep attention away from them, just the way they like it.

EDIT: Charlie hasn't said anything about their tapeouts, so it is possible that they haven't been taped out yet. Here he mentions that it takes ~20-22 weeks from tapeout to shipping, with no metal spins. That puts is ~6 months out from today, . TBH, this sounds too dire to be true.

 

[Alexko]

Considering the battering they have been receiving since the rv770 bomb, I'd guess they'll launch mainstream stuff before B1.

Wouldn't it be wiser to fix what ails Fermi before, in other words finish B1 (or B2 if needed) and then scale it down?

 

[neliz]

Wouldn't it be wiser to fix what ails Fermi before, in other words finish B1 (or B2 if needed) and then scale it down?

they still have quite some dies waiting... if you remember Charlie's risk wafer article from half a year ago. it would totally explain "good" availability a few months after launch as a stop-gap solution for GTX475/485.

 

[rpg.314]

Wouldn't it be wiser to fix what ails Fermi before, in other words finish B1 (or B2 if needed) and then scale it down?

A3 is supposed to be the launch spin. So the functionality has been fixed. What remains is power, clocks and yields. Since the mainstream parts are new/different chips, their power, clocks and yields can be fixed independent of fermi's (atleast from 30k feet).

 

[jimbo75]

When you say its power draw is in line with it's performance, and we are assuming its performance is closer to the 5870 than the 5970 - are you saying its power draw is closer to the 5870 too? That would require it to be below 240 watts I believe.

 

[neliz]

When you say its power draw is in line with it's performance, and we are assuming its performance is closer to the 5870 than the 5970 - are you saying its power draw is closer to the 5870 too? That would require it to be below 240 watts I believe.

It would be more fair to run a 5870 @1Ghz and compare everything like that.

 

[Sontin]

Did you even read the what you quoted? How can there be "little to no" product until July even after the launch, but B1 will somehow be available in massive quantities in May?

Good question but i'm the wrong guy for the answer.
Neliz said that B1 is the "shipping product". How could they ship 70-210k chips without the "shipping product" until may?

 

[Bouncing Zabaglione Bros.]

they still have quite some dies waiting... if you remember Charlie's risk wafer article from half a year ago. it would totally explain "good" availability a few months after launch as a stop-gap solution for GTX475/485.

Depends on where they parked the wafers. If they wanted to change metal layers, then parked wafers are useful. As B1 is a change to silicon layers, I'm not convinced what they've parked will be much use to them. If it is, then there's so little on the parked wafers, it won't save them much time.

 

[Bouncing Zabaglione Bros.]

Good question but i'm the wrong guy for the answer.
Neliz said that B1 is the "shipping product". How could they ship 70-210k chips without the "shipping product" until may?

You've mis-read what Neliz wrote. The end of March launch will be a paper launch, followed by a few Press Editions cards in very short supply. The real "shipping product" won't be at the official launch, it will be in Q3 when Nvidia finally (hopefully) get things fixed with a B1 tapeout. B1 is hoped to be the "real shipping product" whenever B1 is ready - quite a lot later than the official launch with little to no product and PE cards.

 

[Sontin]

You've mis-read what Neliz wrote. The end of March launch will be a paper launch, followed by a few Press Editions cards in very short supply. The real "shipping product" won't be at the official launch, it will be in Q3 when Nvidia finally (hopefully) get things fixed with a B1 tapeout. B1 is hoped to be the "real shipping product" whenever B1 is ready - quite a lot later than the official launch with little to no product and PE cards.

No, that is not what he wrote*. And if you believe that 70-210k cards are a "paper launch followed by a few Press Editions cards in very short supply" then you could say the same about Cypress and Hemlock, too.

*

Availability for May should be about 6 to 7 times that of Rv770/Rv870 at launch

Hence the no-to-limited availability etc. until June/July. It also is the basis for Charlies "handful" of chips for partners, A3 will be here for launch and B1 will be the "shipping" product.

 

[Bouncing Zabaglione Bros.]

No, that is not what he wrote*. And if you believe that 70-210k cards are a "paper launch followed by a few Press Editions cards in very short supply" then you could say the same about Cypress and Hemlock, too.

Where did the 70-210 K number come from? I thought it was 5-10K?

I bought a 5850 last October within a week of launch day. I had to wait a few days for it to come into stock. That's a lot better than Nvidia's "October, November, December, January, March" constantly moving release dates where I'd be waiting six months and counting. You can't compare Cypress where you could always get product in a few days, and Fermi where the product is already half a year late and still not available.

 

[Silus]

So the final clocks have been sent to partners correct ? Were there any leaks of those specs ?

All I've heard is that the GTX 470 clocks weren't final, so the numbers we've seen in the leaked benchmarks may not be the real ones.
As for the GTX 480, it's still being kept secret.

 

[psolord]

So...any word or speculation regarding the idle power consumption of GF100 based cards?

GT200 held it's ground regarding idle power consumption, although it was a big chip. Is there a reason we shouldn't expect the same with GF100 or the opposite?

 

[rpg.314]

Well, what you quoted included more than just dates. silent_guy has shown several times that when he writes about anything BUT dates, he's usually wrong.

Architecturally, perhaps. Overall manufacturing pipeline wise, much less so.

And yes, that's my point, about other tapeouts. He has already confirmed that he doesn't check his sources 100% (surprise, surprise) and recently, after I called him on it, retracted his statement of no other tapeout, for a rumor of another tapeout...

He said he wasn't 100% sure of it, so you can't call that wrong either.

Anyway, there's a dedicated thread in RPSC for this branch of discussion.

 

[Silus]

So...any word or speculation regarding the idle power consumption of GF100 based cards?

GT200 held it's ground regarding idle power consumption, although it was a big chip. Is there a reason we shouldn't expect the same with GF100 or the opposite?

IMO, no. There's probably even more progress in the idle power draw for GF100, than it was for GT200.

 

[Alexko]

A3 is supposed to be the launch spin. So the functionality has been fixed. What remains is power, clocks and yields. Since the mainstream parts are new/different chips, their power, clocks and yields can be fixed independent of fermi's (atleast from 30k feet).

But wouldn't that mean dividing up resources between several dies to fix what is essentially the same problem?

He has already confirmed that he doesn't check his sources 100% (surprise, surprise) and recently, after I called him on it, retracted his statement of no other tapeout, for a rumor of another tapeout...

Seriously, you need to take it down a notch, that's borderline slander. Charlie posted something on the forum, warning that he wasn't 100% sure.

He later said he hadn't heard anything about a new tape-out, and a few days later he apparently had, so he mentioned it.

 

[mczak]

A3 is supposed to be the launch spin. So the functionality has been fixed. What remains is power, clocks and yields. Since the mainstream parts are new/different chips, their power, clocks and yields can be fixed independent of fermi's (atleast from 30k feet).

But if they were supposed to appear in a similar timeframe as GF100, it seems likely nvidia would have just made the same mistakes with them no? Though I'd guess a similar timeframe as GF100 B1 would make sense - use all what you've learned from the mistakes for GF100 Ax, that'll mean they could be ready about the same time.

 

[Sontin]

Where did the 70-210 K number come from? I thought it was 5-10K?

Read his posting.
I read that amd shipped 10k-30k cards for the Cypress launch.

 

[rpg.314]

But wouldn't that mean dividing up resources between several dies to fix what is essentially the same problem?

Or you can play it like amd and put 2 different teams to fix the same problem for both the dies, sharing the work.