Can this be true....475 for the 6800Ultra core now?

[Sunrise]

Won't the A2 core be the revision used for the 6800 Ultra's real world debut? Wasn't the A1 revision to be found only on reviewer/reference cards?

If A2 can be clocked at 475Mhz (and like I said, I know that it can, I've had that confirmed), why would they only ship it at 400Mhz?

The "possibility" that it can go that high, doesn´t make it a "fact" that it will be released with that clocks (and i trust you with that 475Mhz-part).

1. NV already has not the best yields at IBM, 222 million transistors on 0.13µm with IBMs 300mm wafers is actually not that cost-effective, but it improves.

2. NV needs to sell those 6800Ultras with high margins, if ATI doesn´t come close this round, there is no reason to price it lower to be competitive.

3. NV and ATI are both having problems getting RAM that is >600MHz GDDR3 right now, and that won´t improve till end of May/beginning of June at the earliest.

A refresh will be based on that A2-core, maybe they even bring it up close to 500MHz, or they are doing another respin if they think that helps, and they can sell this with better memory in August/September time-frame. Then and only then, prices will be significantly lower for 6800Ultra.

 

[Unknown Soldier]

The problem for Nvidia though is: <-- Remember it's only if ATI rumours are true.

1) If ATI announce the X800 Pro on 26th April and it costs $300
2) If the X800 Pro is at least equal in performance to the 6800 Ultra(Or at least beat the 6800 Ultra at some benchmarks)
3) X800 Pro Stock is immediately available unlike the 6800 Ultra which will only be available after a month
4) ATI then can actually say there card is faster than Nvidia's (I don't think they'll actually come out and say it)
5) Most IHV's will see the R42x as being faster.
6) There's still the X800 XT to be released with stock immediately available

Remember it's only if ATI rumours are true.

The good thing for Nvidia is they already have IHV's in their pocket(Albatron, Shuttle, Asus etc.)
If the current 6800 Ultra is really sold at $399.

Another question. How long would it take Nvidia to Release A2 Stepping??

 

[Geeforcer]

DDR was nowhere near to a new technology when gf3 was released.

It's the price that matters.

 

[Bjorn]

The good thing for Nvidia is they already have IHV's in their pocket(Albatron, Shuttle, Asus etc.)

I thought that IHV (Independent Hardware Vendor) was refering to f.e Nvidia and Ati.

Anyway, i doubt that the AIB partners is a problem for Nvidia. They managed to keep most of them with the NV3X which says a lot

 

[karlotta]

well the power specs one would get/read for the respin is for the PCI-e(x) part and would /could show diff power #s.

Samsung has the GDDr-3 issues.
DDR bac in the GF3 day was much more expensive than DDR today... i think.
Will be nice to have reviews of SHIPPING cards.

 

[Unknown Soldier]

The good thing for Nvidia is they already have IHV's in their pocket(Albatron, Shuttle, Asus etc.)

I thought that IHV (Independent Hardware Vendor) was refering to f.e Nvidia and Ati.

Anyway, i doubt that the AIB partners is a problem for Nvidia. They managed to keep most of them with the NV3X which says a lot

Oops My bad. Change all that IHV to AIB

btw .. what does AIB stand for?

US

 

[Bjorn]

The good thing for Nvidia is they already have IHV's in their pocket(Albatron, Shuttle, Asus etc.)

I thought that IHV (Independent Hardware Vendor) was refering to f.e Nvidia and Ati.

Anyway, i doubt that the AIB partners is a problem for Nvidia. They managed to keep most of them with the NV3X which says a lot

Oops My bad. Change all that IHV to AIB

btw .. what does AIB stand for?

US

Add-in Board (i confess, i had to look it up )

 

[Dave Baumann]

Right now, all Uttar and I are going on is 1. the existence of an A2 revision and 2. that's it. We know an A2 revision exists. We do not know anything at all about it. Forget 475Mhz. You won't see it on the 6800 Ultra. Now, we're thinking that there will be a 6850 Ultra or 6900 Ultra or something with higher clocks.

Are you sure that what hasn't been sent out already isn't A2? Afterall, what is actually stamped on the chip doesn't have to mean much.

Actually, A2 would probably take less power than a 5950U.

Sorry, but if they were about and taking less power than the current ones then there is no way they would have gone to review with the boards they have - first impressions are everything in this game and double slot cooling and messages of 480W PSU's are not messages you want to give unless you have to.

Its not even as though there are a plentiful supply of review samples.

 

[Evildeus]

Well, i do agree with DB. I think that if there's an A2 revision, i don't know why everybody's so excited about it. I for one think it's a bit to much hope and over-reacting.

 

[Bjorn]

Actually, A2 would probably take less power than a 5950U.

Sorry, but if they were about and taking less power than the current ones then there is no way they would have gone to review with the boards they have - first impressions are everything in this game and double slot cooling and messages of 480W PSU's are not messages you want to give unless you have to.

An A2 revision that runs 75 MHz higher and at the same time consumes 27 W less doesn't sound very likely. But perhaps he/they meant less power then the 5950 Ultra if it runs at 400 MHz.

 

[Dave Baumann]

But again, we've supposedly already seen A2 - if this were the case then you'd want to use those in reviews and not go with the large power messages we're hearing. Given the number of review samples that are about then either these are already "A2" or there are such short supplies of any working ASIC's that we won't be seeing any soon, and that I don't believe.

I think its a given that we'll see a little gamesmanship with clocks and revisions and I can certainly see there will be a few "specials" from NVIDIA's AIB's beyond that, but then I think it would be naive to assume that ATI wouldn't expect this. It also seems to be the case that all of a sudden NVIDIA will have no corporate responsibilities to their shareholders and "margins can go hang" and they have an infinite amount of wiggle room regardless of the size of their ASIC and process they are using!

I don’t know if they will or won’t decide 475MHz will be the target or not, I just don’t think they are likely to alter the Ultra specifications days after announcing them and shipping them to reviewers. And, given process differences and probable die sizes you would think that it would be ATI that has the larger wiggle room.

 

[mikechai]

About the init strings in the drivers inf file:-
NVIDIA_NV40.DEV_0040.1 = "NVIDIA Geforce 6800 Ultra"
NVIDIA_NV40.DEV_0041.1 = "NVIDIA Geforce 6800"

As the device id of the previewed engineering samples all have device id of "0041", I keep thinking probably nvidia re-badge the non-ultra as ultra during last minute, and save the real ultra for something else.

Any other explanations ?
________
SUZUKI GSX-R SERIES SPECIFICATIONS

 

[Unknown Soldier]

Just for interest sake.

At 475Mhz the NV40 would still only have a texel fill-rate of 7600 M/Texels/sec compared to the rumoured X800 XT's 9600 M/Texels/sec.

The reason why ATI might feel the X800 Pro compares to the current 6800 Ultra(specs released by Nvidia) is that the X800 Pro has a fill-rate of 6000 M/Texels/sec compared to the 6400 M/Texels/sec of the current 6800 Ultra.

6800 Ultra
256bit bus
400Mhz
16 pipes(32z used in Doom3)
550Mhz GDDR3(1100Mhz)
1 TMUs per pipe
2 PS units per pipe
6 VS units
Fillrate: 6400 Mtexels/sec

X800Pro <-- Rumoured specs from HardOCP
256bit bus
500Mhz
12 pipes <-- Extreme Pipes??
500Mhz GDDR3(1000Mhz)
1 TMUs per pipe
Fillrate: 6000 Mtexels/sec <-- If Clock speed is true of course.

This could mean that the X800 Pro could compete with the 6800 Ultra by equalling or just underpreforming. Hey in some benchmarks it could even be better.

It's all guess work though.

US

 

[Dave Baumann]

They started getting more wafers back and got a better understanding of the yeilds.

 

[Druga Runda]

It would be great that both companies start coming out with new products hammering each other as the start of the season was more than excellent.

I mean NV40 performace is excellent. going even faster than that at the moment is not needed for normal gamers unless we all purchase new monitors so to think that the presented NV40 level of performance will be going mainstream because of X800XT or NV45 is more than appealing to me, but somehow it seems too good to be true.

And in that case the major difference would be price because the real difference between the rumored R420Pro spec and the current NV40 one will be negiligible... maybe 10-15% but with such awsome performace as about a bit less than 2x R9800XT still... let alone even faster cards than NV40... like NV45 or R450 or something allegedly in the pipeline huh...

So give me £100 less for that product (ala R420pro) and you are in the game.

If all those new release come true the R420pro could come in below £200 line quickly

 

[MuFu]

From what I've heard the "480W" figure pertains to the 475/600MHz (likely) card that remains unreleased, not the review samples - hence few reported power issues so far. I agree that if A2 is sampling and does offer massive power consumption reductions + a clock bump compared to A1, they'd send those out to reviewers. It's not as if they'd need many cards or even wait for the qualification process to finish in its entirety. AFAIK 500MHz was the target for A2 so qualifying at 475MHz sounds reasonable.

Plus you can understand how people might get mixed up about rev numbers, short of pulling the HS (which, as Dave mentioned, still might not be a clear indication). There were A0 samples being reported as "the first revision" a while ago and now we hear that the official line from nV is that they are "once again shipping a first revision"; A1.

As an aside - anybody heard anything about Far Cry performance and the current ATi/nV performance gap (talking about R420 vs NV40 here) being largely down to a filtering issue with volumetric effects? It's apparently not related to the current point of speculation regarding Nv3x optimisations.

MuFu.

 

[Dave Baumann]

From what I've heard the "480W" figure pertains to the 475/600MHz (likely) card that's coming, not the review samples - hence few reported power issues so far.

No, it pertains to the Prescott systems it was tested with.

 

[jimmyjames123]

I think its a given that we'll see a little gamesmanship with clocks and revisions and I can certainly see there will be a few "specials" from NVIDIA's AIB's beyond that, but then I think it would be naive to assume that ATI wouldn't expect this.

And wouldn't it be naive to expect that NVDA doesn't know that ATI expects this?

It also seems to be the case that all of a sudden NVIDIA will have no corporate responsibilities to their shareholders and "margins can go hang" and they have an infinite amount of wiggle room regardless of the size of their ASIC and process they are using!

ATI has the same "corporate" responsibilities that NVDA has. ATI has openly stated that they would be willing to sacrifice margins this time around in order to keep the speed crown. I would think that, if anything, NVDA would be even more inclined to sacrifice margins in order to regain the speed crown that they lost with the 9700 Pro. Guess we'll just have to see how things play out. Should be interesting.

 

[Geeforcer]

I think that people should accept 6800U as 400MHz at this point. Now, it could very well be that A2 can be clocked higher, but I am certain that if Nvidia felt they could produce it in quantity in time for launch then that's what review Ultras would have been clocked as. At this point, I'd expect higher-clocked A2s to show up in 512MB cards in a couple of months, but Ultra specks will remain as they are.

 

[mikechai]

As an aside - anybody heard anything about Far Cry performance and the current ATi/nV performance gap (talking about R420 vs NV40 here) being largely down to a filtering issue with volumetric effects? It's apparently not related to the current point of speculation regarding Nv3x optimisations.

MuFu.

..in which R420 being faster ?
________
1300