Nvidia confirms GT-200 with 1 billion transistors

[OutSider]

fudzilla what i can say anymore than that

[AliG]

actually I was hoping for late q3, but looks more like late q4 instead. How ironic, especially with how the r600 launch went. Makes you wonder if we have another r600 on our hands because nvidia's been so quiet about it compared to how they were about g80 (especially now that ati's giving out info it really makes you wonder) not to mention trying to debug a core this complex must be killer (think about it, the current r600 is a cut down of the original, the original was too complex and was full of bugs)

[Barys]

I'd like to go ahead and confirm that I've been hearing similar from my sources as bench. Big surprise, huh bench? We always seem to have the same info on nvidia parts.

Anyway, I'm hearing a late may-early june release, just thought I'd add that.

BTW, for those who forget..one of the number one reason for high-end shortages in the past has been because companies go for high-end ram near it's initial launch, and stuck with not being able to get said ram. Remember the 6800Ultra and X800XT/PE? Both had countless supply problems because of going for what was the best ram at the time.

Sticking with GDDR3 means ram will be in high supply, meaning NVidia can pump out as many cards as they can make gpus.

Doesn`t seem strange you have said that Benchzowner and you have similar info about GT200 and then he is saying GT200=Q4 release and you are saying it`s about May-June So who is right?

Unfortunately my info will depress you.
june = G92b ( 9900XXXs )
GT200 = q4 08 more than likely.

So it means FUdos info about GT200-300 tapeed out is BS? If it`s truth then we should see GT200 in 2-3 months not later.

The second thing - is there ANY possibility (if these specs are true of course) to push 120TMU, 240SP, 32 ROPs and 512-MC in "only" 1B trannies? It`s only about 250M more than G92 which is supposed to have about 2x less TMU, 2x less SP, 2x lessROPs etc. With these rumoured specs (i hope they are really true) GT200 should have about 1.2B trannies at least.


Hmm i can only say i really wait for NVIDIAs "next BIG thing" (worthy G80 successor) and i hope that after crappy GF9 series they will release very powerful GPU (single die).

PS. You have said performance is better than 2xG92 (GF8800GTS) in many cases? But it`s only rumoured performance or someone already has working samples of it and this performance level is what this GPU can really do?

[BenchZowner]

then he is saying GT200=Q4 release and you are saying it`s about May-June So who is right?

Both timeframes can be wrong as well.
Release dates ain't firm, and of course depend on various things.

So it means FUdos info about GT200-300 tapeed out is BS? If it`s truth then we should see GT200 in 2-3 months not later.

The fact that it has been taped out ( if it's true ), doesn't mean that it's going to be released soon.
You can always keep the goodies for the future.
Releasing it now, can kill the prices & sales of the G92 and especially the 9800GX2 & the G92b.

The second thing - is there ANY possibility (if these specs are true of course) to push 120TMU, 240SP, 32 ROPs and 512-MC in "only" 1B trannies? It`s only about 250M more than G92 which is supposed to have about 2x less TMU, 2x less SP, 2x lessROPs etc. With these rumoured specs (i hope they are really true) GT200 should have about 1.2B trannies at least.

Check my posts... people & rumors on the web talk about 1bil, but I'm talking about 1.3b [ info from my source ]

PS. You have said performance is better than 2xG92 (GF8800GTS) in many cases? But it`s only rumoured performance or someone already has working samples of it and this performance level is what this GPU can really do?

That's my performance estimate, if the specs I've posted are true

[y2kbos]

That's my performance estimate, if the specs I've posted are true

Bit disappointing then isnt it?

9800gx2 or 9900gx2 should be very affordable by the time the GT-200 launches with similar performance..

Nvidia have made sli too good lol, killing there high end single card...

[BenchZowner]

Bit disappointing then isnt it?

9800gx2 or 9900gx2 should be very affordable by the time the GT-200 launches with similar performance..

Nvidia have made sli too good lol, killing there high end single card...

Well, I always drop my estimates down, underneath the average.
The performance can be higher than that... depends on various factors.
It will be faster than a 9800GX2/9900GX2... but by how much ? 15% average or so (?)

The interesting part will be the high resolution & details with AA/AF, where the card will have a greater advantage over resolutions lower than 180x1050 ( when compared to the GX2 )

[AliG]

Both timeframes can be wrong as well.
Release dates ain't firm, and of course depend on various things.



The fact that it has been taped out ( if it's true ), doesn't mean that it's going to be released soon.
You can always keep the goodies for the future.
Releasing it now, can kill the prices & sales of the G92 and especially the 9800GX2 & the G92b.



Check my posts... people & rumors on the web talk about 1bil, but I'm talking about 1.3b [ info from my source ]



That's my performance estimate, if the specs I've posted are true

looks like we're on the same page for most things. Taped out doesn't necessarily mean anything at this point since nvidia has time to kill. ATI pushed the rv670 launch extremely hard since they were losing tons of market share and forced nvidia to rush their g92 launch to combat it (imo, I don't think g92 was ever supposed to become part of geforce 8, the g80 design was doing so well I actually think nvidia was planning on releasing g92 as the geforce 9 high end and performance, but was worried about the rv670 head stomping their performance segment). Just look at my last post, you could make rational arguments about whether it was taped out or not very easily

But I don't think g92b will ever be a high end segment. The only reason I can see g92 being shrunken down to 55nm is to improve yields to combat the rv770 in the performance section. That and I think the gt200 dies will be so large there won't be any cut downs for the performance section, all of them will have to be put towards high end for nvidia to make money off it. So if that's true, gt200 could never hurt g92b's sales since there's going to be such a large price difference between performance and high end. Think about it, I mean its not like the 9800gx2 is killing the 8800gts 512mb sales (its taking some, but a very small percentage)


As for 1.3 billion trannies, honestly I couldn't care if it had 5 trillian trannies so long it performed well in games. The r600 proved that numbers on paper mean nothing until they are proven to do so. While g80 is a lot more efficient in current games than the r600, who knows what either the r700 or gt200 will do (especially if the new architecture rumor is true, gt200 doesn't necessarily have to perform like a 240 shader version of g80 as the pipeline can be significantly different; not to mention a lot of the r600's major bottlenecks are finally being removed which can make all the difference).


As for your performance estimate, I have to agree. Without a doubt, the sli gts 512mb will be nothing in comparison to gt200 if gt scales like g80 because we're talking a single die here vs sli's poor scaling (though it has greatly improved, xfire still scales better imo and even that can't do 100% scaling in 10% of the time)

[AliG]

I can't really say for sure whether its taped out or not, but reason could support either answer. For one just because the design is taped out doesn't mean its 100% bug free, just means the design is pretty much finalized and the debugging is in its final stages (especially driver debugging). But also look at the r600's launch, the original r600 rumors talked about a monster card (but all we got was monster power consumption lolz) but ati kept quiet and it turned out to demolish everything previously released by ati, but still no performance crown stealer. So far gt200 is following the same steps, extremely complex and large die, 512 bit memory, rumors of monster. Thing is, usually nvidia gives out a lot more information than this on projects they are confident on. So either they are worried about the rv770 stealing the performance section where the largest sales have been taking place recently (most profit:costs comes from high end though) or something's wrong with gt200 if its not taped out yet. But don't forget, g80 was in volume production months before it launched, so that could mean there are driver issues or samples are still being given to coders, or simply yields are crap so nvidia is stockpiling them. You never know until we actually see a real preview.

As I said, if you think about it rationally based off past events, you could support either arguement.

[BenchZowner]

@AliG

If the specs are true, they can easily disable some clusters, and get the GT & GTS out.
Even with 6 clusters you have a beast card to show.
144 SPs
72 TFUs
72 TMUs

The r600 proved that numbers on paper mean nothing until they are proven to do so

Well... the numbers were clear for people to figure that it won't be any good in gaming ( especially with AA & AF enabled ) since... it only had 16 TFUs, and lacked hardware AA resolve.

It's like designing a RWD Mitsubishi Lancer for off-road races against a 4WD Impreza

[AliG]

true, but still, I'm talking about die costs due to the high size not about the performance. Unless if you can just chop off parts of the die, something tells me none of the gt200 dies are going to the performance section. Perhaps a gts to go with the gtx (or whatever the new naming scheme will be), but nothing else

[BenchZowner]

true, but still, I'm talking about die costs due to the high size not about the performance. Unless if you can just chop off parts of the die, something tells me none of the gt200 dies are going to the performance section. Perhaps a gts to go with the gtx (or whatever the new naming scheme will be), but nothing else

A GTS is planned.
GT ? If ATi comes up with something worth the effort

[JohnZS]

I'm happy with my G92 GTS 512MB, it plays all my games at a good framerate and Image Quality on my 20" NEC LCD.
Well...all except for Crysis, but I'm of the opinion that Crysis is a bit like Farcry in the sense it was built to last the test of time longer than other games released at the same time.
I might upgrade to the 9900 series IF they let me crank up the details @ 1680*1050, otherwise I shall wait until the next big thing. (ATi or nVidia).
John

[AliG]

I realize a gts is planned as they'll have to do something with the chips that don't meet the bin, but unless if g92b will be a mainstream product, I'm almost certain it'll be the gt or whatever the performance models are (who knows, maybe we'll see a x700 from now on).

As for ati coming up with something that compete, I'm sure of that, the biggest concern I have for ati is that unless games come out designed specifically for multi gpu setups (which to be honest would be quite nice to have a few of those every year as it would mean we'd have a few new games like crysis every year as the coders would put in a lot more detail than the regular games), the r700 will be a decent high end, but won't be powerful enough to justify getting it over the rv770 due to budget constraints or gt200 for pure performance

[adamsleath]

gt200 by christmas. but that might be optimistic.

seeing as how there are so many cf and sli options around right now, but crippled by vram, bus and shaders.

[Zytek_Fan]

Please come out within 90 days of April 15
(I get my video card on the 15th)

[GoldenTiger]

My fund is already being saved for GT200 . One shall be mine ! Hope for sooner release...

[STaRGaZeR]

My fund is already being saved for GT200 . One shall be mine ! Hope for sooner release...

November, at least

[Nicksterr]

I hope it has a cooler that exhausts air out the back of the case like previous trends. Oh yeah...hopefully 40%+ performance in games compared to the 9800gtx.

[largon]

Would be nice if nV finally put some sense in PCB layouts - compared to Radeons their boards are horrible; parallel PWMs are scattered around the board, unified GDDR vDD & vDDQ, using old/obsolete or non-documented "top secret" phase controllers - that's what I'm talking about. It would be nice to see them achieve the level of masterpieces ATi has pulled recently - such as the OEM R600 board. This 210mm long beauty is over an inch shorter than 8800GTS but sports far superior 6-phase vGPU PWM with everything in one single clean row, dedicated vDD and vDDQ, double-width video RAM bus, no huge arrays of caps and no dead space.



There's only 3 things I would change in this board to make it perfect:
- R600 GPU
- R600 GPU
- R600 GPU

[Bo_Fox]

Would be nice if nV finally put some sense in PCB layouts - compared to Radeons their boards are horrible; parallel PWMs are scattered around the board, unified GDDR vDD & vDDQ, using old/obsolete or non-documented "top secret" phase controllers - that's what I'm talking about. It would be nice to see them achieve the level of masterpieces ATi has pulled recently - such as the OEM R600 board. This 210mm long beauty is over an inch shorter than 8800GTS but sports far superior 6-phase vGPU PWM with everything in one single clean row, dedicated vDD and vDDQ, double-width video RAM bus, no huge arrays of caps and no dead space.



There's only 3 things I would change in this board to make it perfect:
- R600 GPU
- R600 GPU
- R600 GPU

That is rather interesting.. yeah, and one R770 core by itself should be equal to like 3 R600's. So, two R770's would equal SIX R600 chips, and that is what we are going to see in May/June.

[STaRGaZeR]

That is rather interesting.. yeah, and one R770 core by itself should be equal to like 3 R600's. So, two R770's would equal SIX R600 chips, and that is what we are going to see in May/June.

RV770 is going to be 50% faster on average than RV670, and RV670 has roughly the same perfomance of the R600. How did you get to that conclusion?

[whocaresbg]

RV770 is going to be 50% faster on average than RV670, and RV670 has roughly the same perfomance of the R600. How did you get to that conclusion?

But with specs "800SP/32TMUs" up to 100 % .

[STaRGaZeR]

It's beautiful, eh?

[w0mbat]

No, one RV770 core wont be any near to 3 R600 cores.

[Bo_Fox]

No, one RV770 core wont be any near to 3 R600 cores.

OOPSIE-DAISIE!!! My bad! I do not know why I said that--I must have been thinking that one R700 board with two RV770 cores would be equal to 3 R600 cores at the very least (since one RV770 is 50% faster than one R600), and then later did the math completely wrong!

Oh well, nobody's perfect, right? We aint as reliable as calculators (yeah, I need to use a calculator when adding 2+2).