Sign in with
Sign up | Sign in
Your question

Nvidia claims G92 (Geforce 9800) will be a 1 Teraflop beast

Last response: in Graphics & Displays
Share
May 24, 2007 11:29:07 AM

If that is accurate I'd be curious to see what they are releasing to hit that kind of performance. I would have expected Nvidia's next release to be more of an evolution maybe on 65nm.
Anonymous
May 24, 2007 11:30:39 AM

what do you guys think ATi will bring up against this. i think R700. I don't think R650 will be able to even come close
Related resources
May 24, 2007 11:31:35 AM

bigger question is, how much will that cost!?
Anonymous
May 24, 2007 11:35:49 AM

i would say same as much as highend card. 575$ to 650$
May 24, 2007 12:18:15 PM

If a card 3x as powerful as my current 8800 GTX is going to be released in december, then it would be enough to convince me against just SLI'ing my card with another GTX in order to get the performance I want :D . Hope they do it. Although I suspect theres a healthy dose of hype and marketing there.
May 24, 2007 12:18:50 PM

Looks like it that i got another 6 months more waiting to do. Not going to buy any overpriced 8800 series products considering they're half-a-year-old and will be obsolete in a year. And the fact that R600 failed to live up to its expectations. Well, my current X800 XT will be around to serve me for a little bit longer...
May 24, 2007 12:18:55 PM

I suspect multiple cores.

This would seem like the only reasonable way to go from 330GFlops to 1000GFlops. I can't see them making the processor 3x faster. I could envision 50% faster and then doubling the cores.

The author of the article has little vision.
Off-loaded physics and super real graphics.

In many cases you need to put the horse before the cart.
If there is not any hardware that can play a game, nobody will buy it waiting for the hardware.

People will, however, buy the fastest hardware even if its faster than they need.

I think he is missing the point that this will be a boutique type item.
People with the cash will kill to have this card and NVIDIA will turn a nice profit on this high ticket item.
May 24, 2007 12:29:22 PM

Quote:
guys check this link out.

http://theinquirer.net/default.aspx?article=39829

now we wait for this card until christmas

How dumb is Nvidia's PR? I would have waited till crysis, around august, help sold alot of the 8800 GPU's before releasing this info.
May 24, 2007 12:32:31 PM

I want to see the ATI FanBoys's faces when this monster releases! :lol: 
Time for a
May 24, 2007 12:34:57 PM

I agree. next-gen top GPUs are around 50% faster than the previous one. They don't even come near doubling the performance. Saying that G92 is gonna be x3 faster than G80 sounds like crap PR for kids to me.

Unless.....they are going with something REALLY new, that means, multicore GPUs. Its the only way I could even start to believe that statement. There has been rumors about multicore GPUs in relation to R700. So that can actually be the reason why.

Lets wait and see. Right now my 8800 GTS 320MB OC is running at 1680x1050 just fine with Vista.
a b U Graphics card
May 24, 2007 12:38:04 PM

Dont forget a hefty amount of transistors going to 65nm. Also I suspect therell be 1 gig of really fast ram on it, maybe clocked to 2.8. So with a faster clock you could do it without going multi core
a b U Graphics card
May 24, 2007 12:39:24 PM

this is the same as when the 6 series came out followed by the 7 series, ati barely overcame the 6800, and finally when it did the 7 series came around.

ATi and AMD need another break like the FX and Prescott era (error)
a b U Graphics card
May 24, 2007 12:42:29 PM

Quote:
I want to see the ATI FanBoys's faces when this monster releases! :lol: 
Time for a


i wonder, could they make a 3 slot cooler? :oops:  and two of these babys in SLI would exceed the 1 teraflop by far
May 24, 2007 12:42:36 PM

I hope this does in fact happen, though I highly doubt it will.


And what's with the last paragraph in the article? Is the Inq. unaware that games tend to get more demanding as time goes on?
May 24, 2007 12:48:35 PM

Still doubt it. Normally, you have to wait two generations to actually double the performance of a GPU. Even with the possibilities that a 65nm process opens, I still doubt that it will triple the performance of a G80.

No, it will have to be something really new...like a multicore GPU. I still remember quotes from ATI like "people don't undersatnd the r600" and how excited that made people (including me, who was getting a new GPU). And what did we get? Performance ususally bellow a 8800 GTX. Its actually a good buy if you take into account that it costs the same as a GTS. Bottom line: a dissapointment.

So there's only two ways to go here. Either its something really revolutionary like multicore GPUs or this is just crap and we'll be lucky if the shift to 65nm even gives us double performance.
May 24, 2007 12:50:19 PM

If this is true, my 8800GTX may be out the door by the end of the year. :lol: 
May 24, 2007 12:51:09 PM

Quote:
I hope this does in fact happen, though I highly doubt it will.


And what's with the last paragraph in the article? Is the Inq. unaware that games tend to get more demanding as time goes on?


Yep, that commentary by TI is really ignorant. There's no such thing as a too fast GPU. Just buy it and 6 months latter that GPU will be dragging running the that new uber-FPS at 1920x1080 with AA and AF maxed out.
May 24, 2007 12:52:51 PM

Quote:
If this is true, my 8800GTX may be out the door by the end of the year. :lol: 


Exactely, except that in my case I just spent 300 bucks in an overclocked GTS 320MB, which will make buying a 9800 GTS even more sweet. 8)
May 24, 2007 12:53:20 PM

I wouldn't mind seeing a $250 8800 GTX. I would be the first one out of the door getting them!

There's some games out there that even 8800 GTX has trouble like NWN2 (which is more CPU and GPU intensive than Oblivion) so I wouldn't mind seeing a "better" card.
a b U Graphics card
May 24, 2007 12:54:14 PM

Youre right. It could be hype, or muli or what I said adding a few things, going wider with a 512 bit would help. But in the end, itll probably be close to a T-Flop and be using multis
May 24, 2007 1:03:27 PM

Quote:
If this is true, my 8800GTX may be out the door by the end of the year. :lol: 


Exactely, except that in my case I just spent 300 bucks in an overclocked GTS 320MB, which will make buying a 9800 GTS even more sweet. 8)I only spent $550 on my 8800GTX... :lol: 
May 24, 2007 1:07:29 PM

Quote:
guys check this link out.

http://theinquirer.net/default.aspx?article=39829

now we wait for this card until christmas

How dumb is Nvidia's PR? I would have waited till crysis, around august, help sold alot of the 8800 GPU's before releasing this info.

Or they could just bring the release forward to a few weeks before Crysis and sell an absolute barrel load. On a smaller fab they get more to each wafer so thye'd be cheaper to produce than the existing G80's = more profit per card. I'd buy one (provided price isn't more than £350-400).
May 24, 2007 1:11:38 PM

too bad ati have already had a 9800, it wasn't that bad for its time.. now it doesn't quite compare...
May 24, 2007 1:27:08 PM

I can see it know, some smart eBay guy quadruples the price of his old 9800GPUs with NVIDIA releases theirs. Somebody will buy the old stuff thinking they are getting the new stuff :>
a c 86 U Graphics card
a b Î Nvidia
May 24, 2007 1:52:52 PM

What happened to the 8900-series? Are they going to skip straight to the 9800 or is that just a typo? Anyways 3x faster sounds just too unbelievable, too good to be true 8O :p 
May 24, 2007 1:59:56 PM

the g90 will be a single gpu 9800 so the g92 will probably be a 9800x2 dual gpu, no single card will be hitting a terraflop this side of 2009/10 i think the news sites forgot to meantion maybe its a dual gpu board cos i think they wanted to get you excited r600 gpu is almost 500 gflops and the g80 gpu is almost 350 so if you think nvidia can squeeze another 650 gflops out of a single gpu then your crazy it has to be a dual gpu.
May 24, 2007 2:06:55 PM

The 8900 is coming, but it will just be a faster version of the 8800.

The 9800 is a new series.
May 24, 2007 2:17:21 PM

Quote:
guys check this link out.

http://theinquirer.net/default.aspx?article=39829

now we wait for this card until christmas


According to TG daily:

In regards to CUDA and the 8800GTX "The technology is compatible with 8800-series graphics cards , which are estimated to achieve a peak performance of about 520 GFlops, allowing consumers to build a teraflop system using Nvidia’s SLI bridge."

So that would make the 9800 just like 8800GTX SLI not 3 times as fast.

theinquirer...., whatever
May 24, 2007 2:57:33 PM

Sort of off subject, but on the 680i SLI boards do you have to already be in SLI to use the 8x slot for Physics? and are the NVidia mobos 45nm(bearlake) compatible?

I think the power consumption on these 9800 cards are gonna kill us all, OR they will have to be super duper fast AND efficient...yeah right though.

The reason I asked about the physics slot is because I wonder what are we gonna do with our "old" 8800's? Slamming them into physics seams a little like a waste but better than throwing them away or selling them for 1/4 of their original price. The reason I ask about the 45nm compatibility is I wonder how much current CPU's will bottle neck the hell out of something like a 9800.

I wonder what happens after the 9*** series will they go to a new name instead of Geforce or will they add and X to the front of the name and go back to the beginning or what?
a c 86 U Graphics card
a b Î Nvidia
May 24, 2007 3:36:09 PM

Quote:
The 8900 is coming, but it will just be a faster version of the 8800.

The 9800 is a new series.

ah, ok :) 
a b U Graphics card
a b Î Nvidia
May 24, 2007 5:13:28 PM

Quote:
According to TG daily:

In regards to CUDA and the 8800GTX "The technology is compatible with 8800-series graphics cards , which are estimated to achieve a peak performance of about 520 GFlops, allowing consumers to build a teraflop system using Nvidia’s SLI bridge."

So that would make the 9800 just like 8800GTX SLI not 3 times as fast.

theinquirer...., whatever


Remember that's theoretical performance not actual performance.
nV has never reached that 520GFlops even through CUDA because it can't exploit the extra MUL with any efficiency. It launched at 346Gflops, and only recently reached the mid 400s with new drivers, still well below the 520 quoted at launch. The R580 was accessing almost all of it's 380GFlops through CTM and the R600 acesses close to all it's 480 in the XT.

So really the question becomes whether talking about the theoretical number or the actual number, if it's theoretically close to 1TFlop, and has the same missing MUL issue, then it's going to be closer to the 800-900 range than 900-1000GFlops and like you said just 2 times the difference. If it's actual Gflops, then it's close to 3X the G80 launch and 2.2X current levels.

Considering the speed boost for both architectures, that's pretty much what we all expected, especially if the "deliver almost a teraflop of computing performance" is rounding up significantly.
About 800+Gflops theoretical was what most people expected shoul nV go to 160SPUs and a faster core on 65nm (heck just the SPUs gets you near 700Gflops), so it's not that much of a surprise since it would only need to run at about 800-825mhz, which is doable on the 80nm R600, let alone these new refresshed 65nm chips both will have.

Could be interesting if the architecture dramatically changes, but right now it just looks like the mathematical extension of what we already knew.
Anonymous
May 24, 2007 6:00:02 PM

yeah from what i read on some other places is that G90 and R700 are multi core GPU.
May 24, 2007 6:00:07 PM

Quote:
The R580 was accessing almost all of it's 380GFlops through CTM and the R600 acesses close to all it's 480 in the XT.

Oh and that's why the HD2900XT sucks?
May 24, 2007 6:03:14 PM

3 times the performance of the 8800 GTX? :)  teeheeee, i doubt a single gpu core will do that till 2010 or so.

i think it will be 2 8800 Ultras on one board. dual-GPU. Or it might be a speed bump with two gpu's on one board or dual boards like the 7950's.
Anonymous
May 24, 2007 6:24:42 PM

man why are you always so critical about Nvidia. we know your on ATi side
May 24, 2007 6:40:35 PM

Quote:

Could be interesting if the architecture dramatically changes, but right now it just looks like the mathematical extension of what we already knew.


Usually you are so much more subtle with you're ATI bias. If Nvidia does deliver that kind of performance in 2007 it is not what we all expected anyhow. 2.2 or 3 times is an incredible leap for 1 year.
Either way I'm glad to see that ATI's slip ups are not slowing Nvidia down. I don't care which company delivers more performance as long as they keep bringing it. At this rate maybe only a few more years before we have photo realistic visuals in games.
May 24, 2007 7:00:50 PM

I'm right on track with you. I don't care who makes what as long as it's good. The G92 sounds like a real monster, and although it's waaaaaay too early to know for sure, I think it's a huge leap for just a years worth of research. Now ATI has roughly 6 months to nearly triple it's performance of it's not even released products along with the ones they just put out.
May 24, 2007 7:12:04 PM

Quote:
Remember that's theoretical performance not actual performance.
nV has never reached that 520GFlops even through CUDA because it can't exploit the extra MUL with any efficiency. It launched at 346Gflops, and only recently reached the mid 400s with new drivers, still well below the 520 quoted at launch. The R580 was accessing almost all of it's 380GFlops through CTM and the R600 acesses close to all it's 480 in the XT.

So really the question becomes whether talking about the theoretical number or the actual number, if it's theoretically close to 1TFlop, and has the same missing MUL issue, then it's going to be closer to the 800-900 range than 900-1000GFlops and like you said just 2 times the difference. If it's actual Gflops, then it's close to 3X the G80 launch and 2.2X current levels.

Considering the speed boost for both architectures, that's pretty much what we all expected, especially if the "deliver almost a teraflop of computing performance" is rounding up significantly.
About 800+Gflops theoretical was what most people expected shoul nV go to 160SPUs and a faster core on 65nm (heck just the SPUs gets you near 700Gflops), so it's not that much of a surprise since it would only need to run at about 800-825mhz, which is doable on the 80nm R600, let alone these new refresshed 65nm chips both will have.

Could be interesting if the architecture dramatically changes, but right now it just looks like the mathematical extension of what we already knew.

You could spend less time and just write something like "nVIDIA suck!". :wink:
May 24, 2007 7:32:54 PM

I'm sorry? 1000 GIGAFLOPS?

i like computers

i like being hopefull

i dont like being stupid.

theres no way they can treble the power of the card. They just about doubled it from 7900gtx to 8800gtx. i think the same as some previous guy that multicore is the way forward, or a significant die shrink.
a b U Graphics card
a b Î Nvidia
May 24, 2007 7:33:06 PM

Quote:
The R580 was accessing almost all of it's 380GFlops through CTM and the R600 acesses close to all it's 480 in the XT.

Oh and that's why the HD2900XT sucks?

It's because GFlops doesn't mean much for graphics because it's not raw power that matters most.

Remember it's not the HD2900's shader power that's holding it back it's the back end, which is not part of the GFlops calculation.
May 24, 2007 7:33:19 PM

You know, I wanted to get a Dx10 card for this summer. But it just came back to me that last time I got a first gen DxX card, I was disappoint when the real games of the said DxX came out. For me it was Radeon 9800Pro and Dx9, and the said game was Oblivion, and even, to a lesser extend, FEAR.

That means that I'll have to enjoy Crysis in Dx9 this coming summer. But it also means that I'll be able to play true Dx10 games, that'll come out only at the end of 2008 at best, with the same VPU. That puts me in G92 (90?) and R700 territory of cards.

I... gotta......... resist.... the ... urge to.... upgrade.... :twisted: It'll be tough, but I'll do my best. :wink:
a b U Graphics card
a b Î Nvidia
May 24, 2007 7:49:10 PM

Quote:
man why are you always so critical about Nvidia. we know your on ATi side


You might see it as critical, I see it as factual, the only thing I'm critical of is the the way the data is being represented.

Like I said last time, if you have aproblem with the facts, then put forth something supportable that is contrary to what I said.

You and other want to spin the information into something as in A or B sucks, I don't see anywhere where I said either solution sucks, unlike others in this thread and the other thread you spammed in.

People who know better, know that last time the number was the high theoretical, not actual output, and staying true to form this is likely the theoretical max again, which I showed is in line with what we expected. The 3X number is the previous actual output (which has changed) and the current would be more than 2X actual and under 2X current theoretical of the GTX.

If you have problems with my math, point out the area where you have better numbers, but back it up with more than the missing MUL PR number.

1 TFlop is still a big number, but unless we know how they arrive at that number it doesn't mean much on it's own.
a b U Graphics card
a b Î Nvidia
May 24, 2007 7:57:43 PM

Quote:
You could spend less time and just write something like "nVIDIA suck!". :wink:


I leave that statement to people who don't understand the subject.

Warez and Hangman both use the BS PR numbers that have never been achieved.

If nV manages to reach the 1TFlop PR number with the same level of efficiency that it currently reaches on the G80, then that would mean actual output Should be in the high 800-900 range, which is still impressive. However for anyone to criticize the InQ's numbers while using the inflated 520Gflops number they really need to edjucate themselves better first.

Read my post and then you too can be edjucated. :twisted:
a b U Graphics card
a b Î Nvidia
May 24, 2007 8:03:24 PM

Uh oh, a #1 fan? 8O

Need restraining order. :twisted:
May 24, 2007 8:05:14 PM

That makes two of us. Except i pleasure myself while he makes his post, since we live so close to each other :twisted:
May 24, 2007 8:18:39 PM

OK dude!
Guess you need a or two!
Hey! I`m of your fan too, but i don't know my place! Maybe 2nd or 3rd. I guess we need to make thread named "Who is the biggest TGGA fan?" :D 
May 24, 2007 8:38:03 PM

Quote:
It's because GFlops doesn't mean much for graphics because it's not raw power that matters most.

Remember it's not the HD2900's shader power that's holding it back it's the back end, which is not part of the GFlops calculation.



Then why point out that the GTX can't do as many Gigaflops as the HD2900XT. What you previously stated would be irrelevant. You just proved yourself wrong.
May 24, 2007 8:49:36 PM

His point was that even if a gfx card can do massive gigaflops it doesnt mean it'll perform accordningly, case and point, the r600 has massive gigaflops but the performance doesnt match the g80. So really he's favoring nvidia here, and as such, he cannot be called an ATI fan.

Also, you'd have an easier time saying he favors ati since its canadian than you will with any other method. Sheesh.

Maybe someday we could watch...together :twisted:
May 24, 2007 8:55:25 PM

I guess my point is that when someone makes the statement one way to prove their point and then reverses their actions that actually ends up disproving their first point it makes them look very ignorant. Also he's an ATI fanboy.

Although in the end I understand that raw power doesn't mean anything if you can't get the power to do anything. Hense a P4@3.8 isn't much compared to a C2D@1.8.
!