Nvidia reveals Specifications of GT300

[thilanliyan]

more info:
512-bit mem interface
http://www.brightsideofnews.co...cated-controllers.aspx

delayed to 2010 (caution this article by Charlie) :
http://www.theinquirer.net/inq...t300-delayed-till-2010

 

[Tempered81]

Originally posted by: thilan29
more info:
512-bit mem interface
http://www.brightsideofnews.co...cated-controllers.aspx

delayed to 2010 (caution this article by Charlie) :
http://www.theinquirer.net/inq...t300-delayed-till-2010

Here's what I think: With 300GB/sec memory bandwidth, 512 MIMD redesigned cores on 512bit interface @ 40nm node, 1/2 node whatever....

Nvidia is pissed that ATi's tried and true detailed graphics roadmap has been such a success. With as much underestimating as Nv did regarding ati's plan, they only rehashed and re-released crap, and really nothing much "revolutionary" has come from them since G80. There was huge talk of G100, gt200, g200 when it came out, and we had the intern girl inside XFX labs giving us hints and tidbits. 3dmark06 scores and such. It was just another crappy attempt at new-sku creation.

Now, someone green is mad. Ati's plan went r600->r700>r800>r900, all the way to R10k is road mapped according to wiki. I'm not anticipating much more than a 1.2-1.4x increase in overall raw performance from the red camp. There was quite a jump from hd2000>hd3000>hd4000. IMO, HD4870> HD5870 should net about a 1.4x increase. However, from the sound of things, nvidia is looking for a 2.5x increase in performance jumping to the next generation from GTX280 > GTX380 (or whatever they name it).

GT200-GT300 performance increase percentage is probably going to be greater than rv770-rv870. considering gt200 is faster than rv770 as it is, then you can expect a gtx380 to smash an hd5870. Hopefully with 1200 shaders x2, that R800 will be able to do something about the gtx380. Now timing is everything. October 09-Jan 10 release date window for Nvidia kinda sucks for them - but here's to hoping they get something right, and something with some performance out on the market. If 5870 comes out in June-July and 5870x2 hits the shelves in August-October, then Nvidia really has no competition for that, seeing as how HD4890 OC competes with GTX285, and the soon to be released HD4890x2 will take the performance crown from the GTX295 (based on 4890 crossfire reviews). Now when 5870 hits, there will be nothing for almost 6 months from the green camp, minus a single pcb gtx295 with faster memory. This could be a sweet move and a milestone for ati, doing the first DX11 chip, and controlling high-end market (assuming hd4890x2 is released soon) for 6-months using their small-die mainstream chip approach. That puts NV in an even worse position.

I still expect GT300 to smash the competition. Too bad its going to be Christmas time before we see that. It should be everything that the gloating jensung thought glorified gt200 would be (as many of the anticipating customers expected as well). Nvidia is pissed, and rightfully so because they sat on their butts for so long re releasing gpus. Ati has a long range goal and they are sticking to it. It is giving amazing results so far, but I expect GT300 to be an incredible performer.

So, IMO:
HD5870 - 1.4x times as fast as hd4870 - July 4th 2009
HD5870 X2 - 1.4-1.5x times as fast as hd4870x2 - August 30th 2009
GTX380 - almost as fast as hd5870x2 - Dec 10th 2009.

 

[Wreckage]

Yeah it's looking like the GT300 will be unstoppable, much the same way as the GT200 and 8800 before it.

 

[sankhya]

i think the GT300 will perform like an 295GX2 (if not less)

but the real deal would be the DX11 support it would feature

but then again who really knows till its released later this year(( ..ffssssss...........6 more months of waiting..........cant wait to buy it))))!!!!!!!!!!!!!!!!?

 

[cbn]

http://www.fudzilla.com/index....view&id=13551&Itemid=1

That is what Fudzilla has to say about GT300.

http://www.fudzilla.com/index....iew&id=13550&Itemid=34

Here is what Fudzilla is saying about RV870? (Maybe ATI is changing plans based on the rumors of a 512 core 512-bit DDR5 GT300)

 

[MrK6]

Wow, it's all still just speculation. I'm hoping one of these delivers adequate gaming performance as a single card at 2560x1600.

 

[Blazer7]

Maybe not at such a high res but @1920x1200 and below that?s another story. A single card will most likely be more than enough for most people out there and both nV and ATI know this. Given that less people will resort to CF/SLI setups I wouldn?t be surprised if these cards remain (artificially) expensive for a long time as both companies may see this as lost sales. The only thing I can think of that can ?save? the CF/SLI market is new and more demanding games and I don?t believe that we are gonna see many of those in the near future.

 

[MarcVenice]

Originally posted by: Just learning
http://www.fudzilla.com/index....view&id=13551&Itemid=1

That is what Fudzilla has to say about GT300.

http://www.fudzilla.com/index....iew&id=13550&Itemid=34

Here is what Fudzilla is saying about RV870? (Maybe ATI is changing plans based on the rumors of a 512 core 512-bit DDR5 GT300)

Fudzilla is thinking out loud. And they aren't very smart judging from what they think. How can RV870 not be a DX11 part from the get go? It has to be, because if it launches in q3/q4 as a dx10.1 part, it will take another year to get out a DX11 part. That would be disastrous and it's just plain unthinkable. The old RV870 didn't get canned at all. Also, afaik it's very hard to adjust shadercounts and w/e on the fly. They have build in measures to improve yields, which they might abuse though, sacrificing yields, to gain performance. But that's all the playing room they've got. And they can of course adjust clocks up or downwards.

It's all a whole lot of speculation. Only thing we could bet on is that both Nvidia and AMD will launch later then expected due to yield problems on the 40nm process.

 

[Keysplayr]

Originally posted by: MarcVenice

Originally posted by: Just learning
http://www.fudzilla.com/index....view&id=13551&Itemid=1

That is what Fudzilla has to say about GT300.

http://www.fudzilla.com/index....iew&id=13550&Itemid=34

Here is what Fudzilla is saying about RV870? (Maybe ATI is changing plans based on the rumors of a 512 core 512-bit DDR5 GT300)

Fudzilla is thinking out loud. And they aren't very smart judging from what they think. How can RV870 not be a DX11 part from the get go? It has to be, because if it launches in q3/q4 as a dx10.1 part, it will take another year to get out a DX11 part. That would be disastrous and it's just plain unthinkable. The old RV870 didn't get canned at all. Also, afaik it's very hard to adjust shadercounts and w/e on the fly. They have build in measures to improve yields, and they might abuse though, sacrificing yields, to gain performance. But that's all the playing room they've got. And they can of course adjust clocks up or downwards.

It's all a whole lot of speculation. Only thing we could bet on is that both Nvidia and AMD will launch later then expected due to yield problems on the 40nm process.

Not so far fetched. It is very possible ATI had more than one design in progress and decided on one to battle with. Now ATI may have second thoughts about a 1200 shader (rumor) R870 and might consider that insufficient to combat a 512 core 512 bit GDDR5 GT300. There is enough time to change to the other design, tape out, and have a product by Christmas. They will probably need a GPU with no less than 2000 shaders, 512 bit GDDR5 ( if they are staying with the same architecture ) to compete with a GT300. IMHO.

FUD's DX11 comment was a bit stupid though. All new cards will be DX11.

 

[MarcVenice]

They might have several designs. Saying they need a 2000sp gpu to combat Nvidia is a bit ridiculous (and a lot of speculation) though. They don't WANT to compete with nvidia when it comes to a single gpu. They want a smaller, less powerfull cpu, that gives 75/85% of Nvidia's best gpu's performance, and boom, they double it with a HD 5870X2. That's AMD's tactic, which is no secret at all. No, I'm pretty sure whatever X2 part AMD launches, will handily beat whatever single gpu Nvidia launches. And that's what AMD is aiming for, not to beat or equal Nvidia with a single gpu.

 

[Blazer7]

Marc is right. When it comes to single GPU cards AMD doesn?t have to best nV. Right now they do have the advantage to be the first in the market with a DX11 part and if predictions are correct nV won?t have anything to throw at them for at least 5-6 months. By the time the GT300 is released AMD will have sorted out whatever early production problems they have and will probably be in a position to counter attack with some other products like oced and x2 cards, or more likely start another price war. That will hurt nV even more.

 

[Keysplayr]

Originally posted by: MarcVenice
They might have several designs. Saying they need a 2000sp gpu to combat Nvidia is a bit ridiculous (and a lot of speculation) though. They don't WANT to compete with nvidia when it comes to a single gpu. They want a smaller, less powerfull cpu, that gives 75/85% of Nvidia's best gpu's performance, and boom, they double it with a HD 5870X2. That's AMD's tactic, which is no secret at all. No, I'm pretty sure whatever X2 part AMD launches, will handily beat whatever single gpu Nvidia launches. And that's what AMD is aiming for, not to beat or equal Nvidia with a single gpu.

What's ridiculous about it? An 800 shader ATI GPU still cannot best a 240 shader Nvidia GPU without some serious o/c'ing. It doesn't stand against reason that NV's new shaders won't be stronger/better than current GT200 shaders. Now MORE than double the number of shaders rumored to be on GT300. Do you think 1600 shaders from ATI's current arch is going to be enough? I highly doubt it Marc. They'll need 2000 minimum to keep up, but this is only if they stick to the same 1+4 shader arch. If they seriously improve their shader architecture, which from reports seems unlikely, then they will require less of them IMHO.

Not so ridiculous.

And of COURSE they want to compete with Nvidia when it comes to single GPU. Otherwise, they wouldn't keep pumping up their existing single GPU's and clocking them to the moon.
No matter what ATI tells you, they wan't to be competitive in the top spot with a single GPU. When they can't, they say they never intended to and that was their plan all along. Don't fall for it man.

 

[ShadowOfMyself]

O rly? Then explain why AMD decided to make such a small chip compared to the GT200 when they could have gone with a much larger one and trump it in performance? (Considering the 4890 already matches the GTX280 in alot of games, this is a realistic assumption)

It was obviously a choice... What you are referring to was the case with the HD2900, which was late and huge and still failed to perform, then they came up with the lame excuse that it was a competitor to the GTS and not the GTX, but thats not the case nowadays anymore

 

[Keysplayr]

You're confused. It wasn't ATI that decided to make a small chip. It was Nvidia who decided to make such a large one with a focus on its GPGPU performance. Something ATI really didn't put more than a hint of thought into apparently. ATI didn't decide to make such an anemic chip. That was their design from the get go. Nvidia had much more in mind than just gaming. ATI? Not so much.

 

[Blazer7]

Originally posted by: Keysplayr

And of COURSE they want to compete with Nvidia when it comes to single GPU. Otherwise, they wouldn't keep pumping up their existing single GPU's and clocking them to the moon.
No matter what ATI tells you, they wan't to be competitive in the top spot with a single GPU. When they can't, they say they never intended to and that was their plan all along. Don't fall for it man.

Of course they want to compete but that doesn?t change a thing. AMD will still be the first with a DX11 part out there and with no real competition for half a year. We have to hand it to them. Their smaller-die/higher yield strategy (the rumors are that the 5870 will just be 205mm2), is working.

nV will have a monster of a GPU but it will arrive 6 months late and that?s not the worst. It will undoubtedly be more expensive to produce and as I have already mentioned AMD will have 6 months to prepare their answer. You can add the current economic state to the equation and it?s apparent that no matter whether nV will re-capture their crown it will be AMD that will be laughing all the way to the bank.

Originally posted by: Keysplayr
You're confused. It wasn't ATI that decided to make a small chip. It was Nvidia who decided to make such a large one with a focus on its GPGPU performance. Something ATI really didn't put more than a hint of thought into apparently. ATI didn't decide to make such an anemic chip. That was their design from the get go. Nvidia had much more in mind than just gaming. ATI? Not so much.

Maybe so or maybe, just maybe, the guys at ATI did their homework first because their anemic design IS working. Lower production costs result in cheaper products and during a recession period this is almost always a win-win situation and right now it?s AMD that has the advantage.

By the time the GT300 comes out things may be once again different. As far as GPGPU is concerned, by the time that GPGPU will actually matter to the average Joe things will undoubtedly be different.

The next few months will be very-very interesting.

 

[Idontcare]

Originally posted by: Keysplayr
No matter what ATI tells you, they wan't to be competitive in the top spot with a single GPU. When they can't, they say they never intended to and that was their plan all along. Don't fall for it man.

I too laughed out loud when I read this as the "excuse" for why ATI was under-performing in the single-GPU segment at the time the GT200 came out. You are spot-on, of course.

I laughed because I thought to myself "It's obvious why AMD has to say this, it is their job to do all they can to keep shareholder equity from eroding, but who is actually going to fall for such an obvious transparent marketing excuse?". Then I started reading threads...ah, now I see who the who was.

When I read this oft-quoted marketing "strategy" I can't help but to conjure up the visual of when Peewee falls of his bike in Pee-wee's Big Adventure (yes I admit to having watched it once) and then jumps up to claim "I meant to do that!".

Originally posted by: ATI
[falls off bike after attempting tricks] I meant to do that.

 

[Creig]

The question is, can Nvidia produce the GT300 while still setting the MSRP at a price people will be willing to pay? If the GT300 is indeed a 1.8 billion transistor GPU, you're talking one very big, very complicated, very expensive die to produce.

Remember, Nvidia initially thought they would be getting $650 for each of their high end GTX280s. Then ATI unveiled their 4800 series and all of a sudden Nvidia has to drop the price by $250. Losing $250 of profit per card is huge. The GTX260 was also forced to go from $450 to $300 in order to remain a viable alternative to ATI's offerings. So the GT200 hasn't exactly turned out to be the cash cow Nvidia originally envisioned it to be.

Since GPU design doesn't exactly happen overnight, Nvidia has no alternative but to continue with their big, expensive GPU strategy and come out with the even bigger GT300 in order to have a new product to release. In order for the GT300 to even have a chance of being successful for Nvidia, they will have to get good yields since most people won't put up with $650 price tags.

And yields are an issue they really struggled with on the GT200. Initial reports put the GT200 yield at around 40% while the 4800 was said to be around 70%. If these numbers are accurate, the GT200 was even more expensive to produce than its size alone would imply. Now we have the GT300 rumored to have 400 million more transistors than the GT200, and on a much smaller process to boot.

Originally posted by: Keysplayr
You're confused. It wasn't ATI that decided to make a small chip. It was Nvidia who decided to make such a large one with a focus on its GPGPU performance. Something ATI really didn't put more than a hint of thought into apparently. ATI didn't decide to make such an anemic chip. That was their design from the get go. Nvidia had much more in mind than just gaming. ATI? Not so much.

Focusing on gaming performance doesn't seem like all that bad of a strategy overall. Smaller die, lower costs, higher yields. Have you seen the current results of the PhysX poll on the AT homepage? 76% feel PhysX is 'marginal' to 'not useful' when considering buying a new card and 85% feel the same when buying software. GPGPU performance may be the icing on the video card cake, but it's still the cake that's most important.

The cake is not a lie.

 

[Wreckage]

Originally posted by: Creig


Focusing on gaming performance doesn't seem like all that bad of a strategy overall.

They don't have the fastest card or GPU. So by your measure their strategy failed.

 

[OCGuy]

Originally posted by: Creig
The question is, can Nvidia produce the GT300 while still setting the MSRP at a price people will be willing to pay? If the GT300 is indeed a 1.8 billion transistor GPU, you're talking one very big, very complicated, very expensive die to produce.

Remember, Nvidia initially thought they would be getting $650 for each of their high end GTX280s. Then ATI unveiled their 4800 series and all of a sudden Nvidia has to drop the price by $250. Losing $250 of profit per card is huge. The GTX260 was also forced to go from $450 to $300 in order to remain a viable alternative to ATI's offerings. So the GT200 hasn't exactly turned out to be the cash cow Nvidia originally envisioned it to be.

Since GPU design doesn't exactly happen overnight, Nvidia has no alternative but to continue with their big, expensive GPU strategy and come out with the even bigger GT300 in order to have a new product to release. In order for the GT300 to even have a chance of being successful for Nvidia, they will have to get good yields since most people won't put up with $650 price tags.

And yields are an issue they really struggled with on the GT200. Initial reports put the GT200 yield at around 40% while the 4800 was said to be around 70%. If these numbers are accurate, the GT200 was even more expensive to produce than its size alone would imply. Now we have the GT300 rumored to have 400 million more transistors than the GT200, and on a much smaller process to boot.

Originally posted by: Keysplayr
You're confused. It wasn't ATI that decided to make a small chip. It was Nvidia who decided to make such a large one with a focus on its GPGPU performance. Something ATI really didn't put more than a hint of thought into apparently. ATI didn't decide to make such an anemic chip. That was their design from the get go. Nvidia had much more in mind than just gaming. ATI? Not so much.

Focusing on gaming performance doesn't seem like all that bad of a strategy overall. Smaller die, lower costs, higher yields. Have you seen the current results of the PhysX poll on the AT homepage? 76% feel PhysX is 'marginal' to 'not useful' when considering buying a new card and 85% feel the same when buying software. GPGPU performance may be the icing on the video card cake, but it's still the cake that's most important.

The cake is not a lie.

AT does not represent the GPU market as a whole. Every person that comes in and starts a thread gets recommended an ATi card from the 4XXX series. We recently saw the discreet graphics market share numbers, and learned that this really isnt the big picture.

 

[Creig]

Originally posted by: OCguy
AT does not represent the GPU market as a whole. Every person that comes in and starts a thread gets recommended an ATi card from the 4XXX series.

Sounds like the majority of people here feel that the 4XXX series offers the best value. Which, coincidentally, mirrors a lot of the reviews out there. Not too long ago it was the 8800GT that was being recommended here overwhelmingly. It's all about the best bang for your buck and Nvidia just doesn't have much of it this round.

Originally posted by: OCguy
We recently saw the discreet graphics market share numbers, and learned that this really isnt the big picture.

What do market share numbers have to do with PhysX? Just because somebody buys an Nvidia card doesn't automatically mean that they are gung-ho about PhysX. Over 50% of the more than 11,000 people who voted in that survey said they own a PhysX capable video card. Yet the vast majority of them gave PhysX a great big "Meh".

 

[Creig]

Originally posted by: Wreckage
They don't have the <snip>

Whatever.

 

[Wreckage]

Originally posted by: Creig

Originally posted by: Wreckage
They don't have the <snip>

Whatever.

I know it sucks that I'm right all the time.

 

[error8]

Originally posted by: Wreckage

Originally posted by: Creig

Originally posted by: Wreckage
They don't have the <snip>

Whatever.

I know it sucks that I'm right all the time.

:laugh: :laugh:
Man I laughed with tears at this. :beer: It really lightened up my mood. Thanks for that.

 

[Creig]

Originally posted by: Wreckage

Originally posted by: Creig

Originally posted by: Wreckage
They don't have the <snip>

Whatever.

I know it sucks that I'm right all the time.

No, it's just that every time I read one of your posts I get this mental image of you running around in a green skirt waving green pom-poms. And I dislike making myself that nauseous.

 

[Blazer7]

Originally posted by: Wreckage

Originally posted by: Creig

Originally posted by: Wreckage
They don't have the <snip>

Whatever.

I know it sucks that I'm right all the time.

nV may still lead in performance but for how long? What are they gonna do when the 5870 comes out?

As a consumer, and one considering a new graphics card, I'm a bit worried. For a consumer competition is always a good thing. Right now I don't see how on earth nV can compete with AMD.