Nvidia reveals Specifications of GT300

[imported_Shaq]

Originally posted by: Genx87
I am looking forward to this chip. But of course it will require me to play something other than WoW or Call of Duty W@W to take advantage of it

Nah...just add 4x4 supersampling at 2560 resolution and some combined 8XMSAA and transparency supersampling. LOL

 

[Cookie Monster]

Originally posted by: yacoub
It's just a shame ATi went the wrong way and tightened the bus bandwidth on their 4770 and cranked up the clockspeed to make up for it, instead of running a cooler, lower wattage part with wider bus and lower clockspeeds. 256-bit DDR5 would have been a winner, at least according to my understanding of how that would perform compared to 128-bit w/ higher clocks.

So you expect a $99 dollar part to have GDDR5 + a 256bit memory bus? Not to mention that by increasing from a 128bit bus to 256bit bus will increase the current die size by quite a bit.

 

[AzN]

Originally posted by: Cookie Monster

Originally posted by: yacoub
It's just a shame ATi went the wrong way and tightened the bus bandwidth on their 4770 and cranked up the clockspeed to make up for it, instead of running a cooler, lower wattage part with wider bus and lower clockspeeds. 256-bit DDR5 would have been a winner, at least according to my understanding of how that would perform compared to 128-bit w/ higher clocks.

So you expect a $99 dollar part to have GDDR5 + a 256bit memory bus? Not to mention that by increasing from a 128bit bus to 256bit bus will increase the current die size by quite a bit.

Not to mention 4870 isn't even all that much faster over 4850 with nearly 2x the bandwidth. Majority of bandwidth on 4870 is going to waste. 60gb/s is sufficient for a GPU like RV7xx series. A good 80gb/s should be more than plenty. 4770 does so well compared to 4850 and 4830 because of clock speeds not bandwidth anyway.

It would have been more cost efficient/performance pleasing if AMD added more SP and TMU instead of adding 256bit bus proposed by yacoub.

 

[Idontcare]

Originally posted by: Cookie Monster

Originally posted by: yacoub
It's just a shame ATi went the wrong way and tightened the bus bandwidth on their 4770 and cranked up the clockspeed to make up for it, instead of running a cooler, lower wattage part with wider bus and lower clockspeeds. 256-bit DDR5 would have been a winner, at least according to my understanding of how that would perform compared to 128-bit w/ higher clocks.

So you expect a $99 dollar part to have GDDR5 + a 256bit memory bus? Not to mention that by increasing from a 128bit bus to 256bit bus will increase the current die size by quite a bit.

As a technologist, I'm just forever amazed at how much can be bought for so little every time a new node comes out. All that for $100 retail...amazing when you start contemplating just how much time and effort has gone into making it a possibility over the past 4 yrs.

And in another year they'll be on to 28nm and all this will be yesterday's old news again.

 

[Creig]

Originally posted by: Keysplayr

Originally posted by: error8

Originally posted by: Keysplayr


I'm not certain I agree with NV still using GDDR3 for higher end cards. GDDR5 has been out for a good while now, and I think it's pricing isn't as cost prohibitive as it once was, and may be a contributing reason why 4870 and up have been coming down in price. Also look at 4770 utilizing GDDR5 now.

Thank god for ATi, because if it wasn't for them, we would have still seen ddr3 on Nvidia high end cards, even in 2012. :laugh:

AFAICT, GDDR3 seems to be doing fine. Time to move on? For sure.

If GDDR3 + 512 bit = GDDR5 + 256 bit as far as overall bandwidth goes, then I personally don't see any problem with going either way. The only real difference is that the 512 bit PCB would be more complex, possibly with more layers than a 256 bit board.

I guess you would have to weigh that against the price difference between GDDR3 and GDDR5, plus the energy savings of GDDR5 over GDDR3. But I agree that, in the future, GDDR3 will most definitely be phased out in favor of GDDR5.

 

[Forumpanda]

I am amazed at how consistent and predictably new nodes are coming along, are they simply tossing money after engineering problems until they go away?
Saving small delays here and there it would seem to me that there should be a bit more variance in the introduction of new nodes.

 

[Acanthus]

That chip will be large, even at 40nm.

MIMD processing cores are a lot larger than SIMD.

This spec sheet looks like a dual chip solution to me.

 

[Acanthus]

Originally posted by: Forumpanda
I am amazed at how consistent and predictably new nodes are coming along, are they simply tossing money after engineering problems until they go away?
Saving small delays here and there it would seem to me that there should be a bit more variance in the introduction of new nodes.

A few companies work together to get these technologies off the floor on schedule.

But yes, the R&D budget is enormous, because a multi billion dollar industry relies entirely on its advance.

 

[cbn]

Originally posted by: Azn

Originally posted by: Cookie Monster

Originally posted by: yacoub
It's just a shame ATi went the wrong way and tightened the bus bandwidth on their 4770 and cranked up the clockspeed to make up for it, instead of running a cooler, lower wattage part with wider bus and lower clockspeeds. 256-bit DDR5 would have been a winner, at least according to my understanding of how that would perform compared to 128-bit w/ higher clocks.

So you expect a $99 dollar part to have GDDR5 + a 256bit memory bus? Not to mention that by increasing from a 128bit bus to 256bit bus will increase the current die size by quite a bit.

Not to mention 4870 isn't even all that much faster over 4850 with nearly 2x the bandwidth. Majority of bandwidth on 4870 is going to waste.

That makes sense to considering it looks like GTX 285 has more processing power than HD4870 and roughly the same bandwidth (512 bit GDDR3= 256 bit GDDR5)

 

[evolucion8]

Originally posted by: Just learning

Originally posted by: Azn

Originally posted by: Cookie Monster

Originally posted by: yacoub
It's just a shame ATi went the wrong way and tightened the bus bandwidth on their 4770 and cranked up the clockspeed to make up for it, instead of running a cooler, lower wattage part with wider bus and lower clockspeeds. 256-bit DDR5 would have been a winner, at least according to my understanding of how that would perform compared to 128-bit w/ higher clocks.

So you expect a $99 dollar part to have GDDR5 + a 256bit memory bus? Not to mention that by increasing from a 128bit bus to 256bit bus will increase the current die size by quite a bit.

Not to mention 4870 isn't even all that much faster over 4850 with nearly 2x the bandwidth. Majority of bandwidth on 4870 is going to waste.

That makes sense to considering it looks like GTX 285 has more processing power than HD4870 and roughly the same bandwidth (512 bit GDDR3= 256 bit GDDR5)

Theorically, the HD 4870 can untap more FLOPS than any single GTX GPU, but in real life, is easier to reach the maximum speed with the GTX than with the HD card.

 

[BenSkywalker]

Theorically, the HD 4870 can untap more FLOPS than any single GTX GPU, but in real life, is easier to reach the maximum speed with the GTX than with the HD card.

Actually, it is much easier to hit peak rate using the 48xx core, it just can only do it running nearly useless Vec5, well, calling it nearly useless may be a bit harsh but it can't be used for the overwhelming majority of calculations.

 

[lopri]

Originally posted by: Azn
Not to mention 4870 isn't even all that much faster over 4850 with nearly 2x the bandwidth. Majority of bandwidth on 4870 is going to waste. 60gb/s is sufficient for a GPU like RV7xx series. A good 80gb/s should be more than plenty. 4770 does so well compared to 4850 and 4830 because of clock speeds not bandwidth anyway.

It would have been more cost efficient/performance pleasing if AMD added more SP and TMU instead of adding 256bit bus proposed by yacoub.

I am not sure if this view is correct? For both CPUs and GPUs, we've seen quite a few new memory technology brings minimal performance gain. Take DDR2 to DDR3 transition for desktop CPUs for example. Be it for Phenom or Core 2, the gain was minimal.

RV770 is actually the first thing I've seen that reaps such a massive gain from increased bandwidth. The only difference between HD 4850 and HD 4870 is memory being used, and thus it alone brings 20~30% performance increase at least. I mean, we don't normally expect such a dramatic difference by upgrading memory.

That's why I wondered in the past why NV hadn't mated G92 with GDDR5. Because G92 looked like an ideal candidate for a similar case. (like RV770)

If GT300 is really 512-bit using GDDR5.. That'll be a monster, for sure.

 

[thilanliyan]

Some more info (2.4billion transistors apparently):
http://www.hardware-infos.com/...hp?news=2944&sprache=1

 

[Idontcare]

Originally posted by: thilan29
Some more info (2.4billion transistors apparently):
http://www.hardware-infos.com/...hp?news=2944&sprache=1

Just an incredible number of xtors being put to work there. Amazing if true.

 

[Blazer7]

If anyone cares about Charlie's opinion here you go.

The INQ : A look at the Nvidia GT300 architecture

 

[OCGuy]

Originally posted by: Blazer7
If anyone cares about Charlie's opinion here you go.

The INQ : A look at the Nvidia GT300 architecture

Nope, dont care. That site is not worth the bandwidth.


Anyone who reads into anything that guy says....well, lets just say hes preaching to the choir and knows his audience

Even if GT300 is later, who really cares? If it is as good as expected, ill wait until next summer.

 

[Blazer7]

Originally posted by: OCguy

Originally posted by: Blazer7
If anyone cares about Charlie's opinion here you go.

The INQ : A look at the Nvidia GT300 architecture

Nope, dont care. That site is not worth the bandwidth.


Anyone who reads into anything that guy says....well, lets just say hes preaching to the choir and knows his audience

Yeah, Charlie is not to be trusted when he speaks about nV. However one can have some fun with his predictions.

Even if GT300 is later, who really cares? If it is as good as expected, ill wait until next summer.

I do. I'm also waiting for the GT300 but if they miss Q4 I'll be forced to go for something else.

 

[Rusin]

So 71.5% more transistors get 89% more SPs (GT200 - 270SP --> GT300 - 512SP)?

 

[ilkhan]

Rusin: Im pretty sure GT200 only hits 240SPs.

 

[Rusin]

Originally posted by: ilkhan
Rusin: Im pretty sure GT200 only hits 240SPs.

GT200 has 240 single precision units and 30 double precision units.

Performance of cards is pretty low when there's 64-bit floating point stuff:
GT200: 1/8 of normal performance
RV770: 1/5 of normal
GT300: 1/2 of normal [512 single precision units; two of them could do double precision stuff]

So amount of double precision shaders would go from 30 to 256 with GT200 -> GT300. Doesn't help in games, but scientific calculations and such would benefit from this greatly.

 

[Creig]

Originally posted by: thilan29
Some more info (2.4billion transistors apparently):
http://www.hardware-infos.com/...hp?news=2944&sprache=1

There's no way they're going to get 2.4 billion transistors working on a single die at 40nm. If they ever tried, their yields would be somewhere around 2%.

 

[thilanliyan]

Originally posted by: Creig

Originally posted by: thilan29
Some more info (2.4billion transistors apparently):
http://www.hardware-infos.com/...hp?news=2944&sprache=1

There's no way they're going to get 2.4 billion transistors working on a single die at 40nm. If they ever tried, their yields would be somewhere around 2%.

What makes you so sure?

 

[Idontcare]

Originally posted by: thilan29

Originally posted by: Creig

Originally posted by: thilan29
Some more info (2.4billion transistors apparently):
http://www.hardware-infos.com/...hp?news=2944&sprache=1

There's no way they're going to get 2.4 billion transistors working on a single die at 40nm. If they ever tried, their yields would be somewhere around 2%.

What makes you so sure?

ATI HD4770 is 826m xtors packed into 137 mm^2.

At that xtor density, 2.4B xtors could be packed into a 398 mm^2 die.

Seems feasible to me based on the existing data.

 

[Keysplayr]

Originally posted by: thilan29

Originally posted by: Creig

Originally posted by: thilan29
Some more info (2.4billion transistors apparently):
http://www.hardware-infos.com/...hp?news=2944&sprache=1

There's no way they're going to get 2.4 billion transistors working on a single die at 40nm. If they ever tried, their yields would be somewhere around 2%.

What makes you so sure?

He isn't.

 

[thilanliyan]

Originally posted by: Idontcare
ATI HD4770 is 826m xtors packed into 137 mm^2.

At that xtor density, 2.4B xtors could be packed into a 398 mm^2 die.

Seems feasible to me based on the existing data.

nVidia's transistor density is lower IIRC in the GT200 so they'd have to work on that to get that ~400mm^2 size.