NVIDIA: Beyond G80...

[Ailuros]

Given posts 417/418, allow me to place an innocent bet :

- 650~670MHz core, 1600~1800MHz in the shader clock department.
- 768MB/1.5GB of 2.4GHz GDDR4 memory.
- Dual-slot height self-contained water cooling solution.

Once you're there, if the $999 pricetag should have any merit, go for 1.5GB on-board ram.

If all the above should be true, the price/performance ratio compared to the 8800GTX will be rather ridiculous.

 

[Kocur]

Inkster, I think that you are on spot with memory size. I do not agree on the clocks though.

My bet is as follows:
80nm,
700MHz core,
2000MHz shader domain,
1.5GB of DDR4 memory at 2.2/2.4GHz.

BTW, has the possibility of G80 core having 160 shaders been completely discarded already?

 

[trinibwoy]

Well to begin with I'd discount all theories about a $1000 part being some sort of R600 killer - that doesn't work when cost is 70% more than the competition.

650Mhz is a joke considering there are already overclocked versions of G80 near that mark. So is 1.5GB - please raise your hand if you think another 768MB is going to make an iota of difference

I agree with Ail, so far the proposed clocks would make this part little more than a joke compared to current G80 based cards going for $500-$600 (this has really been a textbook case of lack of competition keeping prices high). Not to mention if R600 is faster than these overclocked G80's and comes in at ~$650.

 

[Kocur]

trinibwoy,

I do not think that 1.5GB is a joke. It will definitely help in the 2560 x 1620 resolution with AA turned on (imho the new card is going to target this very resolution).

I expect the Ultra card to be, on average, 35% faster than 8800GTX in demanding modes (from 20% in games like CoD or Fear to around 50% in shader intensive games like Call of Juarez).

Of course, this card is not intended for price/performance consious buyers.

 

[trinibwoy]

Well it will be interesting to see if R600 comes in a 512MB SKU and how it fares against G80. We don't have any evidence of 768MB providing any significant additional advantage over the GTS or X1950XTX even at 2560x1600 w/AA. So I'm not putting any faith in 1GB or 1.5GB doing any better.

 

[_xxx_]

So is 1.5GB - please raise your hand if you think another 768MB is going to make an iota of difference

As a marketing argument, definitely.

 

[trinibwoy]

Yeah but we're not talking about the 7300-512MB crowd here. I doubt 1.5GB is going to sway many people if performance isn't there.

 

[dnavas]

True, but, the alternative may be less attractive. What if your card stomps the R600 by some "stupidendous" amount ... except memory, which is only 3/4ths as large? Another alternative would be to raise the memory to a gig, but, that would certainly mean a rather larger design effort. Also, if you think NV is targetting bandwidth, how do you explain the 8600GTS? :| I'm not saying I buy 1.5G, but, I wouldn't dismiss it (yet) either.

My thought is to come at it from theoretical peak numbers. For example: 320 shaders by 800Mhz is the competition, so, either you need 160 shaders at larger than 1600Mhz, or 128 shaders at >2Ghz. Place your bets....

I don't see how they tackle the bandwidth comparison without some serious work, unless they just ignore it [And if they don't spend some money/time to get all of those shaders running Folding ... ]

 

[trinibwoy]

Well my point is that no amount of marketing parity is going to make up for that ridiculous price. If R600 has 1GB of RAM and Nvidia feels compelled to one-up them then tacking on another 768MB and asking for $1000 is not the way to go. I don't see why they would be so insecure about RAM size at this point - they must have known about the very real possibility of AMD going to 1GB with R600 back when they were designing their various G80 SKU's.

 

[Geo]

Well my point is that no amount of marketing parity is going to make up for that ridiculous price. If R600 has 1GB of RAM and Nvidia feels compelled to one-up them then tacking on another 768MB and asking for $1000 is not the way to go. I don't see why they would be so insecure about RAM size at this point - they must have known about the very real possibility of AMD going to 1GB with R600 back when they were designing their various G80 SKU's.

When G80 launched we asked them if they knew of any current or near term scenarios where 768MB would be a performance win vs 512MB, and they couldn't point at any. I'd think if anything they'd like the cost advantage of 1GB vs 768MB. Just as, I suspect, AMD likes the cost advantage of 768MB vs 512MB.

 

[Jawed]

Hmm, I'm quite excited about a $1000 8800U. Not to buy, just to see what it can do.

Jawed

 

[Geeforcer]

When G80 launched we asked them if they knew of any current or near term scenarios where 768MB would be a performance win vs 512MB, and they couldn't point at any. I'd think if anything they'd like the cost advantage of 1GB vs 768MB. Just as, I suspect, AMD likes the cost advantage of 768MB vs 512MB.

What price range do we expect 1GB R600 to be in? If its in exsess on $800, then 8800 Ultra will make somewhat more sense.

 

[dnavas]

Well my point is that no amount of marketing parity is going to make up for that ridiculous price.

Depends if the point is to sell them or just have them bench'd

I think 1.5G isn't likely because it raises the bar beyond where you'd want a refresh competing at....
[same reason why 7900 was rumored NOT to have 32 'pipes']

The launch of an 8800 Ultra would seem to indicate that their refresh isn't ready -- either because their 80nm is running behind, or their 65nm isn't quite ready yet. I think that's far more interesting than 1.5G either way.

 

[jimmyjames123]

I have a feeling that the so-called 8800 Ultra will be significantly faster than the 8800 GTX. It wouldn't make much sense to launch a slightly overclocked 8800 GTX, since those are already out on the market. I wouldn't be surprised to see 30% or even higher gains in some instances vs 8800 GTX. We will find out soon I suppose.

 

[icecold1983]

I have a feeling that the so-called 8800 Ultra will be significantly faster than the 8800 GTX. It wouldn't make much sense to launch a slightly overclocked 8800 GTX, since those are already out on the market. I wouldn't be surprised to see 30% or even higher gains in some instances vs 8800 GTX. We will find out soon I suppose.

that never stopped ati. x1950 x 1900. x850 v x800

 

[jimmyjames123]

That's correct, but that was not in the midst of a next generation product launch by a competitor. I imagine that we will see a beefed up card from NV, like the 7800 GTX 512MB was beefed up G70 model launched right around the time when X1800 came into play.

 

[Blazkowicz]

trinibwoy,

I do not think that 1.5GB is a joke. It will definitely help in the 2560 x 1620 resolution with AA turned on (imho the new card is going to target this very resolution).

or it will/might be useful with two cards in SLI. Imagine UE3/Crytek games or tech demos running in those conditions with really high res (2K*2K, 4K*4K..) textures/normal maps/fubar maps etc.
I like much the idea of a 1.5GB card, as it already exist under the quadro name, it's huge, pointless but less pointless than a 10% speed bump over a slightly o/c model. And ATI will have a 1GB model anyway

 

[trinibwoy]

If this 8800 Ultra does exist and is supposed to launch in two weeks you would think there'd be some pictures/info by now. I'm beginning to think this round won't be decided in May.

 

[Geo]

If this 8800 Ultra does exist and is supposed to launch in two weeks you would think there'd be some pictures/info by now. I'm beginning to think this round won't be decided in May.

If an 8800 Ultra launches in two weeks I'll be the most surprised guy in the room. I've been surprised before, of course . . . I'd think it'd make more sense for NVIDIA to wait for R600 launch to know for sure what they are shooting at at this point.

 

[MulciberXP]

I don't see how it matters that they wait till launch to know what they're shooting for. Most likely they already know exactly what the R600 is and what it's capable of. And whats the fastest they could possible do anything to mount a response? Changing the specs slower or faster to better target the R600 would require quite a bit of lead time. It implies the AIBs aren't already in production of their boards, so they'd need time to ramp up and ship the parts out. I have no idea how long that takes. *shrug*