GeForce Titan coming end of February

[blackened23]

Single large GPU vs two small GPUs

Raw performance for 690 might be better, but what happens when a game doesn't scale well? Or worse, doesn't scale at all? The occasional compatibility issues where completely disabling SLI may be required? All those instances are instances were Titan would destroy a 690 since that 690 would be effectively rendered as a very expensive 670.

Then there's the fact that the larger die itself i'm sure is quite a bit more costly to produce. And of course, the epeen tax.

I don't like the price, but it isn't exactly unexpected when put into perspective of nVidia's current offerings at their current prices.

That's not a valid justification of price because the 690 has frame metering technology and all websites that have tested latency with SLI has found excellent results. The only possible justification is the additional VRAM which I find completely stupid , because 6GB VRAM is completely worthless even for multi monitor resolutions. VRAM has been tested extensively by hardocp between 2 and 4GB configurations, with the additional VRAM latency on 4GB cards actually proving slower than 2GB in most cases even at surround resolutions. Why nvidia didn't do a 3GB card is completely baffling.

 

[BallaTheFeared]

Pros

-- Fastest single GPU in the world

-- Better Perf/Watt than the 690, which made AMD's getup look like Fermi

--- Features AMD will attempt to copy, but fail miserably at

-- Enable/Disable DP performance unknown in prior generations of Nvidia cards to increase productively when needed, or to increase Perf Perf/watt when not

-- Dynamic clock adjustments based on several different user set variables

-- Unlocked voltage control

-- Top of the line rear exhaust cooler, less noise than a reference 680, makes the 7970GHz sound worse than it already does

-- 6GB of 6000MHz GDDR

-- Dynamic overclocking that responds to your fan noise tolerance, or excels when a real enthusiast places it where it should be, under water

-- Looks like something Jesus would have turned Gold into

-- Everyone wishes they could have it

-- Industry leading smoothness, unparalleled driver support and features

-- PhysX, TXAA, CUDA

-- Surround +1 on a single card without need for expensive adapters


Did I miss any? I'm sure I did!


Cons

-- Price

-- AMD didn't make it

 

[Keysplayr]

So, how many Titans are you getting then ? :awe:

None.

 

[Imouto]

I'm just thinking that ASUS making a MARS out of this would be awesome. 3K? More?

 

[2is]

That's not a valid justification of price because the 690 has frame metering technology and all websites that have tested latency with SLI has found excellent results. The only possible justification is the additional VRAM which I find completely stupid , because 6GB VRAM is completely worthless even for multi monitor resolutions. VRAM has been tested extensively by hardocp between 2 and 4GB configurations, with the additional VRAM latency on 4GB cards actually proving slower than 2GB in most cases even at surround resolutions. Why nvidia didn't do a 3GB card is completely baffling.

Ummm, i'm not sure if you didn't read what I posted or aren't understanding it. I said nothing about frame latency. I was talking about scaling that ranges from good to bad to none. (SLI scaling and latency/micro-stuttering isn't the same thing) You're also ignoring the actual cost of the card which is higher.

 

[Keysplayr]

Pwned!

In what way?

 

[blastingcap]

Ummm, i'm not sure if you didn't read what I posted or aren't understanding it. I said nothing about frame latency. I was talking about scaling that ranges from good to bad to none. (SLI scaling and latency/micro-stuttering isn't the same thing) You're also ignoring the actual cost of the card which is higher.

I might be wrong but I thought Titan power draw was also lower, and if those slides are real it's also quieter than GTX 680 SLI and thus presumably also GTX 690. Plus voltage unlocked and GPU Boost 2.0 (allegedly better for watercooling based on Anandtech's preview). The big one though is not having to deal with SLI at all if you don't want to.

 

[VulgarDisplay]

Originally Posted by Tweak155
Pwned!

In what way?

Originally Posted by Annisman*
So, how many Titans are you getting then ?

None.

This way.

 

[Ferzerp]

accusing him of having an agenda

He does have an agenda.... Preaching to everyone that anything above bare essentials is stupid and judging everyone that enjoys anything He's consistent across all the subforums that he posts on at least

 

[Elfear]

Pros

-- Fastest single GPU in the world

-- Better Perf/Watt than the 690, which made AMD's getup look like Fermi

--- Features AMD will attempt to copy, but fail miserably at

-- Enable/Disable DP performance unknown in prior generations of Nvidia cards to increase productively when needed, or to increase Perf Perf/watt when not

-- Dynamic clock adjustments based on several different user set variables

-- Unlocked voltage control

-- Top of the line rear exhaust cooler, less noise than a reference 680, makes the 7970GHz sound worse than it already does

-- 6GB of 6000MHz GDDR

-- Dynamic overclocking that responds to your fan noise tolerance, or excels when a real enthusiast places it where it should be, under water

-- Looks like something Jesus would have turned Gold into

-- Everyone wishes they could have it

-- Industry leading smoothness, unparalleled driver support and features

-- PhysX, TXAA, CUDA

-- Surround +1 on a single card without need for expensive adapters


Did I miss any? I'm sure I did!


Cons

-- Price

-- AMD didn't make it

Wow. You were actually sounding pretty reasonable for awhile after you started posting on AT again but it has come full circle with all the cheerleading.

 

[Grooveriding]

GeForce Tit$up, only from nvidia.

 

[sontin]

I miss the good old days. A $549 7970 with crappy drivers, a crappy cooler and 42% more performance over the 6970 was such an amazing offering from AMD.

Damn, what a shame that we don't have a second graphics company on the market to give AMD a little bit of competition.

 

[Tweak155]

I miss the good old days. A $549 7970 with crappy drivers, a crappy cooler and 42% more performance over the 6970 was such an amazing offering from AMD.

Damn, what a shame that we don't have a second graphics company on the market to give AMD a little bit of competition.

Wat?

 

[BallaTheFeared]

Wow. You were actually sounding pretty reasonable for awhile after you started posting on AT again but it has come full circle with all the cheerleading.

You have a problem with my list?

Could you point out the issues instead of resorting to personal attacks?

I never said it was the definitive list, I only added what I thought were pros and cons to the card. I'm fully open to any critiques some might have, please let me know what you think.

 

[Haserath]

You have a problem with my list?

Could you point out the issues instead of resorting to personal attacks?

I never said it was the definitive list, I only added what I thought were pros and cons to the card. I'm fully open to any critiques some might have, please let me know what you think.

Your list sounds like marketing on some points. Maybe?

 

[BoFox]

Indeed but how do you know the sp on both the 5870 and 6870 are same? they must have tweaked individual sp.NV increased the number of cuda cores in Kepler but made them weaker compared to Fermi.They choose this path as they hope applications in the future will benefit more from extra shaders.Your observation is valid as very few consumer apps stress all the shaders.

I actually had a conspiracy theory that the 6870 was using VLIW4 shaders, despite AMD saying that it used VLIW5. It still does seem like as if HD 6870 actually has 1280 rather than 1120 shaders......... perhaps I'll try to approach AMD engineers about it again, in a different way (that is twice as kind). It looks like a good topic to discuss in the VP thread - I'm thinking of inquiring a bit more into this, as hardware specs need to be discussed at some point, in how they affect video card performance. There seems to be some kind of algorithm that shows how performance scales with certain specs, in relation to each other.. like a GPU "Theory of Relativity".

What I've noticed is that GPUs in the past usually went all-out on ROPs and bus width until GPGPU came into the picture. With GPGPU applications, the primary focus on GPU core muscle becomes much greater than on bandwidth or ROP back-end output. For example, with Bitcoin mining, my 5870 does not show any penalty as I underclock the memory by a full GHz.. AT ALL. I'm also pretty sure that if my 5870 had half the ROPs (16 rather than 32) while having the same bandwidth, it would still churn out just as many Mhps.. or at least the difference would be within 1-2% rather than over 10% for most games when using 8x AA at 1920x1200 or higher. Most likely, the difference would be closer to 0%.

G80 and GT200 were still pretty formidable with bandwidth and pixel fillrate capacity, "relatively". Fermi really showed a relative lack of progress with bandwidth, while Kepler is showing lack with BOTH bandwidth and ROPs. On the AMD side, relatively - R700 showed lack with ROPs, Evergreen showed lack with bandwidth, and Tahiti with ROPs.

P.S. - ADDED :
If it were not for GPGPU, Nvidia would have definitely cut down on the number of shader processors used for Titan, perhaps using only 13 SMX instead of 14 (cut-down from 15), and increased the number of ROPs from 48 to 64, along with bus width from 384-bit to 512-bit. Then we'd have the sweet spot 4GB versions along with a special 8GB version. That way, GPU muscle could be more fully unlocked - especially when using 8x AA with 2560x1600 or higher.

 

[BoFox]

Would any of you be surprised if NV releases a cut-down 13 SMX version with 3GB of RAM, and call it GTX 780 - a couple months later on?

The bandwidth and ROP count might remain the same (along with the core clock), so that it stays quite attractive (without losing too much performance) while retaining a $649 MSRP or so.

I'd also expect NV to completely remove the 1/3 DP functionality, by locking it at 1/24 DP - reserving fast DP for Titan only.....


P.S. --

Yeah, if AMD is really smart, AMD should be officially launching HD 7990 (dual 7970GE) as an official model number on the same day as Titan is launched (this Thursday), to steal thunder away from it. It should be even faster than GTX 690, for the same price as both 690 and Titan.

The sweet thing would be that the 7990 would kill Titan at DP performance (being nearly 2x as fast - especially when oc'ed).

I know that it would piss off Asus, as they are trying to sell Ares II for $1500, but it's further oc'ed (50Mhz higher) with a special watercooled design and quality PCB stuff.

If review sites had both Titan and 7990 to cover on the same launch day, it would definitely dampen the impact of Titan quite a bit!!!! The 7990 would be destroying it in most benchmarks for the same $, give about 60% more DP performance, etc..

P.S.S. -
If 7990 is truly to be GHz Edition and not vanilla 7970x2, then it might have to be watercooled. AMD did that with FX-8150, so they definitely could do that with the card, with an additional benefit of vastly cutting down on power draw due to cooler temps reducing leakage (and also allowing for 2 of them to run together in CF with less chances of having heat issues).

 

[blastingcap]

Would any of you be surprised if NV releases a cut-down 13 SMX version with 3GB of RAM, and call it GTX 780 - a couple months later on?

No. In fact that is what I am expecting. So many big GK110 dies can't all be perfect or even near-perfect; I am sure they have plenty of 13 SMX parts and maybe even sub-13 SMX parts among all the Tesla rejects.

 

[f1sherman]

Too much emo; it's not like they're taking our guns or water.
This is how complaining should be done.

If NVIDIA's selling it for $1000, that's probably the best decision for them, I'm sure they've run the numbers. For us, it's a bit disappointing, because it's essentially a pointless product for the vast majority of gamers: we can't afford it, and it's so expensive that it's not going to push down the prices of other products either (GTX 680, HD 7970 GE, etc.). Effectively, Titan makes no difference.

That's not what usually happens when a GPU vendor comes up with a newer, faster GPU. At worst, it costs about $600 and doesn't push down the rest of the stack very much. Sometimes, it's actually affordable and triggers price cuts for just about everything else. NVIDIA's no charity, I don't blame them, I'm just disappointed.

http://forum.beyond3d.com/showpost.php?p=1709197&postcount=1071

Yeah, if AMD is really smart, AMD should be officially launching HD 7990 (dual 7970GE) as an official model number on the same day as Titan is launched (this Thursday), to steal thunder away from it.

My thoughts exactly

No. In fact that is what I am expecting. So many big GK110 dies can't all be perfect or even near-perfect; I am sure they have plenty of 13 SMX parts and maybe even sub-13 SMX parts among all the Tesla rejects.

Those dies need to go somewhere and NV is like a whaling industry - nothing goes to waste

 

[Keysplayr]

This way.

I don't get it. Should I get it? Probably something stupid.

 

[Hauk]

Hey don't leave us behind buddy.
Bridge may be our home, but we love Crysis 3 as much as the next guy.

$199 and some foodstamps plx

Maybe not Titan, but a solid Kepler like 670 would be nice for $199

 

[Keysplayr]

Listen guys can we stop with the freaking poverty stories here? If you have the money, and want the Titan, you buy it. I don't want to hear that this is a 1st world problem. We aren't all going to give up all our worldly possessions and become priests, monks and nuns.

 

[iMacmatician]

Would any of you be surprised if NV releases a cut-down 13 SMX version with 3GB of RAM, and call it GTX 780 - a couple months later on?

The bandwidth and ROP count might remain the same (along with the core clock), so that it stays quite attractive (without losing too much performance) while retaining a $649 MSRP or so.

I'd also expect NV to completely remove the 1/3 DP functionality, by locking it at 1/24 DP - reserving fast DP for Titan only.....

I wouldn't be surprised at all (although depending on specs it may still be called "Titan." I'm half expecting an even further cut down card with 11 or 12 SMXes, a 320-bit bus, and performance not far above that of the 680.

 

[Rubycon]

Limited to 3 way SLI?
These means no TitanX (dual GPU card)??!

 

[Smoblikat]

Limited to 3 way SLI?
These means no TitanX (dual GPU card)??!

Linus has a card and he said it can do quadSLI