NVIDIA Kepler speculation thread

[RecessionCone]

Meh...I'm a Physicist / Materials Scientist and I'm fine with channel capacity or throughput, but bandwidth - at least in Physics - has a very specific meaning and it's not throughput or channel capacity. It is used to calculate those parameters, but it is one of many variables.

Should have guessed you're a physicist.
Physicists always think they know everything - at least that's the joke amongst the electrical engineers. I'm just trying to explain to you why it makes sense to use the term bandwidth to describe bits/second: because bits/second is fundamentally related to the spectral bandwidth of the signal carrying information. At a constant signal to noise ratio, there's a linear relationship between the two, so the concepts are compatible.

 

[Mize]

Should have guessed you're a physicist.
Physicists always think they know everything - at least that's the joke amongst the electrical engineers. I'm just trying to explain to you why it makes sense to use the term bandwidth to describe bits/second: because bits/second is fundamentally related to the spectral bandwidth of the signal carrying information. At a constant signal to noise ratio, there's a linear relationship between the two, so the concepts are compatible.

LOL. As long as the EEs get a laugh and get their work done on time I won't fire them
Physics was the Bachelor's my MSE and PhD are Materials Science focused on sensors and modeling. Even had a nice 3D wavelet in my dissertation thanks to a brilliant Mathematician I worked with. Anyway, not to derail any further I'll give you that they're, as you said, related (integrating over the space of one yields the other) and that they are colloquially the same for IT and computer EE...but, find me a dyed in the wool, artist of an analog, RF EE and he'll cringe too (I know this from experience). Don't worry though, it's just because we're old an haven't kept up with the times For us old folks we'll call it "digital bandwidth" so we don't get confused...or probably better we call the old one "analog bandwidth" and get used to "bandwidth = digital bandwidth"

 

[Tridam]

So it's 4x 48-wide SIMDs? Operating as 16x2-clock+32x1-clock with the 16 lanes having one subset of instructions and the 32 lanes having another subset?

Think simple, simple is more efficient

 

[jaredpace]

 

[Rangers]

There were a ton available yesterday...and they sold out very fast. I was showing stock until after lunch yesterday of about 8 different boards. Sorry you didn't notice, but, like you're branding of B3D as Nvidia territory, your lack of observation doesn't change facts.

Back to more important things, I was guessing AMD would win the hires war this round. I wonder if it's simply the memory per card? 2GB seems a bad choice IMHO.

I have not seen any at any point...a little skeptical of your account. When I have seen them the instant I attempted to add to cart I was met with a out of stock error. I guess I wasnt around for the specific 7 hours out of the last 48 I was supposed to be to see them, that doesnt count as widely available to me.

I mean you can check for yourself right now, speaking of not very observant http://www.newegg.com/Product/Produ...315498&IsNodeId=1&name=GeForce GTX 600 series

I'm sure a few are floating in and out of stock here and there but it's hardly good availability and certainly AMD got hammered for that.

For memory, 2GB is good imo, it's enough yet will let them save a few bucks versus 3GB, and they will have 4GB versions apparently (lets hope not as scarce as say 2.5GB 570's were though).

 

[Silent_Buddha]

I have not seen any at any point...a little skeptical of your account. When I have seen them the instant I attempted to add to cart I was met with a out of stock error. I guess I wasnt around for the specific 7 hours out of the last 48 I was supposed to be to see them, that doesnt count as widely available to me.

Same thing I've been saying about 7970. All we know is that demand is higher than supply. We have no idea what supply levels are like.

I'm sure a few are floating in and out of stock here and there but it's hardly good availability and certainly AMD got hammered for that.

Unfairly hammered certainly. But same situation, we have no idea if absolute supply numbers are the same, worse, or better than it was for 7970. We have no way to even start to infer where demand is at. All we know is that demand is greater than supply. That's it. No idea what demand is like. No idea what supply is like.

Although I suppose to be fair the same people that complained about stock levels for 7970 should be complaining just as loudly about stock levels for GTX 680.

Or people can just stop trying to determine supply levels when there's Zero way to even start coming up with any kind of number for what supply is at.

Regards,
SB

 

[AlphaWolf]

I checked at 8-9am my time yesterday (utc-6) and Newegg had none. There were, however, other e-tailers with product available, but have since sold out. I've not seen one at MSRP yet. We'll see how it shakes out over the coming month.

 

[Domell]

The only thing i wonder about the newest nVIDIA GPU is why it`s highend GPU has a codename from midrange/performance GPU ( in the past GF104/114, G94 etc.).

 

[A1xLLcqAgt0qc2RyMz0y]

The only thing i wonder about the newest nVIDIA GPU is why it`s highend GPU has a codename from midrange/performance GPU ( in the past GF104/114, G94 etc.).

This has been answered many many times already.

When Nvidia was designing Kepler they had internal targets as to what they thought AMD's GCN would perform at. Once Nvidia saw the performance of the 7970 they knew that their mid-range GPU would be able to match or exceed it. It also allowed Nvidia to delay the release of the BigK GK110 until later in the year when yields and supply of 28nm capacity of TSMC increases.

"Honestly, we expected more from our competitor’s new architecture."

http://www.nordichardware.com/news/...cted-more-from-radeon-hd-7900q-exclusive.html


If you really have complaints as to why Nvidia's mid-range GPU can outperform AMD's high-range ask AMD why.

 

[Voxilla]

If you really have complaints as to why Nvidia's mid-range GPU can outperform AMD's high-range ask AMD why.

I wouldn't say it can outperform AMD in everything. Also, if AMD would for example reduce the memory bus to 256 bit, use same NV speed memory, and boost the clock to say 1.1 Ghz, it would perform better as it does, at least in games.
In situations where memory bandwidth is more important, like for compute applications, the 7970 is simply better. Where is it still better, in double precision, 1 TFLOP versus 125 GFLOP. Overall the AMD chip is better if you consider both compute and graphics. NV choose this time to deemphasize general purpose compute in favor of graphics, something the GPGPU community is less than thrilled with. For pure game graphics and as a console GPU, it is a more efficient design though.

 

[cal_guy]

I'm sure the fact that the GK110 only taped out last month didn't have anything to do with it either.

 

[A1xLLcqAgt0qc2RyMz0y]

I'm sure the fact that the GK110 only taped out last month didn't have anything to do with it either.

If you're referring to Charlie of S|A for your data then you should realize that he has been wrong many times before in regards to Nvidia's tape-outs.

 

[CarstenS]

fact = linky-linky to nvidia statement or pic-or-it-didn't-happen

 

[UniversalTruth]

I wouldn't say it can outperform AMD in everything. Also, if AMD would for example reduce the memory bus to 256 bit, use same NV speed memory, and boost the clock to say 1.1 Ghz, it would perform better as it does, at least in games.
In situations where memory bandwidth is more important, like for compute applications, the 7970 is simply better. Where is it still better, in double precision, 1 TFLOP versus 125 GFLOP. Overall the AMD chip is better if you consider both compute and graphics. NV choose this time to deemphasize general purpose compute in favor of graphics, something the GPGPU community is less than thrilled with. For pure game graphics and as a console GPU, it is a more efficient design though.

The most important advantage here for nvidia is that they have the freedom to put a very very low price for their card. I mean if they put 300$, that will hurt severely AMD and they will have to either "enjoy" very limited interest for their card, or be forced to decrease the price which means inevitable margins loss.
They are lucky that nvidia doesn't want to push them in the red.

 

[jimbo75]

The most important advantage here for nvidia is that they have the freedom to put a very very low price for their card. I mean if they put 300$, that will hurt severely AMD and they will have to either "enjoy" very limited interest for their card, or be forced to decrease the price which means inevitable margins loss.
They are lucky that nvidia doesn't want to push them in the red.

And what do you think would happen to the entire market if AMD decides to sell the 7870 for $150?

 

[UniversalTruth]

And what do you think would happen to the entire market if AMD decides to sell the 7870 for $150?

In many countries around the world the nvidia brand (respectively the interest in their products) is so much stronger that I would say "nothing will happen". Or the damage on nvidia will be very limited...

 

[Rangers]

The most important advantage here for nvidia is that they have the freedom to put a very very low price for their card. I mean if they put 300$, that will hurt severely AMD and they will have to either "enjoy" very limited interest for their card, or be forced to decrease the price which means inevitable margins loss.
They are lucky that nvidia doesn't want to push them in the red.

It isn't all that much smaller than Tahiti. Plus they've shown no interest in a price war.

Caymen 6970 was bigger than Tahiti, and AMD sold it for 350 (actually seemed closer to 300 with rebates and sales later on) with no problems.

Even if in 6 months you see something like, GK110=599, GK104=399, it wouldn't cause AMD problems. But I dont even expect to see that.

In many countries around the world the nvidia brand (respectively the interest in their products) is so much stronger that I would say "nothing will happen". Or the damage on nvidia will be very limited...

Heh, you keep thinking that. I guess that is why AMD never prices their products below it's performance area yet still sells fine? EG, 6970 was priced relative to Nvidia competition where it's performance warranted, etc for every other card?

Or by the same token, Nvidia never charges twice as much for the same performing card as AMD (or even a little more, that I see). If so many people only buy Nvidia, they could do that, simple economics.

Heck, if anything 680 price maybe says they have to undercut the competition to sell

 

[UniversalTruth]

Heh, you keep thinking that.

Yeah, most probably because I read such stories around the web... It's not my own idea if you mean that.

My understanding and guess to why AMD keeps asking such ridiculous prices is that despite some "potentially" lower prices, they don't enjoy higher interst in their products to compensate that.

 

[jimbo75]

Yeah, most probably because I read such stories around the web... It's not my own idea if you mean that.

My understanding and guess to why AMD keeps asking such ridiculous prices is that despite some "potentially" lower prices, they don't enjoy higher interst in their products to compensate that.

In that case they'd be sitting on a mountain of unsold cards. Last I checked it was Nvidia who had to rebrand the 570's to 560 Ti 448 Core because the 570's weren't selling.

 

[itsmydamnation]

Yeah, most probably because I read such stories around the web... It's not my own idea if you mean that.

My understanding and guess to why AMD keeps asking such ridiculous prices is that despite some "potentially" lower prices, they don't enjoy higher interst in their products to compensate that.

that makes zero sense, we dont have the markets interest so we will price our selves out of the market.... good one there buddy.

also looking at DX 11 GPU's on steam paints the opposite picture
http://store.steampowered.com/hwsurvey/videocard/