next gen kepler to support dx 11.1, also take a year to rollout all cards

[Ninja Prime]

Originally Posted by lee63

I read it....and I cant believe where it came from or is he being sarcastic ??????

http://semiaccurate.com/2012/01/19/n...-clear-winner/

One of his usual jokes, I bet hes got a NV marketing slide with all the bars ten times has high as they should be, with a bunch of random benches/metrics where they win lol.

[Viral]

It could be relative to performance and expected price. If GK104 is within 5-10% of 7970 and costs $350, nvidia will officially have pretty much a 4870 vs GTX 280 scenario, only reversed. That would be pretty hilarious if true

[shadow001]

Originally Posted by Viral

It could be relative to performance and expected price. If GK104 is within 5-10% of 7970 and costs $350, nvidia will officially have pretty much a 4870 vs GTX 280 scenario, only reversed. That would be pretty hilarious if true

Given that it only has a 256 bit memory bus, it'll have a serious disadvantage in memory bandwith compared to the HD7970 for one, and you can only use 8 memory modules( each uses a 32 bit interface) and the card packs at 2 GB or ram onboard, or 4 GB if modules with twice the density exist.(i'm not sure they do).


Then there's the GK104 designation wich as i mentioned in a previous post is usually reserved for midrange products, and the current fermi version(GF104) is used the GTX560 cards right now, so i could see this GK104 version probably get pretty close to the performance of a GTX580, and for 300$ it would be a no brainer really...


Mind you i'm assuming that the information from charlie is accurate as his hatred for Nvidia in general is pretty much legendary at this point....

[Ninja Prime]

Originally Posted by Viral

It could be relative to performance and expected price. If GK104 is within 5-10% of 7970 and costs $350, nvidia will officially have pretty much a 4870 vs GTX 280 scenario, only reversed. That would be pretty hilarious if true

That would be something, but that would require JHH to swallow his pride, better odds of the sun turning blue.

[shadow001]

Saw this just now

Geforce GTX 660 / Ti

Nvidia's next-generation 28nm GPU architecture, codenamed Kepler, is officially expected to launch in early Q2 2012 according to the latest schedule we have seen from the company. Although the company will technically be launching one quarter behind AMD, we can still expect to see a few noteworthy design wins in the enthusiast and performance segments, beginning with GK110 in April and following up with GK104, GK107 and GK108.

The latest report from sources overseas claims that Nvidia's GK104 GPU, the successor to 40nm GF114 (see: Geforce GTX 560, Geforce GTX 560 Ti), will feature a 256-bit memory interface, will pack 2GB of memory capacity and should have a 225W TDP. Nevertheless, the GK104 will most likely be branded as Geforce GTX 660 and may very well have a Geforce GTX 660 Ti variant later down the release roadmap.

Many analysts are expecting 28nm GK104 to get a big boost in CUDA cores. Some estimates are projecting as many as 768 CUDA cores or "well above 2 teraflops" of raw performance, which roughly equates to 50-percent more compute power than the current flagship Geforce GTX 580 single-GPU card with its 1.56 teraflops. While we don't have exact information on these specifications yet, we expect more information on the exact details of CUDA cores, texture units and ROPs in the very near future.

http://www.fudzilla.com/home/item/25...specs-revealed


Looks like the GK104 is going to be a solid midrange card to replace the GTX560 series, but even if it packs more shaders and increases it's single precision math ability to 2+ terraflops, the HD7970 has 3.7 terraflops, and the 256 bit memory bus also means less memory(2 GB) and less memory bandwith too, so this is no direct competitor to the HD7970....


That leaves the GK110 wich is a dual GPU card according to rumors, but in any case the cards will only start showing up in april so it's still 3 months out...

[Logical]

Most of the rumours are still stating that the Memory bus for GK104 will be 256 bit or 384 bit. What is interesting is that there are rumours that we might see this GPU as early as Febuary.

[Redeemed]

Originally Posted by shadow001

Saw this just now





http://www.fudzilla.com/home/item/25...specs-revealed


Looks like the GK104 is going to be a solid midrange card to replace the GTX560 series, but even if it packs more shaders and increases it's single precision math ability to 2+ terraflops, the HD7970 has 3.7 terraflops, and the 256 bit memory bus also means less memory(2 GB) and less memory bandwith too, so this is no direct competitor to the HD7970....


That leaves the GK110 wich is a dual GPU card according to rumors, but in any case the cards will only start showing up in april so it's still 3 months out...

I must confess I'm not too savy on nVidia's codenames. However, if I'm not mistaken the "GTX660" will be slower than the 7970. Which means the "GTX680" will either be as fast, or just slightly faster. That only leaves the "GTX690" but that will have to compete with the 7990.

If all of that is true, then I wonder how tRollo will implement his damage control? The 7970 is a pretty decent increase over the 6970. Not so much over the GTX580. If the GTX680 is as fast or just barely faster than the 7970 how will it be an upgrade over a GTX580 or a 7970? And wouldn't that be less of a performance increase over previous gen than what the 7970 brought? Thinking about it, it just doesn't seem to make sense. I expect the GTX680 to lead the 7970 just as the GTX580 lead the 6970. Maybe even moreso.

[shadow001]

Originally Posted by Redeemed

I must confess I'm not too savy on nVidia's codenames. However, if I'm not mistaken the "GTX660" will be slower than the 7970. Which means the "GTX680" will either be as fast, or just slightly faster. That only leaves the "GTX690" but that will have to compete with the 7990.

If all of that is true, then I wonder how tRollo will implement his damage control? The 7970 is a pretty decent increase over the 6970. Not so much over the GTX580. If the GTX680 is as fast or just barely faster than the 7970 how will it be an upgrade over a GTX580 or a 7970? And wouldn't that be less of a performance increase over previous gen than what the 7970 brought? Thinking about it, it just doesn't seem to make sense. I expect the GTX680 to lead the 7970 just as the GTX580 lead the 6970. Maybe even moreso.

Well put it this way, even when JHH did the presentation of the future roadmap for Nvidia over the next few years, the charts used primarily focused of single and dual precision floating point math, such as this:





Disregading the timetable itself as kepler didn't make it in 2011, we just see dual precision floating point math as the main metric wich isn't related to gaming at all but important for cuda and GP-GPU in general.....So the main point is that a carefull balance will have to be struck between gaming and GP-GPU performance, as that's where they sell the Quadro and tesla versions of the cards, wich are priced far higher than the gaming version using the same GPU and make Nvidia far more money for every card sold.


So when it comes to gaming, i do believe it'll be faster than the HD7970, but only enough to beat it by some margin without compromising too much on GP-GPU abilities, wich if you're wondering, the HD7970 also got a serious boost there as it's GCN architecture allows it to have 1.2 terraflops dual precision and 3.7 terraflops single precision math....On dual precision alone it's twice as much as the GF110 found in the GTX580.


On the chart we see 5 Gigaflops per watt for Kepler as the target, so assuming the usual 300 watt limit for PCI-e express power limit for single GPU cards, it'll give kepler 1.5 terraflops dual precision math ability if the entire 300 watts was dedicated to the GPU alone(wich it can't be as there's other stuff that needs power like the memory) ....No idea for single precision since it isn't listed.



Basically it's not just about gaming anymore....

[shadow001]

Here's a quote from the anandtech review and why GP-GPU is getting so much attention


http://www.anandtech.com/show/5261/a...-7970-review/2....Bottom of the page.

So long as you only want to handle the highly parallel workloads VLIW is fine. But using VLIW as the basis of a compute architecture is going is limit what tasks your processor is sufficiently good at. If you want to handle a wider spectrum of compute workloads you need a more general purpose architecture, and this is the situation AMD faced.

But why does AMD want to chase compute in the first place when they already have a successful graphics GPU business? In the long term GCN plays a big part in AMD’s Fusion plans, but in the short term there’s a much simpler answer: because they have to.

In Q3’2011 NVIDIA’s Professional Solutions Business (Quadro + Tesla) had an operating income of 95M on 230M in revenue. Their (consumer) GPU business had an operating income of 146M, but on a much larger 644M in revenue. Professional products have much higher profit margins and it’s a growing business, particularly the GPU computing side. As it stands NVIDIA and AMD may have relatively equal shares of the discrete GPU market, but it’s NVIDIA that makes all the money. For AMD’s GPU business it’s no longer enough to focus only on graphics, they need a larger piece of the professional product market to survive and thrive in the future. And thus we have GCN.

Basically gaming isn't enough anymore and AMD wants of piece of that compute market, wich is highly profitable, hence why the GCN shader architecture was developed and they didn't stick with VLIW as in previous AMD GPU's...

[K007]

http://www.youtube.com/watch?v=ioZNQcHHF8E&t=2m0s

This ones for tRollo i think.

[Rollo]

Originally Posted by K007

http://www.youtube.com/watch?v=ioZNQcHHF8E&t=2m0s

This ones for tRollo i think.

I had no idea you are a YouTube "reviewer" K007. Nice hat!

[Ninja Prime]

Originally Posted by shadow001

On the chart we see 5 Gigaflops per watt for Kepler as the target, so assuming the usual 300 watt limit for PCI-e express power limit for single GPU cards, it'll give kepler 1.5 terraflops dual precision math ability if the entire 300 watts was dedicated to the GPU alone(wich it can't be as there's other stuff that needs power like the memory) ....No idea for single precision since it isn't listed.

To clarify this, they are judging against Tesla supercomputing boards. The first fermi computing board has a DP flops rating of 515 DP flops and a TDP of 238 watts. This amounts to 2.16 DP flops per watt. To hit their claimed 2.5 times this, they would hit 5.42 DP flops per watt. So, 5.42 x 238 watts = 1.28 teraflops DP, and since NVs DP rate is half their SP rate, SP flops would be 2.56 teraflops SP. Doesn't even beat the 6970(2.7teraflops), let alone the 7970(3.79 teraflops), at similar power.

It also depends on what TDP they are targeting for this 2.5x number. As lower end tends to be more efficent, maybe this 2.5x number only applies to a 200 watt product, that still puts them at 1.08 teraflops DP, still in line with the rumored "over 1 teraflop DP" number. I suspect this might be the case, and their high end will be a dual GPU board product.