Even More GeForce GTX 980 and GM204 Specs Tumble Out

[xenocide]

Indeed, this is a great conversation! My take is that I won't any better off in replacing my 780Ti with a 980. Since I'm extremely happy with my 780Ti there's no good reason to replace it.

The 980 is not intended to be a replacement for 780/780 Ti owners.

 

[Ferrum Master]

Continuing reading latest nvidia patents... they seem to be on trolling rampage...

http://www.4-traders.com/NVIDIA-COR...ated-Vapor-Chamber-Published-Online-19027641/

 

[HumanSmoke]

Continuing reading latest nvidia patents... they seem to be on trolling rampage...

Talking of trolling...
1. This is a graphics card thread
2. The patent is for a specific heatpipe cooler (they also patented a fan design as well, but you probably missed that), and just about everyone has a patent for their particular implementation. Here's one of Intel's, and here's another of Nvidia's from 2007, and here's the one Asetek are using for AMD cards which is basically a full cover waterblock allied with a HDT cooler base - not overly original, but because of implementation it doesn't really affect anyone unless you try to copy it wholesale.

Back under that Latvian bridge with ya!

 

[Sony Xperia S]

4K is barely relevant as a statistic in usage, yet a significant portion of forum users across the net are whining and moaning about the lack of 5K and 8K support.

This is just a pure exaggeration, and actually not true.

Give the consumer what they want and they just want more

Again exaggeration, probably false too.

they'll whine .... they'll whine ...... they'll never use anyway

The number of people of confuse "tech enthusiast" with "hyper-critical user with unrealistic expectations" seems to grow larger every product cycle. Intel could no doubt transform their 18-core E5-2699v3 into a desktop part, but how would they continue to keep topping something like that?

Die shrinking? Optimising further the architecture?

The moment they slip up and "only" produce a 30-core next time around you have a bunch of people complaining because the rate of progress isn't to their satisfaction

Again pure exaggeration. The companies slip up very often now and what? In this regard, I think the status quo will remain the same.

Sometimes, you just have to apply real world values rather than a wish list

Sometimes, the limitations are purely in someone's imagination. Who defines what "real" world value is or not?

The entire post of yours sounds very pessimistic and sceptic.

 

[Ferrum Master]

Back under that Latvian bridge with ya!

Actually this is a graphics cards leaked specs speculation thread alias gossips. And trying to understand how really the card clocks in memory department now, dig up few post up too skullface... and last submitted patents are in chain with this delta compression method.

 

[HumanSmoke]

Give the consumer what they want and they just want more

Again exaggeration, probably false too.

Why say probably, can't you be bothered in doing any basic research?
Here, let me help - How about something from the Harvard Business Review for starters.
As for the rest, it's either evident or it isn't. Either way...

 

[Vayra86]

This is just a pure exaggeration, and actually not
true.



Again exaggeration, probably false too.







Die shrinking? Optimising further the architecture?



Again pure exaggeration. The companies slip up very often now and what? In this regard, I think the status quo will remain the same.



Sometimes, the limitations are purely in someone's imagination. Who defines what "real" world value is or not?

The entire post of yours sounds very pessimistic and sceptic.

Actually no, he sounds very sensible, contrary to the fully self-centered approach you've adopted. Every single statement he made is bang on the money and proven every single day on every single techsite comment section.

 

[Ferrum Master]

Actually no, he sounds very sensible, contrary to the fully self-centered approach you've adopted. Every single statement he made is bang on the money and proven every single day on every single techsite comment section.

first post?

 

[RCoon]

Am I the only person who looks at core count and wonders if the 9xx series has a hope of matching the 780 and 780ti?
I mean I get it, core count isn't everything, but it counts for a lot. Full fat 780ti is in excess of 2800 cores. Full fat 980 is only 2048. Forgive me for being a caveman, but I don't see how the 970 or 980 are going to have a hope in hell competing against flagship kepler gpu's. Sure they're power efficient, and might have a little performance increase from architectural improvement, but I don't see 2000 cores being better than 2800, or even that close.

If some GPU engineer would enlighten me how these cards will be remotely on equal footing, (and I mean really close, equal), then maybe the rumoured prices will make some sense to me.

 

[HumanSmoke]

Am I the only person who looks at core count and wonders if the 9xx series has a hope of matching the 780 and 780ti?
I mean I get it, core count isn't everything, but it counts for a lot. Full fat 780 is in excess of 2800 cores. Full fat 980 is only 2048. Forgive me for being a caveman, but I don't see how the 970 or 980 are going to have a hope in hell competing against flagship kepler gpu's. Sure they're power efficient, and might have a little performance increase from architectural improvement, but I don't see 2000 cores being better than 2800, or even that close.

If some GPU engineer would enlighten me how these cards will be remotely on equal footing, (and I mean really close, equal), then maybe the rumoured prices will make some sense to me.

Well, I'm no GPU engineer, but I pretty much agree with your confusion!
It looks like the GM 204's lack of cores and bandwidth are offset to a degree by a large L2 cache, ROP count (probably a little overkill for a 256-bit card), colour compression (still dubious on that one), and a huge increase in clockspeed. If rumours are right, the stock clock on the 980 is 1126M which is almost 30% higher than the 780 Ti, and the boost clock just widens the gap.
Still hard to think that the GM 204 will live with the GK 110 (and Hawaii) at high resolutions and image quality settings though. We'll find out soon enough.

 

[rtwjunkie]

@RCoon: Thats exactly what I've been saying in here too. They are taking this mid-level GM204 and charging a top-end price (if the price rumor is true), for simply giving a 980 their flagship naming convention. If they did it with the GM200/210 I could understand it...

 

[RCoon]

Well, I'm no GPU engineer, but I pretty much agree with your confusion!
It looks like the GM 204's lack of cores and bandwidth are offset to a degree by a large L2 cache, ROP count (probably a little overkill for a 256-bit card), colour compression (still dubious on that one), and a huge increase in clockspeed. If rumours are right, the stock clock on the 980 is 1126M which is almost 30% higher than the 780 Ti, and the boost clock just widens the gap.
Still hard to think that the GM 204 will live with the GK 110 (and Hawaii) at high resolutions and image quality settings though. We'll find out soon enough.

Yeah, all pretty strange to me, that's why I don't think these speculated prices are correct at all. However the colour compression worries me. I know the 285 uses it (and it works), but I'm really not a fan of my images being messed with. I don't care how amazing their compression form is, it's going to erk me.
I understand it's not meant to compete with the 780ti, but it's still about 200+ cores away from the 780 (which I assume they will offset by setting core and boost clocks so high).
I'm just going to assume they'll eventually make the die larger to fit in more cores, which will inevitably increase the heat and power, but being Maxwell it will be more efficient. One thing we don't know is just how well the Maxwell archi scales. Then there's the inevitable 970ti with 14-15 SMM's in the future.

Quite simply I don't know, I'm just sceptical about these specs in their current form, and the price makes no sense to me.

@RCoon: Thats exactly what I've been saying in here too. They are taking this mid-level GM204 and charging a top-end price (if the price rumor is true), for simply giving a 980 their flagship naming convention. If they did it with the GM200/210 I could understand it...

Well they did it with the 6xx series, why not do it again.

 

[FourtyTwo]

Am I the only person who looks at core count and wonders if the 9xx series has a hope of matching the 780 and 780ti?

A "core" is a generic expression, it doesn't indicate any kind of performance level.
The cores of the GM 204 can very well have different performance than the previous generation.

 

[rtwjunkie]

Well they did it with the 6xx series, why not do it again.

Good point!

 

[RCoon]

it doesn't indicate any kind of performance level.

By that logic 1 core = 2880 cores. It does in fact indicate performance when you're looking at two architectures built on the same 28nm process. Performance increases will be minimal at best (+-10%), further improved by the compression technique both companies seem to be adopting.

 

[Sony Xperia S]

Actually no, he sounds very sensible, contrary to the fully self-centered approach you've adopted. Every single statement he made is bang on the money and proven every single day on every single techsite comment section.

Self-centred approach goes only out of nvidia. I'm thinking about the good of customers, and on the second level, the good of the company which actually tends to adopt a very unsustainable policy with these ever growing price increases and ever declining sales.

 

[rtwjunkie]

Self-centred approach goes only out of nvidia. I'm thinking about the good of customers, and on the second level, the good of the company which actually tends to adopt a very unsustainable policy with these ever growing price increases and ever declining sales.

Declining sales, yes. But not because of any business practice. Low-end card sales for both GPU companies have nearly dried up with so many CPU's now sporting decent low-end graphics. As for whether Nvidia is hurting, no. Their prices aren't hurting the company in the least. I believe last month's sales figures for the previous quarter showed Nvidia made as much money as AMD, and AMD's figures were COMBINED GPU, CPU and APU figures. So, Nvidia will charge whatever the market will allow. As long as we buy at high prices, there is no reason for Nvidia to lower them It's unfortunate for some, but true.

 

[Ferrum Master]

I'm really not a fan of my images being messed with. I don't care how amazing their compression form is, it's going to erk me.

Well AMD at least operates still with 10bit RGB data still, geforces still use 8bit and nothing more, leaving exclusive thing for quadro cards... thus leaving the question open for full fledged Maxwell Quadro and his likeness...

The patents are way to non specific and describe various methods picking up the data mask and then stacking it in rows telling we already sent this batch and is same with that thus reducing workload... and the methods can be lossy and lossless, that depends on the will of the manufacturer... it even can be dynamic... I hope they are not so evil writing the block firmware...

So we get a obviously a tweaked Titan with Tegra features, and cut down to 680 sizes for better yield, higher clocks as we are using the same tech node, but more laid down, less leakage(thumbs up TSMC), and milking money again...

And there was a upper comment about placebo thing... peps using plain TN's 6bit frc and shouting I don't see the difference - will actually also never notice the difference just hardware wise... just my five cents.

 

[CookieMonsta]

Am I the only person who looks at core count and wonders if the 9xx series has a hope of matching the 780 and 780ti?
I mean I get it, core count isn't everything, but it counts for a lot. Full fat 780ti is in excess of 2800 cores. Full fat 980 is only 2048. Forgive me for being a caveman, but I don't see how the 970 or 980 are going to have a hope in hell competing against flagship kepler gpu's. Sure they're power efficient, and might have a little performance increase from architectural improvement, but I don't see 2000 cores being better than 2800, or even that close.

If some GPU engineer would enlighten me how these cards will be remotely on equal footing, (and I mean really close, equal), then maybe the rumoured prices will make some sense to me.

According to Anandtech, NVIDIA designed the Maxwell SMM (128 cores) to deliver 90% of the Kepler SMX (192 cores) performance at a much greater energy efficiency. From this, the Kepler SMX has 50% more cores, thus we can safely say that the Maxwell SMM (i.e. 128 Maxwell cores) delivers 40% more performance than Kepler cores.

Therefore, the 2048 Maxwell cores of the GTX 980 is equivalent 2867 Kepler cores of the same clockspeed. I.e. I'm confident the 980 will easily match or beat GK110 at least in shading power.

This massive improvement in IPC has already been seen on the Laptops. The 860M (640 Maxwell) overclocked to 1300mhz can deliver 90% of the GTX 870M (1344 cores) performance at stock.

Secondly, the large L2 cache of the Maxwell cores supposedly has a magnifying effect on the VRAM bandwidth, this is difficult to estimate but I think the figures of +50% to outright doubling on the bandwidth has been thrown around by the Beyond3D guys.

Apparently, the cost of this extra 40% IPC from the SMM units is the larger die area due to the more compartmentalized nature of Maxwell vs Kepler.

 

[GhostRyder]

Declining sales, yes. But not because of any business practice. Low-end card sales for both GPU companies have nearly dried up with so many CPU's now sporting decent low-end graphics. As for whether Nvidia is hurting, no. Their prices aren't hurting the company in the least. I believe last month's sales figures for the previous quarter showed Nvidia made as much money as AMD, and AMD's figures were COMBINED GPU, CPU and APU figures. So, Nvidia will charge whatever the market will allow. As long as we buy at high prices, there is no reason for Nvidia to lower them It's unfortunate for some, but true.

There in lies the problem in my book, it's not so much them charging the prices but the consumer buying it. We can whine and moan all we want but if we fork over the money what are we telling them ("them" being generic).

Continuing reading latest nvidia patents... they seem to be on trolling rampage...

http://www.4-traders.com/NVIDIA-COR...ated-Vapor-Chamber-Published-Online-19027641/

Well while I agree it's just another lock out patent that could be used in the wrong way it's just for their recent design on the cooler. However after reading it seems to be very vague which could be a problem in the future.

Am I the only person who looks at core count and wonders if the 9xx series has a hope of matching the 780 and 780ti?
I mean I get it, core count isn't everything, but it counts for a lot. Full fat 780ti is in excess of 2800 cores. Full fat 980 is only 2048. Forgive me for being a caveman, but I don't see how the 970 or 980 are going to have a hope in hell competing against flagship kepler gpu's. Sure they're power efficient, and might have a little performance increase from architectural improvement, but I don't see 2000 cores being better than 2800, or even that close.

If some GPU engineer would enlighten me how these cards will be remotely on equal footing, (and I mean really close, equal), then maybe the rumoured prices will make some sense to me.

Well I do my math by looking over the specs of the GTX 750ti vs the GTX 650ti. The 650ti has 768 cores at 928base clock on a 128bit bus giving 5.4gbs for memory speed while the GTX 750ti has 640 cores at 1020 (Boost of 1085) on a 128bit bus giving the same 5.4gbs for memory speed as well. Comparing the two the performance difference generally runs in the range of ~25% depending on the game of course which shows that even with a core deficit of 20% the 750ti performs 25% better (But that also has to take into account the boost clock of 1085 vs the 928 clock on the 650ti). So I guess it just comes down to if and how much better the new architecture that has evolved from the first taste of the GM architecture and its scaling with more cores on top of these extreme core clocks. As far as the architecture is concerned the person above linked to Anandtech which is what I was about to do so he beat me to it that does a decent job with explaining the new architectural changes.

Like I said I am still going to wait until we see the real deal benchmarks and everything is all out in the open.

 

[Avlin]

see this updated post on geeks3d.com :

http://www.geeks3d.com/20140915/nvi...d-gtx-980-latest-news-specs-gpu-z-msi-gaming/

 

[utengineer]

see this updated post on geeks3d.com :

http://www.geeks3d.com/20140915/nvi...d-gtx-980-latest-news-specs-gpu-z-msi-gaming/

Interesting. Looks like CompuBench has already done their performance testing on the GTX980 versus the Titan. At Screen: 3840 x 2160!

http://compubench.com/compare.jsp?b...1=cl&D2=NVIDIA+GeForce+GTX+TITAN+Black&cols=2

 

[D007]

Looks like I'm not upgrading til next release then.

Interesting. Looks like CompuBench has already done their performance testing on the GTX980 versus the Titan. At Screen: 3840 x 2160!

http://compubench.com/compare.jsp?benchmark=compu20&D1=NVIDIA GeForce GTX 980&os1=Windows&api1=cl&D2=NVIDIA GeForce GTX TITAN Black&cols=2

Unless those specs are right.
4k looks like a big jump on this series.

 

[utengineer]

Unless those specs are right.
4k looks like a big jump on this series.

.

I mentioned this before and I am not a GPU guru, but the GTX980 64 ROP should benefit 4k+ users. The Titan only has 48.

 

[GhostRyder]

.

I mentioned this before and I am not a GPU guru, but the GTX980 64 ROP should benefit 4k+ users. The Titan only has 48.

Yes and that is going to make a nice difference for this series of cards on top of the extra 1gb over the previous generation. I really like these cards already more and more especially once I saw the 3 DP on it. Its almost tempting me...