First rumours GK110(GeForce 700) - Page 5

Noisiv · **Memory**: 4GB OCZ 5-4-4-4-12

Yup. 50% over 680 would be my guess

Something liek

850MHz GPU (boost), 1500MHz DDR5(6000 Mbps), 250W TDP

Lavans · 01-21-2013, 21:16 | posts: 4,540 | Location: Austin

Quote:

Originally Posted by Noisiv

Yup. 50% over 680 would be my guess

No way. Not even on new architectures do you see a performance increase that high.

Noisiv · **Memory**: 4GB OCZ 5-4-4-4-12

Quote:

Originally Posted by Lavans

No way. Not even on new architectures do you see a performance increase that high.

Compared to GTX 680

14/8 times more Cuda Cores

+

850/1050 less MHz

=

14/8 * 850/1050 = 42%

Consider more mature TSMC process, empirical fact that more "workers" operating at low freq.(GK110) should be more TDP friendly than less workers(shaders) at higher freq.(GTX 680),

then optimization or two here and there,
(it's already known that GK110 has some new soup like Dynamic Parallelism, and Hyper-Q),

and 50% more bandwidth...

50% is rather realistic IMHO. TDP is crucial bit here.

Lavans · *Last edited by Lavans; 01-21-2013 at 22:08.*

In the past there have been numerous new architectures, which has offered similar differences in specs, that have failed to give more than 35% performance boost. Calculating theoretical percentages based off specs is a poor way of estimating the actual jump in performance. Also, Cuda cores don't count for much in games, and not many games use them. I'm really looking forward to what Hyper-Q can do in games/apps that DO use Cuda, but I'm not holding my breath.

Maybe, when the next generation of consoles come out, and the graphical standard is raised, we'll see a 50% increase. But ATM, games just aren't optimized enough, nor do they take enough advantage of the hardware we're currently running them with. Heck, even with my GTX680, there's a number of games that I'm playing at QFHD and 2xAA + FXAA with little issue.

Noisiv · **Memory**: 4GB OCZ 5-4-4-4-12

http://www.computerbase.de/artikel/g...e-8800-gtx/28/

That's healthy +50% across the board ^^

Lavans, GK110 is same arch. (albeit with a twist) and almost twice the size of your GK104.

Calculating theoreticals from specs is how theoreticals should be calculated.
Estimating performance is more of a guess-work, but it's pretty straight forward in this case

Lavans · 01-21-2013, 22:25 | posts: 4,540 | Location: Austin

Meh, I'll remain skeptical at this point. People estimated the GK104 to be more powerful than what it really is. The issue is two fold - 1) games don't take advantage of the hardware and 2) AMD/Nvidia sees that games are failing to utilize their products so they end up cutting back on the final product. Nvidia themselves stated that this is the exact reason why they cut back on some of the GK104 features, such as direct compute. Also, looking at the difference in performance between the GTX680 and HD7970, bus bandwidth doesn't count for much at 1920x1080.

---TK--- · 01-21-2013, 22:37 | posts: 14,935 | Location: New Jersey, USA

I been seeing 40% or so from a 680. I`m sold

Lavans · 01-21-2013, 22:41 | posts: 4,540 | Location: Austin

40% at what price range? That's another thing to keep in mind

It's possible to see a performance jump that high, but you could expect it to be a repeat of the 8800GTX, being priced at around $700.

Spets · 01-21-2013, 22:51 | posts: 1,444

Can't wait to see some benchmarks from this beast. I hope the price won't be any higher than $600, but I'll still be picking one up. Wonder how high it'll clock on air.

@TK, Our cards are finally being released

---TK--- · 01-21-2013, 22:56 | posts: 14,935 | Location: New Jersey, USA

Quote:

Originally Posted by Spets

Can't wait to see some benchmarks from this beast. I hope the price won't be any higher than $600, but I'll still be picking one up. Wonder how high it'll clock on air.

@TK, Our cards are finally being released

was planning on 2 at launch, but defo getting just 1 if the price is any indication $899

cowie · 01-21-2013, 23:44 | posts: 10,949 | Location: new jersey

I would hope it would stay inline price wise with other top end card launches of the past.

Matt26LFC · 01-22-2013, 00:50 | posts: 2,302 | Location: UK

Is this actual fact, or is it just another rumor been released? Tbh unless Nvidia themselves have come out with something I'm hardly going to be getting excited about it. Also 50% performance increase, not likely.

8800GTX was a long time ago, pretty sure we've not seen a performance leap like that since then. If Nvidia did release another card with a performance increase like that I'd buy it and then not buy another card for a few years and most likely nor would anyone else! See where I'm going with that.

abollo · 01-23-2013, 23:15 | posts: 27

tsunami231 · 01-24-2013, 00:54 | posts: 2,368 | Location: USA

Comeback for 384bit memory bus Can only hope lower end cards use it too or at lest 256bit. All rumores till such time they are not.

yasamoka · *Last edited by yasamoka; 01-24-2013 at 01:27.*

BTW, just a small calculation:

a 384-bit memory bus running at an effective 6200MHz would yield around 290GB/s of memory bandwidth. Just check out the 7970 Ghz Edition running 6000MHz effective on a 384-bit memory bus. It gets 288GB/s.

Wonder where that 250GB/s came from. There isn't even ECC error correction on consumer graphics lines to warrant any reduction in theoretical memory bandwidth.

HeavyHemi · 01-24-2013, 01:39 | posts: 3,111 | Location: Wooing whilst wearing only socks.

Quote:

Originally Posted by yasamoka

BTW, just a small calculation:

a 384-bit memory bus running at an effective 6200MHz would yield around 290GB/s of memory bandwidth. Just check out the 7970 Ghz Edition running 6000MHz effective on a 384-bit memory bus.

Wonder where that 250GB/s came from. There isn't even ECC error correction on consumer graphics lines to warrant any reduction in theoretical memory bandwidth.

GDDR5 implements error correction via spec. JESD212 covers this.

phatbx133 · 01-24-2013, 01:40 | posts: 706

They better be good driver out of box.

Agent-A01 · 01-24-2013, 01:54 | posts: 5,706 | Location: USA

Quote:

Originally Posted by phatbx133

They better be good driver out of box.

Ill take it a week earlier with no driver if it means i can just hold it for a while lol

yasamoka · 01-24-2013, 02:05 | posts: 1,161 | Location: Lebanon

Quote:

Originally Posted by HeavyHemi

GDDR5 implements error correction via spec. JESD212 covers this.

Yes, but why would we have 2 cards that use GDDR5 to spec, where one has lower memory frequencies and higher bandwidth, and one has higher memory frequencies but lower bandwidth over the same memory bus width?

ESlik · 01-24-2013, 05:54 | posts: 1,666 | Location: Sault Ste. Marie Ont. CA

I have been gaming for over 12 years. In that 12 years,I have purchased whatever the newest and best card was. From the Radeon 9800 right up to the 680 GTX I have now. Never, did any of the cards give a performance boost of anywhere near 50%. It just ain't gonna happen. Period!

Agent-A01 · 01-24-2013, 06:42 | posts: 5,706 | Location: USA

Quote:

Originally Posted by ESlik

I have been gaming for over 12 years. In that 12 years,I have purchased whatever the newest and best card was. From the Radeon 9800 right up to the 680 GTX I have now. Never, did any of the cards give a performance boost of anywhere near 50%. It just ain't gonna happen. Period!

a 480 is more than 40% faster than 285. next

SLI-756 · 01-24-2013, 07:05 | posts: 3,233 | Location: Sunny Scotland

Quote:

Originally Posted by Agent-A01

Ill take it a week earlier with no driver if it means i can just hold it [up in front of your faces] for a while lol

epeen yeah?

HeavyHemi · 01-24-2013, 07:46 | posts: 3,111 | Location: Wooing whilst wearing only socks.

Quote:

Originally Posted by yasamoka

Yes, but why would we have 2 cards that use GDDR5 to spec, where one has lower memory frequencies and higher bandwidth, and one has higher memory frequencies but lower bandwidth over the same memory bus width?

Depends upon what optimization or lack of they may be using in the controller.

Lavans · *Last edited by Lavans; 01-24-2013 at 08:27.*

What's the difference in performance between the Tesla K10 and K20? It seems that comparing those two cards will give us a rough idea of what to expect, best case scenario. Though I can't seem to find any decent benchmarks that compare the two in a workstation environment.

Lane · *Last edited by Lane; 01-24-2013 at 09:04.*

Quote:

Originally Posted by Lavans

What's the difference in performance between the Tesla K10 and K20? It seems that comparing those two cards will give us a rough idea of what to expect, best case scenario. Though I can't seem to find any decent benchmarks that compare the two in a workstation environment.

Hardly, because the clock are so low, The 680/670 have a 1.5x ( 50% ) advantage over the Tesla one... should be more easy with 2x 670 ( 2688 SP ) and look core speed difference.

The K20x and K20 due to low clock speed dont have a really impressive SP performance ( 3.52 and 3.95 Tflops, for give you an example the GTX680 have around 3.1Tflops (~ 30% Tflops performance ). The K10 peak at 4.52Tflops with ( 2x GK104 ).

Now The K10 is not good at DP, but the K20x and K20 are really good at this ( 1/3 DP rate ) .

the difference on gaming will more likely be the ROP, TMU, number of SP, core speed.. ( memory controller on some cases ( high resolution. ) and ofc without knowing the clock speed. and be sure of the number of SP, speculation is a bit hard.