Overclock.net › Forums › Industry News › Hardware News › [Guru3D]Nvidia Slide reveals numbers on Single and Double precision for Flagship Pascal GPU
New Posts  All Forums:Forum Nav:

[Guru3D]Nvidia Slide reveals numbers on Single and Double precision for Flagship Pascal GPU - Page 14

post #131 of 168
Quote:
Originally Posted by iLeakStuff View Post

To you people that think Pascal will be atleast 2x as fast as Maxwell. Here is some food for thought (from Nvidia`s presentation at GDC 2015).
Compare Maxwell > Pascal to the jump from Fermi to Kepler where we saw 85% better performance with GPUs.

Don`t be surprised if the performance jump isnt as big as you thought and Pascal is much about introducing stacked DRAM and NVLink



What zealord said. And fine with me. If Pascal doesn't hit my performance targets then I'll simply look to Polaris or not upgrade.
post #132 of 168
Quote:
Originally Posted by VeerK View Post

30% performance increase for the $500 mark? Damn, if the 1080 shows up in June RIP 980ti
I think they will stop making the 980ti a little before release, just like the 780ti and the 980/970 release. So RIP indeed.
If they don't that will either be weird, or the 1080 will be a low power potato for gaming laptops.
I hope it won't.
My monitor isn't very portable.
post #133 of 168
If GP104 is indeed 30% faster than 980 Ti, comes with 8GB GDDR5X and has no competition from AMD, I guarantee it won't go for just $500, guaranteed.
post #134 of 168
It might.
I had mixed feelings for a while when I recommended the about to be released 970 to a friend at work over the 770. I couldn't believe a card just as good could be sold for half the price of a 780ti, especially when mine still seemed pretty new.
Seems like that just happened. My 780tis still work great too.
post #135 of 168
All the 3.5GB jokes aside (actually it has some merit, as otherwise having a fully functional 4GB would've drove up the cost due to better binning needed),

1. 970 was on a very mature 28nm process, so all the savings in fab costs have been reaped. Beyond 28nm cost per transistor goes UP instead, so even from a materials cost perspective it'll be more expensive.
2. 970 was clearly intended to squeeze AMD out of the market make the 290X and 290 irrelevant, which is also why AMD had a firesale on those cards when 970 launched.
3. But most importantly, 970 is just on par with 780 Ti, not 30% faster. Even the 980 after all the driver optimizations ended up being about 15% faster. And guess how much it sold for? $550 at launch, $500 after 980 Ti dropped. wink.gif

So yeah, a card that's 30% faster than 980 Ti and comes with more vram is not going to be sold for $500 period, unless AMD pulls another HD 4870 level of upset. nVidia is a business and if they can increase their margins they will. They already have 80% of the market, it makes 0 business sense for them to launch the hypothetical 1080 mentioned above for $500 as they'd just be pissing away money.
Edited by magnek - 2/21/16 at 2:24pm
post #136 of 168
Ok you've got me. Maybe $550, and that's just because it won't be much faster than 980ti oc'd too.
Nvidia still gets money when the getting is good, and if better cards are just 6 months away (or less if amd doesn't totally mess up) people sitting on their wallets is going to cost them more than a reduced markup. They have been making a lot of money by moving a lot of cards. Right now they have the best name in almost every regard for discrete gpus. If amd unloads a properly working ~17B transistor gcn flagship with some of the same efficiency improvements nvidia came up with maxwell, nvidia won't have that. If new games don't really need new gpus they lose that selling point too.
There's a lot of people with their hands by their wallets looking for a reason. Nvidia will find one. If performance improvements are so so, deals can be persuasive.
IMHO
post #137 of 168
Quote:
Originally Posted by Serandur View Post

You ballparked based on demonstrably false calculations of architectural improvements (not realizing that a 60% power reduction from process node alone is directly equivalent to a 2.5x increase in perf/watt)

False. The 2.5x figure does not correlate to the 60% power reduction figure by Samsung.

AMD stated that this figure was based on 70% 14LPP improvements and 30% architectueral improvements.

If we take 1x to mean fury from 2.5x that leaves us with 1.5x. What is 30% of 1.5x? 0.45. 45% boost. My 0.5 figure was not far off the mark.

Also 55-60% equates to up to 2.1x-2.2x, not 2.5x. 50% is 2x. If we look at it this way it's a 30% architectural boost. Your math is wrong wink.gif

So yeah...an up to 45% architectural boost from Fiji to Polaris is what we're looking at.

http://www.pcper.com/reviews/Graphics-Cards/AMD-Radeon-Technologies-Group-Previews-Polaris-Architecture

https://forum.beyond3d.com/threads/amd-polaris-arctic-islands-r-4-series-speculation-rumors-and-discussion.56719/page-24

Ps nice rant but you posted all that for nothing.
Edited by Mahigan - 2/24/16 at 9:06am
Kn0wledge
(20 items)
 
Pati3nce
(14 items)
 
Wisd0m
(10 items)
 
Reply
Kn0wledge
(20 items)
 
Pati3nce
(14 items)
 
Wisd0m
(10 items)
 
Reply
post #138 of 168
Quote:
Originally Posted by Mahigan View Post

False. The 2.5x figure does not correlate to the 60% power reduction figure by Samsung.

AMD stated that this figure was based on 70% 14LPP improvements and 30% architectueral improvements.

If we take 1x to mean fury from 2.5x that leaves us with 1.5x. What is 30% of 1.5x? 0.45. 45% boost. My 0.5 figure was not far off the mark.

Also 55-60% equates to up to 2.1x-2.2x, not 2.5x. 50% is 2x. If we look at it this way it's a 30% architectural boost. Your math is wrong wink.gif

So yeah...an up to 45% architectural boost from Fiji to Polaris is what we're looking at.

http://www.pcper.com/reviews/Graphics-Cards/AMD-Radeon-Technologies-Group-Previews-Polaris-Architecture

https://forum.beyond3d.com/threads/amd-polaris-arctic-islands-r-4-series-speculation-rumors-and-discussion.56719/page-24

Ps nice rant but you posted all that for nothing.
Do you have a serious problem with basic math or something? It's extremely simple.

100% - 60% = 40%

100/40 = 2.5



Pretend that a given 28nm GPU performs at 10,000 GFLOPs and consumes 300 watts. Now pretend it's shrunk down to an equivalently-performing 14nm one with a 60% power reduction.

0.60 x 300 = 180

Now subtract that 60% (180) from 300 watts to get a 60% reduction and you get 120 watts. Both GPUs are performing at 10,000 GFLOPS: one uses 300 watts, one uses 120 watts. Performance per watt is calculated as performance divided by watts. So


GPU 1 - 10,000 GFLOPS / 300 watts = 33.33 GFLOPS/watt


GPU 2 - 10,000 GFLOPS / 120 watts = 83.33 GFLOPS/watt

Now for how they compare

GPU 2/GPU 1 = 83.33 / 33.33 = 2.5x performance/watt


Nice try, but you fail at math.
Ice-nine
(17 items)
 
  
CPUMotherboardGraphicsRAM
i7-6700K Asus Z170-DELUXE Gigabyte G1 GTX 980 Ti Corsair Vengeance LPX DDR4 3000 
Hard DriveHard DriveOptical DriveCooling
Samsung 850 Evo 500 GB Seagate Barracuda 7200 RPM 2 TB Asus DVD Drive DRW-24B1ST/BLK/B/AS Phanteks PH-TC14PE 
OSMonitorKeyboardPower
Windows 8.1 Pro AOC Q2770PQU Corsair K95 RGB - Cherry MX Blue EVGA Supernova G2 750W 
CaseMouseMouse PadAudio
NZXT Phantom 410 Gunmetal Edition Logitech G502 Proteus Core Razer Goliathus Speed Edition Small Cyber Acoustics Satellite CA-3602 
Audio
Oppo PM-3 Closed Back Planar Magnetic Headphones 
  hide details  
Reply
Ice-nine
(17 items)
 
  
CPUMotherboardGraphicsRAM
i7-6700K Asus Z170-DELUXE Gigabyte G1 GTX 980 Ti Corsair Vengeance LPX DDR4 3000 
Hard DriveHard DriveOptical DriveCooling
Samsung 850 Evo 500 GB Seagate Barracuda 7200 RPM 2 TB Asus DVD Drive DRW-24B1ST/BLK/B/AS Phanteks PH-TC14PE 
OSMonitorKeyboardPower
Windows 8.1 Pro AOC Q2770PQU Corsair K95 RGB - Cherry MX Blue EVGA Supernova G2 750W 
CaseMouseMouse PadAudio
NZXT Phantom 410 Gunmetal Edition Logitech G502 Proteus Core Razer Goliathus Speed Edition Small Cyber Acoustics Satellite CA-3602 
Audio
Oppo PM-3 Closed Back Planar Magnetic Headphones 
  hide details  
Reply
post #139 of 168
Quote:
Originally Posted by Serandur View Post

Do you have a serious problem with basic math or something? It's extremely simple.

100% - 60% = 40%

100/40 = 2.5



Pretend that a given 28nm GPU performs at 10,000 GFLOPs and consumes 300 watts. Now pretend it's shrunk down to an equivalently-performing 14nm one with a 60% power reduction.

0.60 x 300 = 180

Now subtract that 60% (180) from 300 watts to get a 60% reduction and you get 120 watts. Both GPUs are performing at 10,000 GFLOPS: one uses 300 watts, one uses 120 watts. Performance per watt is calculated as performance divided by watts. So


GPU 1 - 10,000 GFLOPS / 300 watts = 33.33 GFLOPS/watt


GPU 2 - 10,000 GFLOPS / 120 watts = 83.33 GFLOPS/watt

Now for how they compare

GPU 2/GPU 1 = 83.33 / 33.33 = 2.5x performance/watt


Nice try, but you fail at math.

No, you're not taking into consideration what AMD stated about Polaris.

It is up to 60% power savings but this assumes the same clock speeds from GCN3 to Polaris.

AMD stated that their figure is 70% 14LPP and 30% architectural improvements. I've linked the article where they say this.

You're calculating based on less information than I am hence why your results are incorrect.
Quote:
Even RTG staff were willing to admit that the move to 14nm FinFET process tech was the majority factor for the improvement we are seeing here, something on the order of a 70/30 split

Edited by Mahigan - 2/24/16 at 12:16pm
Kn0wledge
(20 items)
 
Pati3nce
(14 items)
 
Wisd0m
(10 items)
 
Reply
Kn0wledge
(20 items)
 
Pati3nce
(14 items)
 
Wisd0m
(10 items)
 
Reply
post #140 of 168
If GP104 will come in june-july, probably in march 2017, GP100, first probably titan pascal, then geforce GP100 flavor ?

Also, what ppl discuss now its kinda plausible. The same boring perf. increase. GP104 +30-40% over 980 Ti (and way more over 980 vanilla, since GP104 is his successor). So if 980 ti is kinda 30-40% over 980, GP104 should be like 80% over 980 vanilla ? (stock nvidia clocks. vs stock nv clocks) ?

And in vram, 8gb ddr5x for 1080 and 6gb for 1070 ? smile.gif Could be nice ?

And amd's r9 series the exact same, performance vs. nvidia, and prices. With the same 10% perf. +/- between amd's and nvidia's cards.


And when GP100 comes out, if we compare it to his successor, 980 ti GM200, could be like more than 100% performance increase ? smile.gif
PC
(15 items)
 
  
CPUMotherboardGraphicsRAM
Intel® Core™ i7 3770K @4.5Ghz (Multiplier OC) 1... MSI Z77A-GD65 [10.10 BIOS] MSi GTX 980 GAMING + MX4 @1544mhz/1.262v; +450M... Geil Black Dragon 4x4Gb 1866Mhz [XMP] 
Hard DriveCoolingOSMonitor
Samsung 850 EVO 500Gb [AHCI] Noctua NH-D14 Windows 10 Home 64-Bit 10586.420 (FPP/retail key) BenQ XL2411T [Dual-Link DVI, 144Hz] AUO M24HW01... 
KeyboardPowerCaseMouse
Razer BlackWidow Ultimate 2014 [green switches] Enermax Revolution 87+ 750W and APC Back-UPS 950VA Phanteks Evolv Atx Tempered Glass Black + ATX 2... Razer Mamba 2015 Tournament Edition 
Mouse PadAudio
SteelSeries QcK Mini Asus Xonar DGX 
  hide details  
Reply
PC
(15 items)
 
  
CPUMotherboardGraphicsRAM
Intel® Core™ i7 3770K @4.5Ghz (Multiplier OC) 1... MSI Z77A-GD65 [10.10 BIOS] MSi GTX 980 GAMING + MX4 @1544mhz/1.262v; +450M... Geil Black Dragon 4x4Gb 1866Mhz [XMP] 
Hard DriveCoolingOSMonitor
Samsung 850 EVO 500Gb [AHCI] Noctua NH-D14 Windows 10 Home 64-Bit 10586.420 (FPP/retail key) BenQ XL2411T [Dual-Link DVI, 144Hz] AUO M24HW01... 
KeyboardPowerCaseMouse
Razer BlackWidow Ultimate 2014 [green switches] Enermax Revolution 87+ 750W and APC Back-UPS 950VA Phanteks Evolv Atx Tempered Glass Black + ATX 2... Razer Mamba 2015 Tournament Edition 
Mouse PadAudio
SteelSeries QcK Mini Asus Xonar DGX 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Guru3D]Nvidia Slide reveals numbers on Single and Double precision for Flagship Pascal GPU