Overclock.net › Forums › Industry News › Hardware News › [wccftech] GeForce 800 Series Arrives in Second Half of 2014
New Posts  All Forums:Forum Nav:

[wccftech] GeForce 800 Series Arrives in Second Half of 2014 - Page 52

post #511 of 1139
Quote:
Originally Posted by Majin SSJ Eric View Post

Where the hell does Forbes come up with that nonsense? There is no way on earth that Nvidia is going to be releasing a replacement for the 780Ti in March (just 5 months after launching the 780Ti). None. My guess is that we will see a repeat of the 680 with a smallish-die GM104 release as the 880 during the holiday season (probably Oct/Nov) and a large die GM110 held back for a second generation Maxwell release well into 2015. Just a guess but I reiterate that there is no way that we see a GTX 880 this month. That is just utter nonsense...

I don't think the higher end Maxwells will launch before late summer (at the very earliest), however replacing the 780Ti after 5 months wouldn't be unprecedented.

480 -> 580 = 8 months
9800GTX -> GTX 280 = 2.5 months
GTX 280 -> GTX 285 = 6 months

etc.

It's probably going to be a small die first however I don't think we should rule out a big die release just because GK100 was a failure... Other than that all other big die NVs have been released...

They are going to want it out as fast as possible since the HPC crowd is always screaming for more.
 
Benching
(17 items)
 
 
CPUMotherboardGraphicsRAM
[i7 5960X @ 4.8GHz] [Rampage V Extreme] [Titan 1400MHz (1500MHz bench)] [Various] 
Hard DriveCoolingCoolingCooling
[250GB 840EVO +2x SpinpointF3 1TB RAID0] [LD PC-V2 SS Phase Change] [XSPC X2O 750 pump/res] [Monsta 360 full copper + EK XT 360 + XT 240] 
MonitorPowerCaseAudio
[Crossover 27Q LED-P 1440p+ASUS 1200p+LG 1080p] [Corsair AX1200] [Dimastech Easy v3.0] [Sennheiser HD558s] 
CPUCPUMotherboardGraphics
FX 8320, FX 8350, Phenom II x2 555BE i7 3930K, i7 860, i7 4770K, 68x Celeron D CVF, commando, 2x RIVE, Z87X-OC Asus 4870x2, Sapphire 4870 
GraphicsGraphicsGraphicsGraphics
2x 5870, 5850, 5830, 5770 2x 3870x2, 3870 GTX Titan, GTX 480, GTX 590 GTX 285, GTX 260, 4x 9800GT, 8800GTX 
RAMHard DriveCoolingCooling
4x4GB vengeance, 2x4GB predatorX, 2x1GB OCZ DDR2 Intel X25-M 80GB LD PC-V2 SS Phase Change OCN Marksman 
CoolingCoolingOSPower
2x old tek slims (GPU) Various watercooling stuff win7, winxp AX1200 
Case
test bench / cardboard box 
  hide details  
Reply
 
Benching
(17 items)
 
 
CPUMotherboardGraphicsRAM
[i7 5960X @ 4.8GHz] [Rampage V Extreme] [Titan 1400MHz (1500MHz bench)] [Various] 
Hard DriveCoolingCoolingCooling
[250GB 840EVO +2x SpinpointF3 1TB RAID0] [LD PC-V2 SS Phase Change] [XSPC X2O 750 pump/res] [Monsta 360 full copper + EK XT 360 + XT 240] 
MonitorPowerCaseAudio
[Crossover 27Q LED-P 1440p+ASUS 1200p+LG 1080p] [Corsair AX1200] [Dimastech Easy v3.0] [Sennheiser HD558s] 
CPUCPUMotherboardGraphics
FX 8320, FX 8350, Phenom II x2 555BE i7 3930K, i7 860, i7 4770K, 68x Celeron D CVF, commando, 2x RIVE, Z87X-OC Asus 4870x2, Sapphire 4870 
GraphicsGraphicsGraphicsGraphics
2x 5870, 5850, 5830, 5770 2x 3870x2, 3870 GTX Titan, GTX 480, GTX 590 GTX 285, GTX 260, 4x 9800GT, 8800GTX 
RAMHard DriveCoolingCooling
4x4GB vengeance, 2x4GB predatorX, 2x1GB OCZ DDR2 Intel X25-M 80GB LD PC-V2 SS Phase Change OCN Marksman 
CoolingCoolingOSPower
2x old tek slims (GPU) Various watercooling stuff win7, winxp AX1200 
Case
test bench / cardboard box 
  hide details  
Reply
post #512 of 1139
Good point Al. I just don't see there being any need for Nvidia to rush things at this point...
post #513 of 1139
Quote:
Originally Posted by Majin SSJ Eric View Post

Good point Al. I just don't see there being any need for Nvidia to rush things at this point...

Yeah, as I said... Wouldn't rule it out. I agree there's no immediate need buuuut a lot of good reasons exist timing wise for them to try to get it out ASAP.
The Sig Rig
(16 items)
 
  
CPUMotherboardGraphicsRAM
i7 2600K @ 4.xghz MSI Z68A-GD65 (G3) MSI GTX 970 Gaming SLI @ 1504mhz core/8020mhz m... 16GB DDR3 2x 8gb @ 1866mhz C8 
Hard DriveHard DriveOptical DriveCooling
240GB Sandisk Extreme A few 2tb hdd's. DVDRW Corsair H75 CLC on the CPU 
OSMonitorKeyboardPower
Windows 8.1 Pro 64-bit ACER B326hk 4k 60hz 3840x2160 IPS sst display Coolermaster Cherry MX Brown switch mechanical ... Corsair TX 950w 
CaseMouseMouse PadAudio
Fractal Design Define R3 Black Pearl Logitech MX518 EverGlide Titan "Medium" Soundblaster Z PCIE 
  hide details  
Reply
The Sig Rig
(16 items)
 
  
CPUMotherboardGraphicsRAM
i7 2600K @ 4.xghz MSI Z68A-GD65 (G3) MSI GTX 970 Gaming SLI @ 1504mhz core/8020mhz m... 16GB DDR3 2x 8gb @ 1866mhz C8 
Hard DriveHard DriveOptical DriveCooling
240GB Sandisk Extreme A few 2tb hdd's. DVDRW Corsair H75 CLC on the CPU 
OSMonitorKeyboardPower
Windows 8.1 Pro 64-bit ACER B326hk 4k 60hz 3840x2160 IPS sst display Coolermaster Cherry MX Brown switch mechanical ... Corsair TX 950w 
CaseMouseMouse PadAudio
Fractal Design Define R3 Black Pearl Logitech MX518 EverGlide Titan "Medium" Soundblaster Z PCIE 
  hide details  
Reply
post #514 of 1139
Quote:
Originally Posted by Alatar View Post

I don't think the higher end Maxwells will launch before late summer (at the very earliest), however replacing the 780Ti after 5 months wouldn't be unprecedented.

480 -> 580 = 8 months
9800GTX -> GTX 280 = 2.5 months
GTX 280 -> GTX 285 = 6 months

etc.

TSMC stopped half nodes after 55nm . So the GTX 280 (65 nm) - > GTX 285 (55 nm) comparison does not hold now. Moreover Nvidia was late with GF100 Fermi (GTX 480). If they had been on time the gap would be the usual 1 year. TSMC 20nm is in volume production this quarter. Apple is most probably getting the maximum allocation at this leading node. So Nvidia, Qualcomm, AMD will have to wait a bit longer to get enough wafers for a full production run. The earliest we can expect GM104 is Aug-Sep . latest would be Nov-Dec. I think late Q3 / early Q4 is most likely.
Quote:
It's probably going to be a small die first however I don't think we should rule out a big die release just because GK100 was a failure... Other than that all other big die NVs have been released...

They are going to want it out as fast as possible since the HPC crowd is always screaming for more.

A massive 550 sq mm die on an immature bleeding edge node with yield problems is not going to happen. Nvidia learnt that lesson hard with GF100. The smart thing is the Kepler way. A mid range die at 300 - 330 sq mm which can be manufactured at acceptable yields and sold at flagship prices. The problem is the flagship prices have been steadily increasing in both camps.

GTX 480 / GTX 580- 500
GTX 680 - 500
GTX 780 - 650
GTX 780Ti - 700

HD 5870 - 380
HD 6970 - 370
HD 7970 - 550
R9 290X - 550 (but 700 in US due to retailer price gouging and mining craze)

GM104 aka GTX 880 can easily be sold at USD 700 if it comes with a 40% increase in perf over GTX 780 Ti. AMD will most probably launch later this gen as they are most likely to get HBM (high bandwidth memory) out on their 20nm flagship. AMD too can sell at USD 650 - 700 if they match the GTX 880. Enthusiasts have to get used to these ever increasing prices if they always want the best perf money can buy.
Edited by raghu78 - 3/1/14 at 11:02pm
Fragbox
(14 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 2400 DH67BL AMD Radeon HD 6900 Series XFX HD 6950 2GB 
RAMRAMHard DriveOptical Drive
Corsair Corsair DDR3 1333 Mhz 2 GB Corsair DDR3 1333 Mhz 2 GB Western Digital Caviar Green SONY DVD-RW AD-7260S 
OSMonitorKeyboardPower
Windows 7 Professional 32 bit BENQ G2420HD Logitech K200 Seasonic VX550 psu 
CaseMouse
ANTEC 200 V2 Logitech mouse 
  hide details  
Reply
Fragbox
(14 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 2400 DH67BL AMD Radeon HD 6900 Series XFX HD 6950 2GB 
RAMRAMHard DriveOptical Drive
Corsair Corsair DDR3 1333 Mhz 2 GB Corsair DDR3 1333 Mhz 2 GB Western Digital Caviar Green SONY DVD-RW AD-7260S 
OSMonitorKeyboardPower
Windows 7 Professional 32 bit BENQ G2420HD Logitech K200 Seasonic VX550 psu 
CaseMouse
ANTEC 200 V2 Logitech mouse 
  hide details  
Reply
post #515 of 1139
meh...I will just wait for Volta with stacked VRAM. My GTX 560 still gets 60 FPS completely maxed out on Grid 2.
post #516 of 1139
Quote:
Originally Posted by raghu78 View Post

TSMC stopped half nodes after 55nm . So the GTX 280 - > GTX 285 comparison does not hold now. Moreover Nvidia was late with GF100 Fermi (GTX 480). If they had been on time the gap would be the usual 1 year. TSMC 20nm is in volume production this quarter. Apple is most probably getting the maximum allocation at this leading node. So Nvidia, Qualcomm, AMD will have to wait a bit longer to get enough wafers for a full production run. The earliest we can expect GM104 is Aug-Sep . latest would be Nov-Dec. I think late Q3 / early Q4 is most likely.

That part of the post wasn't about current process tech, or the circumstances surrounding examples I mentioned. I was just replying to Eric who seemed to imply that Nvidia would be opposed to replacing the previous flagship so soon from a business point of view.
Quote:
A massive 550 sq mm die on an immature bleeding edge node with yield problems is not going to happen. Nvidia learnt that lesson hard with GF100. The smart thing is the Kepler way.

This mostly just depends on when they want to tape out the higher end designs and how many respins / steppings they need to get it working like they want.

If tape out is early enough and everything goes really well with the very first stepping then the parts might be out pretty quickly.

For example we know that GK110 taped out in early 2012. What would have happened if GK100 made it to market?

It's not quite as simple as "no big die on a new process anymore after fermi". I'm not saying it will happen or that it wont happen. But there are tons of things to consider with that one, especially if you compare to fermi as you need to take the 40nm problems into account. And since GK100 apparently was a thing before it was canceled I don't think NV's plans for Kepler were originally meant to be any different from their Fermi approach.

This time with 20nm you need to consider:

-NV has had training with Maxwell already (GM107)
-What problems does NV have with 20nm at this point in time (if any)
-etc.

Lots of small things go into the final strategy for bringing out the different parts to the market. And at this point it's a pure guessing game.
Quote:
A mid range die at 300 - 330 sq mm which can be manufactured at acceptable yields and sold at flagship prices.The problem is the flagship prices have been steadily increasing in both camps.

GTX 480 / GTX 580- 500
GTX 680 - 500
GTX 780 - 650
GTX 780Ti - 700

HD 5870 - 380
HD 6970 - 370
HD 7970 - 550
R9 290X - 550 (but 700 in US due to retailer price gouging and mining craze)

No argument with this one. The problem really was the 7970/680 generation. Those two designs took a die usually sold for the $250-$350 price range and made it $500+. That's where the real jump happened.

The later cards after that were mostly reactionary for both camps. Hawaii maybe being the only one at correct prices (well at least at MSRP) compared to the original small AMD dies. Problem is that it came way too late, has supply issues and gets overpriced because of that.
Quote:
GM104 aka GTX 880 can easily be sold at USD 700 if it comes with a 40% increase in perf over GTX 780 Ti. AMD will most probably launch later this gen as they are most likely to get HBM (high bandwidth memory) out on the first 20nm flagship. AMD too can sell at USD 650 - 700 if they match the GTX 880. Enthusiasts have to get used to these ever increasing prices if they always want the best perf money can buy.

I personally wont pay any more than $500 max for a sub 350mm^2 die.

I know that 20nm is supposed to be pretty expensive but the price hike already happened with 28nm. I'm hoping that we've already reached the point where upping the price of the high end would only cause higher margins with lower sales, lower revenue and lower actual profit.

The high margins are actually only good as long as you can sell everything you can manufacture.

Because that's what it all boils down to in the end. Making the most money you can. They need to find a proper price to move X volume of cards.

The only thing I know is that I wont be buying a GM204 GTX 880 at $700.
Edited by Alatar - 3/1/14 at 11:32pm
 
Benching
(17 items)
 
 
CPUMotherboardGraphicsRAM
[i7 5960X @ 4.8GHz] [Rampage V Extreme] [Titan 1400MHz (1500MHz bench)] [Various] 
Hard DriveCoolingCoolingCooling
[250GB 840EVO +2x SpinpointF3 1TB RAID0] [LD PC-V2 SS Phase Change] [XSPC X2O 750 pump/res] [Monsta 360 full copper + EK XT 360 + XT 240] 
MonitorPowerCaseAudio
[Crossover 27Q LED-P 1440p+ASUS 1200p+LG 1080p] [Corsair AX1200] [Dimastech Easy v3.0] [Sennheiser HD558s] 
CPUCPUMotherboardGraphics
FX 8320, FX 8350, Phenom II x2 555BE i7 3930K, i7 860, i7 4770K, 68x Celeron D CVF, commando, 2x RIVE, Z87X-OC Asus 4870x2, Sapphire 4870 
GraphicsGraphicsGraphicsGraphics
2x 5870, 5850, 5830, 5770 2x 3870x2, 3870 GTX Titan, GTX 480, GTX 590 GTX 285, GTX 260, 4x 9800GT, 8800GTX 
RAMHard DriveCoolingCooling
4x4GB vengeance, 2x4GB predatorX, 2x1GB OCZ DDR2 Intel X25-M 80GB LD PC-V2 SS Phase Change OCN Marksman 
CoolingCoolingOSPower
2x old tek slims (GPU) Various watercooling stuff win7, winxp AX1200 
Case
test bench / cardboard box 
  hide details  
Reply
 
Benching
(17 items)
 
 
CPUMotherboardGraphicsRAM
[i7 5960X @ 4.8GHz] [Rampage V Extreme] [Titan 1400MHz (1500MHz bench)] [Various] 
Hard DriveCoolingCoolingCooling
[250GB 840EVO +2x SpinpointF3 1TB RAID0] [LD PC-V2 SS Phase Change] [XSPC X2O 750 pump/res] [Monsta 360 full copper + EK XT 360 + XT 240] 
MonitorPowerCaseAudio
[Crossover 27Q LED-P 1440p+ASUS 1200p+LG 1080p] [Corsair AX1200] [Dimastech Easy v3.0] [Sennheiser HD558s] 
CPUCPUMotherboardGraphics
FX 8320, FX 8350, Phenom II x2 555BE i7 3930K, i7 860, i7 4770K, 68x Celeron D CVF, commando, 2x RIVE, Z87X-OC Asus 4870x2, Sapphire 4870 
GraphicsGraphicsGraphicsGraphics
2x 5870, 5850, 5830, 5770 2x 3870x2, 3870 GTX Titan, GTX 480, GTX 590 GTX 285, GTX 260, 4x 9800GT, 8800GTX 
RAMHard DriveCoolingCooling
4x4GB vengeance, 2x4GB predatorX, 2x1GB OCZ DDR2 Intel X25-M 80GB LD PC-V2 SS Phase Change OCN Marksman 
CoolingCoolingOSPower
2x old tek slims (GPU) Various watercooling stuff win7, winxp AX1200 
Case
test bench / cardboard box 
  hide details  
Reply
post #517 of 1139
Quote:
Originally Posted by Alatar View Post

This mostly just depends on when they want to tape out the higher end designs and how many respins / steppings they need to get it working like they want. If tape out is early enough and everything goes really well with the very first stepping then the parts might be out pretty quickly. For example we know that GK110 taped out in early 2012. What would have happened if GK100 made it to market?

GK100 was simply not feasible as Nvidia had yield issues even with a 294 sq mm chip. 28nm initial yields were a problem in H1 2012 and subsequently got better in H2 2012. Nvidia made record revenue / profits / margins in H2 2012.

http://www.xbitlabs.com/news/graphics/display/20120216125300_Nvidia_Kepler_Yields_Lower_Than_Expected_Chief_Executive_Officer.html
https://nvidianews.nvidia.com/Releases/NVIDIA-Reports-Financial-Results-for-Third-Quarter-Fiscal-Year-2013-8b3.aspx

"We are ramping our Kepler generation very hard, and we could use more wafers. The gross margin decline is contributed almost entirely to the yields of 28nm being lower than expected. That is, I guess, unsurprising at this point,” said Jen-Hsun Huang, chief executive officer of Nvidia, during a conference call with financial analysts.

The question is about manufacturability / yields at the bleeding edge especially with higher wafer costs due to added process steps like dual patterning immersion lithography at 20nm, completely new 3D FINFET transistor device at foundries 16nm/14nm and subsequent yield struggles due to added process steps and increased process complexity. Even Intel is making the transition to dual patterning immersion litho only at 14nm and you can already see their struggles. Nvidia has been publicly vocal about the increase in cost per transistor at 20 nm planar ( dual patterning immersion litho) and 16FF (FINFET 3D transistor device). The cost per transistor is a multi-variable function of yield(t) , scaling factor and wafer cost. yield(t) which is the yield over time has been a problem at all leading edge foundry nodes from 40nm. Nvidia was again vocal about yield struggles at 40nm and 28nm at the very early stage of the process ramp. TSMC is the best of the foundries with a 1- 2 year lead at the leading node. The rest like Samsung and GF are well behind TSMC and have even greater yield problems.

http://www.altera.com/technology/system-design/articles/2012/20nm-systems-era.html
http://www.extremetech.com/computing/123529-nvidia-deeply-unhappy-with-tsmc-claims-22nm-essentially-worthless

ARM also acknowledges the complete lack of cost savings per transistor (in fact even a slight increase in cost) at TSMC 20nm / TSMC 16FF

http://www.extremetech.com/computing/176422-arm-announces-cortex-a17-the-new-cpu-core-that-finally-replaces-cortex-a9
Quote:
It's not quite as simple as "no big die on a new process anymore after fermi". I'm not saying it will happen or that it wont happen. But there are tons of things to consider with that one, especially if you compare to fermi as you need to take the 40nm problems into account. And since GK100 apparently was a thing before it was canceled I don't think NV's plans for Kepler were originally meant to be any different from their Fermi approach.

This time with 20nm you need to consider:

-NV has had training with Maxwell already (GM107)
-What problems does NV have with 20nm at this point in time (if any)
-etc.

TSMC 20nm is costlier on a per transistor cost till Q1 2015 according to the graph from Nvidia's presentation. this means a GM104 even with the same transistor count as GK110 will be costlier even though it will be of a much smaller die size. So do not be surprised if the GM104 chip launches at USD 700 - 800.
Quote:
I know that 20nm is supposed to be pretty expensive but the price hike already happened with 28nm. I'm hoping that we've already reached the point where upping the price of the high end would only cause higher margins with lower sales, lower revenue and lower actual profit. The high margins are actually only good as long as you can sell everything you can manufacture. Because that's what it all boils down to in the end. Making the most money you can. They need to find a proper price to move X volume of cards. The only thing I know is that I wont be buying a GM204 GTX 880 at $700.

Here again wafer allocation and process yield controls how many chips you can manufacture. With Apple getting the highest priority/wafer allocation at TSMC 20nm , Nvidia is going to be supply constrained and also yield constrained. So if Nvidia is confident of selling every GM104 chip they can manufacture at a given price of USD 700 - 800 they will do so. Same logic as Titan. can you sell the volume of chips at the stated price ? If yes then Nvidia will sell at the best possible price. As far as you not buying a GM104 at $700 , Nvidia should not have a problem finding more hardcore enthusiasts who are ready to give their money for bragging rights thumb.gif
Edited by raghu78 - 3/2/14 at 12:21am
Fragbox
(14 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 2400 DH67BL AMD Radeon HD 6900 Series XFX HD 6950 2GB 
RAMRAMHard DriveOptical Drive
Corsair Corsair DDR3 1333 Mhz 2 GB Corsair DDR3 1333 Mhz 2 GB Western Digital Caviar Green SONY DVD-RW AD-7260S 
OSMonitorKeyboardPower
Windows 7 Professional 32 bit BENQ G2420HD Logitech K200 Seasonic VX550 psu 
CaseMouse
ANTEC 200 V2 Logitech mouse 
  hide details  
Reply
Fragbox
(14 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 2400 DH67BL AMD Radeon HD 6900 Series XFX HD 6950 2GB 
RAMRAMHard DriveOptical Drive
Corsair Corsair DDR3 1333 Mhz 2 GB Corsair DDR3 1333 Mhz 2 GB Western Digital Caviar Green SONY DVD-RW AD-7260S 
OSMonitorKeyboardPower
Windows 7 Professional 32 bit BENQ G2420HD Logitech K200 Seasonic VX550 psu 
CaseMouse
ANTEC 200 V2 Logitech mouse 
  hide details  
Reply
post #518 of 1139
Quote:
Originally Posted by raghu78 View Post

The question is about manufacturability at the bleeding edge especially with higher wafer costs due to added process steps like dual patterning immersion lithography at 20nm, completely new 3D FINFET transistor device at foundries 16nm/14nm and subsequent yield struggles due to added process steps and increased process complexity. Even Intel is making the transition to dual patterning immersion litho only at 14nm and you can already see their struggles. Nvidia has been publicly vocal about the increase in cost per transistor at 20 nm planar ( dual patterning immersion litho) and 16FF (FINFET 3D transistor device). The cost per transistor is a multi-variable function of yield(t) , scaling factor and wafer cost. yield(t) which is the yield over time has been a problem at all leading edge foundry nodes from 40nm. Nvidia was again vocal about yield struggles at 40nm and 28nm at the very early stage of the process ramp. TSMC is the best of the foundries with a 1- 2 year lead at the leading node. The rest like Samsung and GF are well behind TSMC and have even greater yield problems.

http://www.altera.com/technology/system-design/articles/2012/20nm-systems-era.html
http://www.extremetech.com/computing/123529-nvidia-deeply-unhappy-with-tsmc-claims-22nm-essentially-worthless

ARM also acknowledges the complete lack of cost savings per transistor (in fact even a slight increase in cost) at TSMC 20nm / TSMC 16FF

http://www.extremetech.com/computing/176422-arm-announces-cortex-a17-the-new-cpu-core-that-finally-replaces-cortex-a9
TSMC 20nm is costlier on a per transistor cost till Q1 2015 according to the graph from Nvidia's presentation. this means a GM104 even with the same transistor count as GK110 will be costlier even though it will be of a much smaller die size. So do not be surprised if the GM104 chip launches at USD 700 - 800.

-The same slides also show 40nm being more expensive than 55nm back when fermi launched.
-The 20nm and 14nm* numbers are pure predictions
-The article is 2 years old and the slides are probably even older than that

*the slide saying 14nm shows you how old the thing is, there are no 14nm TSMC processes so the slide was made before even Nvidia knew what TSMC was going to do after 20nm.

What I'm trying to say here (and it seems like I'm not quite getting through) is that the circumstances are different from the last few times Nvidia made a process switch, and the actual info we have on 20nm is either hearsay or hopelessly outdated. It might be accurate at the moment or it might not be accurate.

We don't know how wafer cost estimates have changed since early 2012/ late 2011 when the slides were made, we don't know how the yields will be for individual companies and their designs

Which is why I'm saying it's almost pure guesswork at this stage unless you have insider information.
Quote:
Here again wafer allocation and process yield controls how many chips you can manufacture. With Apple getting the highest priority/wafer allocation at TSMC 20nm , Nvidia is going to be supply constrained and also yield constrained. So if Nvidia is confident of selling every GM104 chip they can manufacture at a given price of USD 700 - 800 they will do so. Same logic as Titan. can you sell the volume of chips at the stated price ? If yes then Nvidia will sell at the best possible price. As far as you not buying a GM104 at $700 , Nvidia should not have a problem finding more hardcore enthusiasts who are ready to give their money for bragging rights thumb.gif

But are they going to be supply constrained? That's the thing your argument relies on.

The supply constraint again relies on a lot of things.

-You can fit more GM204 per wafer than GK110
-GM204 would not yield at the same rate as GK110 (on a mature process it would yield better but we don't know how the yields would be at 20nm)
-How many wafers would be allocated on 20nm to GM204 compared to the GK110 wafer allocation on 28nm (28nm has tegra, GK110, GK104, GK106, GK107, GK108, GM107 etc. while beginning 20nm might only have one design being produced)
-how many chips can they produce before launch

And after that you have to take into account things like Nvidia often being OK with short supply shortages on launch, (GK104 springs to mind).

So in the end question of can they sell all of them at $700-$800 is a really problematic one because we barely know anything about the supply.

What we do however know is that only very small amounts (compared to other GPUs) of 780Tis and Titans sell. Just look at any top sellers list out there.
 
Benching
(17 items)
 
 
CPUMotherboardGraphicsRAM
[i7 5960X @ 4.8GHz] [Rampage V Extreme] [Titan 1400MHz (1500MHz bench)] [Various] 
Hard DriveCoolingCoolingCooling
[250GB 840EVO +2x SpinpointF3 1TB RAID0] [LD PC-V2 SS Phase Change] [XSPC X2O 750 pump/res] [Monsta 360 full copper + EK XT 360 + XT 240] 
MonitorPowerCaseAudio
[Crossover 27Q LED-P 1440p+ASUS 1200p+LG 1080p] [Corsair AX1200] [Dimastech Easy v3.0] [Sennheiser HD558s] 
CPUCPUMotherboardGraphics
FX 8320, FX 8350, Phenom II x2 555BE i7 3930K, i7 860, i7 4770K, 68x Celeron D CVF, commando, 2x RIVE, Z87X-OC Asus 4870x2, Sapphire 4870 
GraphicsGraphicsGraphicsGraphics
2x 5870, 5850, 5830, 5770 2x 3870x2, 3870 GTX Titan, GTX 480, GTX 590 GTX 285, GTX 260, 4x 9800GT, 8800GTX 
RAMHard DriveCoolingCooling
4x4GB vengeance, 2x4GB predatorX, 2x1GB OCZ DDR2 Intel X25-M 80GB LD PC-V2 SS Phase Change OCN Marksman 
CoolingCoolingOSPower
2x old tek slims (GPU) Various watercooling stuff win7, winxp AX1200 
Case
test bench / cardboard box 
  hide details  
Reply
 
Benching
(17 items)
 
 
CPUMotherboardGraphicsRAM
[i7 5960X @ 4.8GHz] [Rampage V Extreme] [Titan 1400MHz (1500MHz bench)] [Various] 
Hard DriveCoolingCoolingCooling
[250GB 840EVO +2x SpinpointF3 1TB RAID0] [LD PC-V2 SS Phase Change] [XSPC X2O 750 pump/res] [Monsta 360 full copper + EK XT 360 + XT 240] 
MonitorPowerCaseAudio
[Crossover 27Q LED-P 1440p+ASUS 1200p+LG 1080p] [Corsair AX1200] [Dimastech Easy v3.0] [Sennheiser HD558s] 
CPUCPUMotherboardGraphics
FX 8320, FX 8350, Phenom II x2 555BE i7 3930K, i7 860, i7 4770K, 68x Celeron D CVF, commando, 2x RIVE, Z87X-OC Asus 4870x2, Sapphire 4870 
GraphicsGraphicsGraphicsGraphics
2x 5870, 5850, 5830, 5770 2x 3870x2, 3870 GTX Titan, GTX 480, GTX 590 GTX 285, GTX 260, 4x 9800GT, 8800GTX 
RAMHard DriveCoolingCooling
4x4GB vengeance, 2x4GB predatorX, 2x1GB OCZ DDR2 Intel X25-M 80GB LD PC-V2 SS Phase Change OCN Marksman 
CoolingCoolingOSPower
2x old tek slims (GPU) Various watercooling stuff win7, winxp AX1200 
Case
test bench / cardboard box 
  hide details  
Reply
post #519 of 1139
Quote:
Originally Posted by Alatar View Post

-The same slides also show 40nm being more expensive than 55nm back when fermi launched.

Than its no surprise that GTX 285 was $380 at launch and GTX 480 was $500 at launch.
Quote:
-The 20nm and 14nm* numbers are pure predictions
-The article is 2 years old and the slides are probably even older than that
the nvidia and arm presentation talk about

*the slide saying 14nm shows you how old the thing is, there are no 14nm TSMC processes so the slide was made before even Nvidia knew what TSMC was going to do after 20nm.

What I'm trying to say here (and it seems like I'm not quite getting through) is that the circumstances are different from the last few times Nvidia made a process switch, and the actual info we have on 20nm is either hearsay or hopelessly outdated. It might be accurate at the moment or it might not be accurate. We don't know how wafer cost estimates have changed since early 2012/ late 2011 when the slides were made, we don't know how the yields will be for individual companies and their designs.

http://www.anandtech.com/show/7739/arm-cortex-a17

Nvdia's presentation is quite old. But the ARM A17 presentation which has the graph for Number of transistors / $ had an embargo date of Feb 11, 2014. they mention 16nm / 14nm because ARM products will be manufactured by TSMC ,Samsung and GF. Also TSMC 16FF , Samsung 14nm ,GF 14XM are of the same generation with a 20nm BEOL and a 3D FINFET transistor device. the transistor cost info from ARM presentation is very recent and quite reflective of what the fabless design industry is seeing now in 2014 and anticipate for 2015 wrt yields, wafer cost and transistor cost / $ .

Quote:
Which is why I'm saying it's almost pure guesswork at this stage unless you have insider information. But are they going to be supply constrained? That's the thing your argument relies on. The supply constraint again relies on a lot of things.

-You can fit more GM204 per wafer than GK110
-GM204 would not yield at the same rate as GK110 (on a mature process it would yield better but we don't know how the yields would be at 20nm)

Please understand that Nvidia / ARM talk about Number of transistors per $ or cost per transistor after taking into account the fact that 20nm process has a 1.8x transistor density increase over 28nm. But increased wafer costs and lower yields still make 20nm not better than 28nm on a transistors / $ basis.
Quote:
How many wafers would be allocated on 20nm to GM204 compared to the GK110 wafer allocation on 28nm (28nm has tegra, GK110, GK104, GK106, GK107, GK108, GM107 etc. while beginning 20nm might only have one design being produced) . how many chips can they produce before launch

The question is how many 20nm wafers get allocated to Nvidia given the clout Apple has . Apple is going to be sucking dry TSMC 20nm with the flagship 20nm Apple A8 which will drive tens of millions of ipads and iphones in Q3 / Q4 2014. Add to that Qualcomm 20nm modems, maybe even next gen 20nm flagship SOCs and then Nvidia / AMD 20nm flagship GPUs.

http://www.qualcomm.com/media/releases/2014/02/24/qualcomm-announces-worlds-first-commercial-20-nm-lte-advanced-chipset
Quote:
And after that you have to take into account things like Nvidia often being OK with short supply shortages on launch, (GK104 springs to mind). So in the end question of can they sell all of them at $700-$800 is a really problematic one because we barely know anything about the supply. What we do however know is that only very small amounts (compared to other GPUs) of 780Tis and Titans sell. Just look at any top sellers list out there.

Nvidia can sell a GTX 780 Ti at USD 700 fine enough with 28nm being available in plenty with rock solid yields and excellent cost. So it stands to reason that Nvidia will have no issues selling a 20nm GTX 880 which is 40% faster than GTX 780 Ti especially with supply constraints and yield challenges.
Edited by raghu78 - 3/2/14 at 1:29am
Fragbox
(14 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 2400 DH67BL AMD Radeon HD 6900 Series XFX HD 6950 2GB 
RAMRAMHard DriveOptical Drive
Corsair Corsair DDR3 1333 Mhz 2 GB Corsair DDR3 1333 Mhz 2 GB Western Digital Caviar Green SONY DVD-RW AD-7260S 
OSMonitorKeyboardPower
Windows 7 Professional 32 bit BENQ G2420HD Logitech K200 Seasonic VX550 psu 
CaseMouse
ANTEC 200 V2 Logitech mouse 
  hide details  
Reply
Fragbox
(14 items)
 
  
CPUMotherboardGraphicsGraphics
Intel Core i5 2400 DH67BL AMD Radeon HD 6900 Series XFX HD 6950 2GB 
RAMRAMHard DriveOptical Drive
Corsair Corsair DDR3 1333 Mhz 2 GB Corsair DDR3 1333 Mhz 2 GB Western Digital Caviar Green SONY DVD-RW AD-7260S 
OSMonitorKeyboardPower
Windows 7 Professional 32 bit BENQ G2420HD Logitech K200 Seasonic VX550 psu 
CaseMouse
ANTEC 200 V2 Logitech mouse 
  hide details  
Reply
post #520 of 1139
Looking forward to see what the 880 has to offer.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [wccftech] GeForce 800 Series Arrives in Second Half of 2014