Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [WCCF] Nvidia's Volta could launch as early as 2H 2017 ON 16nm node
New Posts  All Forums:Forum Nav:

[WCCF] Nvidia's Volta could launch as early as 2H 2017 ON 16nm node - Page 13

post #121 of 272
Quote:
Originally Posted by Asmodian View Post

Yes, but you want to spend less money while Nvidia wants you to spend more money. Lots of releases with just enough of an improvement would make more money.

I wouldn't be surprised to see a 16nm Volta next year.

Except you don't need a new architecture(and all the extra R&D cost associated with that) to achieve "Lots of releases with just enough of an improvement" with all the current Pascal chips on hand for 2017.

Nvidia made plenty of $$ from reusing GTX 680's GK104 chip for GTX 770 a year later(and 670's GK104 in GTX 760), rushing out a GM204 in 2013 would have costed them hundred of millions more in R&D and made selling GK110 at $650/700 impossible. The end result would be less, not more profit.

That's where the "rush Volta out next year" logic fails in its business model. What would maximize NV's revenue when Vega comes out is something similiar what they did in 2013:

  • 1080 price drop to $400-500 as 1170
  • Titan X 2016 rebranded as 1180, takes the $650-$700 price point
  • 1070 price drop to 1060's price point and become 1160
  • Full core Titan X Black succeed to the $1200 price point

Rushing out a 16nm GV104 as X80 Volta that's probably faster than GP102(just like GM204 was faster than GK110) would make no sense and kill the 1080Ti/Titan X Black market for no comparable benefit to offset the extra R&D cost. What Volta could do is show up in late late 2017 or early 2018 with a GV107 small chip as the 1150/1150Ti.
Edited by sherlock - 7/27/16 at 10:44am
Skylake 1080 FTW
(18 items)
 
XPS 15
(7 items)
 
CPUMotherboardGraphicsRAM
i7-6700K  ASUS Maximus VIII Ranger EVGA 1080 FTW TridentZ DDR4-3000 
Hard DriveHard DriveHard DriveHard Drive
Samsung SM951 128GB Samsung 840 Pro 256 GB Muskin Reactor 1TB Seagate Baracuda 2TB ST2000DM001 HDD 
Hard DriveCoolingOSMonitor
Seagate Baracuda 2TB ST2000DM001 HDD Corsair H110i GT Windows 10 Pro  Acer XB321HK 32" 4K G-sync 
KeyboardPowerCaseMouse
Corsair K70 Rapid Fire EVGA SuperNova 650 P2 NZXT H440 Black Logitech G900  
Mouse PadOther
Razer Vespula CyberPower CP1500PFCLCD - PFC Sinewave UPS Syst... 
CPUMotherboardGraphicsRAM
i7-7700HQ 2.8Ghz(3.8Ghz Turbo) Dell XPS 15-9560 Nvidia GTX 1050 8GB DDR4-2400 
Hard DriveOSMonitor
256GB PCIE SSD Windows 10 Pro 15.6" 1080p 
  hide details  
Reply
Skylake 1080 FTW
(18 items)
 
XPS 15
(7 items)
 
CPUMotherboardGraphicsRAM
i7-6700K  ASUS Maximus VIII Ranger EVGA 1080 FTW TridentZ DDR4-3000 
Hard DriveHard DriveHard DriveHard Drive
Samsung SM951 128GB Samsung 840 Pro 256 GB Muskin Reactor 1TB Seagate Baracuda 2TB ST2000DM001 HDD 
Hard DriveCoolingOSMonitor
Seagate Baracuda 2TB ST2000DM001 HDD Corsair H110i GT Windows 10 Pro  Acer XB321HK 32" 4K G-sync 
KeyboardPowerCaseMouse
Corsair K70 Rapid Fire EVGA SuperNova 650 P2 NZXT H440 Black Logitech G900  
Mouse PadOther
Razer Vespula CyberPower CP1500PFCLCD - PFC Sinewave UPS Syst... 
CPUMotherboardGraphicsRAM
i7-7700HQ 2.8Ghz(3.8Ghz Turbo) Dell XPS 15-9560 Nvidia GTX 1050 8GB DDR4-2400 
Hard DriveOSMonitor
256GB PCIE SSD Windows 10 Pro 15.6" 1080p 
  hide details  
Reply
post #122 of 272
Quote:
Originally Posted by ToTheSun! View Post

Let's just agree that it's very refined clump of moving electrons.

We are all Pascal GPUs, and share common bonds. thumb.gif

Pun intended.
post #123 of 272
Quote:
Originally Posted by sherlock View Post

Nvidia made plenty of $$ from reusing GTX 680's GK104 chip for GTX 770 a year later, rushing out a GM204 in 2013 would have costed them hundred of millions more in R&D and made selling GK110 at $650/700 impossible. The end result would be less, not more profit.

That's where the "rush Volta out next year" logic fails. What would get Maximize NV's revenue when Vega comes out is exactly what they did in late Nov 2013:

  • 1080 price drop to $400-500
  • Titan X 2016 rebranded as 1080Ti, takes the $650-$700 price point
  • Full core Titan X Black succeed to the $1200 price point

Rushing out a 16nm GV104 as X80 Volta that's probably than GP102(just like GM204 was faster than GK110) would make no sense and kill the 1080Ti/Titan X Black market for no comparable benefit to offset the extra R&D cost.

Nvidia is always doing R&D on new chips, you cannot lay all those people off after they finish a new design and you do not hire a new team to design each architecture. R&D costs are not spiking with each new architecture. If Volta is done, $0 is saved by not releasing it. Worrying about R&D money spent on designing Pascal is the sunk cost fallacy, that money has been spent and nothing you do now will change that.

The reason for the 780Ti and Titan Black was because Nvidia did not have another architecture ready, not because they had Maxwell ready to go but were just waiting for the right time. If Nvidia had Maxwell ready when the 780Ti was released you can be sure they would have released the Titan X then instead of the Titan Black.

Edit: Also, more time selling Pascal is less time selling Volta, people who already bought Pascal might buy Volta and Volta doesn't cost them more to make. Why not sell Volta as soon as it is ready?
Edited by Asmodian - 7/27/16 at 10:58am
Desktop
(19 items)
 
RAID
(17 items)
 
 
CPUMotherboardGraphicsRAM
Intel i9-7900X @ 4.7GHz Asus ROG Rampage VI Apex Titan X (Pascal) @ 2.05GHz 32GB DDR4 4000-17-18-18-41-1T 
Hard DriveHard DriveCoolingOS
480GB - Intel Optane 900P 2TB - Samsung 960 Pro EK Custom Water Loop Windows 10 Pro x64 
MonitorMonitorMonitorKeyboard
Acer XB270HU bprz HTC Vive LG OLED55C7P Logitech G810 
PowerCaseMouseMouse Pad
Seasonic PRIME 1200 Platinum Old Marble Slab Logitech G900 Logitech G440 
AudioAudioOther
Sennheiser HD 600 Creative SoundBlasterX AE-5 Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
CPUMotherboardGraphicsRAM
i7-5960X @ 4.2GHz Asus Rampage 5 Extreme Nvidia GeForce GT 545 32GB DDR4 (2400-12-12-12-28-1T) 
Hard DriveHard DriveHard DriveOptical Drive
Samsung 950 Pro M.2 512GB HGST NAS 4TB x8 - 21.8TB RAID6 Western Digital Black 4TB Samsung SH-S183L 
CoolingOSMonitorKeyboard
Noctua NH-D15 Windows 10 Pro Asus VG278H WASD "CODE" Keyboard 
PowerCaseMouseOther
SeaSonic Platinum-1000 DIYPC Alpha-GT3 Logitech G700s Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
Other
Adaptec RAID 71605 
  hide details  
Reply
Desktop
(19 items)
 
RAID
(17 items)
 
 
CPUMotherboardGraphicsRAM
Intel i9-7900X @ 4.7GHz Asus ROG Rampage VI Apex Titan X (Pascal) @ 2.05GHz 32GB DDR4 4000-17-18-18-41-1T 
Hard DriveHard DriveCoolingOS
480GB - Intel Optane 900P 2TB - Samsung 960 Pro EK Custom Water Loop Windows 10 Pro x64 
MonitorMonitorMonitorKeyboard
Acer XB270HU bprz HTC Vive LG OLED55C7P Logitech G810 
PowerCaseMouseMouse Pad
Seasonic PRIME 1200 Platinum Old Marble Slab Logitech G900 Logitech G440 
AudioAudioOther
Sennheiser HD 600 Creative SoundBlasterX AE-5 Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
CPUMotherboardGraphicsRAM
i7-5960X @ 4.2GHz Asus Rampage 5 Extreme Nvidia GeForce GT 545 32GB DDR4 (2400-12-12-12-28-1T) 
Hard DriveHard DriveHard DriveOptical Drive
Samsung 950 Pro M.2 512GB HGST NAS 4TB x8 - 21.8TB RAID6 Western Digital Black 4TB Samsung SH-S183L 
CoolingOSMonitorKeyboard
Noctua NH-D15 Windows 10 Pro Asus VG278H WASD "CODE" Keyboard 
PowerCaseMouseOther
SeaSonic Platinum-1000 DIYPC Alpha-GT3 Logitech G700s Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
Other
Adaptec RAID 71605 
  hide details  
Reply
post #124 of 272
Quote:
Originally Posted by Asmodian View Post

Nvidia is always doing R&D on new chips, you cannot lay all those people off after they finish a new design and you do not hire a new team to design each architecture. R&D costs are not spiking with each new architecture. If Volta is done, $0 is saved by not releasing it. Worrying about R&D money spent on designing Pascal is the sunk cost fallacy, that money has been spent and nothing you do now will change that.

The reason for the 780Ti and Titan Black was because Nvidia did not have another architecture ready, not because they had Maxwell ready to go but were just waiting for the right time. If Nvidia had Maxwell ready when the 780Ti was released you can be sure they would have released the Titan X then instead of the Titan Black.

Edit: Also, more time selling Pascal is less time selling Volta, people who already bought Pascal might buy Volta and Volta doesn't cost them more to make. Why not sell Volta as soon as it is ready?
completely agree.they'll probably produce enough gp104/102/106 chips to sell them till 2h 17 and then it's volta time. I personally upgrade once a year around summer time and my budget is around $600 max and that includes getting most of it from selling the old card. So I can either buy a new volta card or buy nothing.
Edited by Klocek001 - 7/27/16 at 11:11am
post #125 of 272
Quote:
Originally Posted by sherlock View Post

Except you don't need a new architecture(and all the extra R&D cost associated with that) to achieve "Lots of releases with just enough of an improvement" with all the current Pascal chips on hand for 2017.

Nvidia made plenty of $$ from reusing GTX 680's GK104 chip for GTX 770 a year later(and 670's GK104 in GTX 760), rushing out a GM204 in 2013 would have costed them hundred of millions more in R&D and made selling GK110 at $650/700 impossible. The end result would be less, not more profit.

That's where the "rush Volta out next year" logic fails in its business model. What would maximize NV's revenue when Vega comes out is something similiar what they did in 2013:

  • 1080 price drop to $400-500 as 1170
  • Titan X 2016 rebranded as 1180, takes the $650-$700 price point
  • 1070 price drop to 1060's price point and become 1160
  • Full core Titan X Black succeed to the $1200 price point

Rushing out a 16nm GV104 as X80 Volta that's probably faster than GP102(just like GM204 was faster than GK110) would make no sense and kill the 1080Ti/Titan X Black market for no comparable benefit to offset the extra R&D cost. What Volta could do is show up in late late 2017 or early 2018 with a GV107 small chip as the 1150/1150Ti.

They got away with doing what they did with Kepler because GK110 offered a s substantial boost in performance over GK104. (30% according to TPU) The full GP102 would only have 7% more shaders, and unless they ramp up the clocks, it's only going to perform 10% better than Titan XP in the very best case scenario, and that's simply not enough of an upgrade.

Basically, the situation we have now would be analogous to them launching the 680 in March 2012, and then launching OG Titan only 2 months later in May 2012. If that's how Kepler ended up panning out, I find it hard to believe they would drag it out for another year and launch nothing but rebrands until 2014.
post #126 of 272
Quote:
Originally Posted by Asmodian View Post

Nvidia is always doing R&D on new chips, you cannot lay all those people off after they finish a new design and you do not hire a new team to design each architecture. R&D costs are not spiking with each new architecture. If Volta is done, $0 is saved by not releasing it. Worrying about R&D money spent on designing Pascal is the sunk cost fallacy, that money has been spent and nothing you do now will change that.

The reason for the 780Ti and Titan Black was because Nvidia did not have another architecture ready, not because they had Maxwell ready to go but were just waiting for the right time. If Nvidia had Maxwell ready when the 780Ti was released you can be sure they would have released the Titan X then instead of the Titan Black.

Edit: Also, more time selling Pascal is less time selling Volta, people who already bought Pascal might buy Volta and Volta doesn't cost them more to make. Why not sell Volta as soon as it is ready?

The problem is you assume Volta is done(Fermi/Kepler/Maxwell each took 20-24 month to develop, and you think Volta is done in 12 month?) and nothing can be done in 2017 to improve it(and HBM2 price would also go down further the later you push the Volta release date), whereas I think it is still in development for late 2017/early 2018. You can have GV104 out in early 2H 2017, but that's going to mean take R&D off GV100 chips development and overtime which means higher cost. Or you could just leave those people where they are and refine Pascal for less.

If you push out Volta within a year of Pascal that means less not more time to sell Volta, people would just assume NV is now on a 1 year architecture development cycle and by H2 2018 will be waiting for the next architecture. Nvidia is the marketing leadership role, and like Intel that means it would make more business sense to scale back R&D and milk existing product more like they did with GK110,760 & 770, accelerating architecture release cycle from 20-24 month to 12 month would be the polar opposite.
Edited by sherlock - 7/27/16 at 12:33pm
Skylake 1080 FTW
(18 items)
 
XPS 15
(7 items)
 
CPUMotherboardGraphicsRAM
i7-6700K  ASUS Maximus VIII Ranger EVGA 1080 FTW TridentZ DDR4-3000 
Hard DriveHard DriveHard DriveHard Drive
Samsung SM951 128GB Samsung 840 Pro 256 GB Muskin Reactor 1TB Seagate Baracuda 2TB ST2000DM001 HDD 
Hard DriveCoolingOSMonitor
Seagate Baracuda 2TB ST2000DM001 HDD Corsair H110i GT Windows 10 Pro  Acer XB321HK 32" 4K G-sync 
KeyboardPowerCaseMouse
Corsair K70 Rapid Fire EVGA SuperNova 650 P2 NZXT H440 Black Logitech G900  
Mouse PadOther
Razer Vespula CyberPower CP1500PFCLCD - PFC Sinewave UPS Syst... 
CPUMotherboardGraphicsRAM
i7-7700HQ 2.8Ghz(3.8Ghz Turbo) Dell XPS 15-9560 Nvidia GTX 1050 8GB DDR4-2400 
Hard DriveOSMonitor
256GB PCIE SSD Windows 10 Pro 15.6" 1080p 
  hide details  
Reply
Skylake 1080 FTW
(18 items)
 
XPS 15
(7 items)
 
CPUMotherboardGraphicsRAM
i7-6700K  ASUS Maximus VIII Ranger EVGA 1080 FTW TridentZ DDR4-3000 
Hard DriveHard DriveHard DriveHard Drive
Samsung SM951 128GB Samsung 840 Pro 256 GB Muskin Reactor 1TB Seagate Baracuda 2TB ST2000DM001 HDD 
Hard DriveCoolingOSMonitor
Seagate Baracuda 2TB ST2000DM001 HDD Corsair H110i GT Windows 10 Pro  Acer XB321HK 32" 4K G-sync 
KeyboardPowerCaseMouse
Corsair K70 Rapid Fire EVGA SuperNova 650 P2 NZXT H440 Black Logitech G900  
Mouse PadOther
Razer Vespula CyberPower CP1500PFCLCD - PFC Sinewave UPS Syst... 
CPUMotherboardGraphicsRAM
i7-7700HQ 2.8Ghz(3.8Ghz Turbo) Dell XPS 15-9560 Nvidia GTX 1050 8GB DDR4-2400 
Hard DriveOSMonitor
256GB PCIE SSD Windows 10 Pro 15.6" 1080p 
  hide details  
Reply
post #127 of 272
Quote:
Originally Posted by sherlock View Post

The problem is you assume Volta is done(Fermi/Kepler/Maxwell each took 20-24 month to develop, and you think Volta is done in 12 month?) and nothing can be done in 2017 to improve it(and HBM2 price would also go down further the later you push the Volta release date), whereas I think it is still in development for late 2017/early 2018. You can have GV104 out in early 2H 2017, but that's going to mean take R&D off GV100 chips development and overtime which means higher cost. Or you could just leave those people where they are and refine Pascal for less.

If you push out Volta within a year of Pascal that means less not more time to sell Volta, people would just assume NV is now on a 1 year architecture development cycle and by H2 2018 will be waiting for the next architecture. Nvidia is the marketing leadership role, and like Intel that means it would make more business sense to scale back R&D and milk existing product more like they did with GK110,760 & 770, accelerating architecture release cycle from 20-24 month to 12 month would be the polar opposite.

Of course this does assume Volta is done. I am sure they aren't doing overtime to push Volta out but given my limited experiance in R&D you cannot precisely schedule completion dates, you have to plan to be done earlier incase something happens.

From what we can tell everyone has had a lot of trouble moving below 20nm so plans didn't work out. This means that the ideal planned schedule went out the window and they have been working on Volta all this time.

I completely agree that Nvidia will slow down R&D over the long term if AMD continues being uncompetitive, but AMD hasn't been uncompetitive in the GPU space for very long yet. Nvidia just recently got caught by the 290X, the entire 700 series was not what Nvidia would have liked, too many big dies and damaging their new money making Titan brand.

A new method (also the original method with the first Titan); full big die only in Quadro with a cut one for the Titan. Everyone else gets the medium or small die. Profit. If AMD cannot beat the medium die Nvidia has no reason to ever move a big die below $1000.
Desktop
(19 items)
 
RAID
(17 items)
 
 
CPUMotherboardGraphicsRAM
Intel i9-7900X @ 4.7GHz Asus ROG Rampage VI Apex Titan X (Pascal) @ 2.05GHz 32GB DDR4 4000-17-18-18-41-1T 
Hard DriveHard DriveCoolingOS
480GB - Intel Optane 900P 2TB - Samsung 960 Pro EK Custom Water Loop Windows 10 Pro x64 
MonitorMonitorMonitorKeyboard
Acer XB270HU bprz HTC Vive LG OLED55C7P Logitech G810 
PowerCaseMouseMouse Pad
Seasonic PRIME 1200 Platinum Old Marble Slab Logitech G900 Logitech G440 
AudioAudioOther
Sennheiser HD 600 Creative SoundBlasterX AE-5 Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
CPUMotherboardGraphicsRAM
i7-5960X @ 4.2GHz Asus Rampage 5 Extreme Nvidia GeForce GT 545 32GB DDR4 (2400-12-12-12-28-1T) 
Hard DriveHard DriveHard DriveOptical Drive
Samsung 950 Pro M.2 512GB HGST NAS 4TB x8 - 21.8TB RAID6 Western Digital Black 4TB Samsung SH-S183L 
CoolingOSMonitorKeyboard
Noctua NH-D15 Windows 10 Pro Asus VG278H WASD "CODE" Keyboard 
PowerCaseMouseOther
SeaSonic Platinum-1000 DIYPC Alpha-GT3 Logitech G700s Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
Other
Adaptec RAID 71605 
  hide details  
Reply
Desktop
(19 items)
 
RAID
(17 items)
 
 
CPUMotherboardGraphicsRAM
Intel i9-7900X @ 4.7GHz Asus ROG Rampage VI Apex Titan X (Pascal) @ 2.05GHz 32GB DDR4 4000-17-18-18-41-1T 
Hard DriveHard DriveCoolingOS
480GB - Intel Optane 900P 2TB - Samsung 960 Pro EK Custom Water Loop Windows 10 Pro x64 
MonitorMonitorMonitorKeyboard
Acer XB270HU bprz HTC Vive LG OLED55C7P Logitech G810 
PowerCaseMouseMouse Pad
Seasonic PRIME 1200 Platinum Old Marble Slab Logitech G900 Logitech G440 
AudioAudioOther
Sennheiser HD 600 Creative SoundBlasterX AE-5 Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
CPUMotherboardGraphicsRAM
i7-5960X @ 4.2GHz Asus Rampage 5 Extreme Nvidia GeForce GT 545 32GB DDR4 (2400-12-12-12-28-1T) 
Hard DriveHard DriveHard DriveOptical Drive
Samsung 950 Pro M.2 512GB HGST NAS 4TB x8 - 21.8TB RAID6 Western Digital Black 4TB Samsung SH-S183L 
CoolingOSMonitorKeyboard
Noctua NH-D15 Windows 10 Pro Asus VG278H WASD "CODE" Keyboard 
PowerCaseMouseOther
SeaSonic Platinum-1000 DIYPC Alpha-GT3 Logitech G700s Mellanox ConnectX-3 MCX312A-XCBT 10 GbE Adapter 
Other
Adaptec RAID 71605 
  hide details  
Reply
post #128 of 272
Quote:
Originally Posted by Asmodian View Post

Of course this does assume Volta is done. I am sure they aren't doing overtime to push Volta out but given my limited experiance in R&D you cannot precisely schedule completion dates, you have to plan to be done earlier incase something happens.

From what we can tell everyone has had a lot of trouble moving below 20nm so plans didn't work out. This means that the ideal planned schedule went out the window and they have been working on Volta all this time.

I completely agree that Nvidia will slow down R&D over the long term if AMD continues being uncompetitive, but AMD hasn't been uncompetitive in the GPU space for very long yet. Nvidia just recently got caught by the 290X, the entire 700 series was not what Nvidia would have liked, too many big dies and damaging their new money making Titan brand.

A new method (also the original method with the first Titan); full big die only in Quadro with a cut one for the Titan. Everyone else gets the medium or small die. Profit. If AMD cannot beat the medium die Nvidia has no reason to ever move a big die below $1000.

I work in R&D in a completely different field (pharma), but you basically plan for failures and have not one but multiple contingency plans. Hell even when something gets to clinical trials, you still keep a backup project running because you can't predict the future. Oh and the deadlines are intentionally very aggressively/borderline unrealistic to ensure enough maneuverability in case something tanks.

OT aside, I have to sadly agree with the part in bold. But I'll also add that the days of the full big die going for $500 is over, and we'll never see those days again.
post #129 of 272
Quote:
Originally Posted by ZealotKi11er View Post

So we will see 1180 in 1 year. Not bad.

No... they're going to skip a generation I bet. 1160, 1170, 1180 just don't sound good. 1280 sounds cool.
Name
(14 items)
 
  
CPUMotherboardGraphicsRAM
Intel i5 3570k 4.6GHz 1.37v ASRock Z77 Extreme6 NITRO R9 Fury 16GB 1866 Kingston HyperX DDR3 
Hard DriveHard DriveHard DriveCooling
2TB Seagate Hard drive 640GB Hitachi Hard Drive 480GB Crucial M500 SSD Hyper 212 EVO 
OSMonitorKeyboardPower
Windows 7 Ultimate 64-bit/Zorin OS 49" 4K 240Hz Sony TV Corsair K70 RGB Rapidfire Corsair AX860i 
CaseMouse
Haf X Logitech G400s 
  hide details  
Reply
Name
(14 items)
 
  
CPUMotherboardGraphicsRAM
Intel i5 3570k 4.6GHz 1.37v ASRock Z77 Extreme6 NITRO R9 Fury 16GB 1866 Kingston HyperX DDR3 
Hard DriveHard DriveHard DriveCooling
2TB Seagate Hard drive 640GB Hitachi Hard Drive 480GB Crucial M500 SSD Hyper 212 EVO 
OSMonitorKeyboardPower
Windows 7 Ultimate 64-bit/Zorin OS 49" 4K 240Hz Sony TV Corsair K70 RGB Rapidfire Corsair AX860i 
CaseMouse
Haf X Logitech G400s 
  hide details  
Reply
post #130 of 272
Quote:
Originally Posted by Asmodian View Post

Of course this does assume Volta is done. I am sure they aren't doing overtime to push Volta out but given my limited experiance in R&D you cannot precisely schedule completion dates, you have to plan to be done earlier incase something happens.

From what we can tell everyone has had a lot of trouble moving below 20nm so plans didn't work out. This means that the ideal planned schedule went out the window and they have been working on Volta all this time.

I completely agree that Nvidia will slow down R&D over the long term if AMD continues being uncompetitive, but AMD hasn't been uncompetitive in the GPU space for very long yet. Nvidia just recently got caught by the 290X, the entire 700 series was not what Nvidia would have liked, too many big dies and damaging their new money making Titan brand.

A new method (also the original method with the first Titan); full big die only in Quadro with a cut one for the Titan. Everyone else gets the medium or small die. Profit. If AMD cannot beat the medium die Nvidia has no reason to ever move a big die below $1000.

I agree with this bolded point. R and D is a huge cost for them and one way to deal with the market slowing down is less spent on R and D. However, i believe Nvidia will charge 800 dollars or so for their big die. The concession is they will try to sell a titan version first at a huge premium, and the card they release for 800 dollars will be cut down but close to the same performance. This card will come later depending on the competition and even with no competition, 6 months will be when these cards are release. If Nvidia prices their cards too high, they are more likely to be charged for monopoly practices. Even if AMD graphic division dies off, as long as they don't abuse their monopoly position, I don't think there will be antitrust issues. This is because discrete graphics are not a necessity and is a pure luxury. However if they continue to rise prices much further than they are today, then Nvidia will be investigated.

Big dies are bound to have chip candidates that don't make it to titan grade. These won't be throw away and will be saved for a chip similar to the gtx 780. This card will be the volume card and overall generates more profit than titan. I can't see titan xp having much volume at this price. Enough to sell out but that's because of yields and because many of these chips are being taken for professional use. I think nvidia is going to price their further cut down from titan's at 66% of the price of their titan cards.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Rumors and Unconfirmed Articles
Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [WCCF] Nvidia's Volta could launch as early as 2H 2017 ON 16nm node