• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

NVIDIA Volta Unveiled (GV100 new GPU Architecture)

ethomaz said:
Depend on clock but new Titan will be based on GV102 that probably will have 5376 SPs.

1700Mhz = ~18.2TFs
1800Mhz = ~19.3TFs
1900Mhz = ~20.4TFs
2000Mhz = ~21.5TFs
Click to expand...

Yeah that's about right and with all other improvements including bandwidth etc... i'm expecting for GV102 fully Unlock Titan caliber card to be around 60-75% faster then Titan Xp
 
So we know this is going to be called the 20 series and not the 11 series?

Anyway, Square Enix used the CG character models in real time for the HD version of Type-0. Imagine if they went ahead and used the CG models for the PC version of FFXV, even just as a future-proof feature.

Personally I think my 1070 is gonna be fine for a while if I stick to 1080p gaming, maybe even until 2019 or 2020.
 
The best thing about pascal is how much power it gave laptops. Next year this time, you will get 15" laptops that are as fast as 1080ti, add to the 6cores mobile Intel cpu, laptops are going to overpower next gen consoles before they are even born(assuming they are sticking with AMD apu)
 
longdi said:
The best thing about pascal is how much power it gave laptops. Next year this time, you will get 15" laptops that are as fast as 1080ti, add to the 6cores mobile Intel cpu, laptops are going to overpower next gen consoles before they are even born(assuming they are sticking with AMD apu)
Click to expand...

I wonder if Volta laptops will be a notable jump.
 
R10Neymarfan said:
So wait this can run Kingsglaive in real time?

Give me dat FFXV with Kingsglaive fidelity and Gameworks pls
Click to expand...

It seems like with Volta they can push Luminousto greatly approximate Kingsglaive in a scene like this.

That said, I absolutely want the PC version to basically be a crazy(but scalable) tech demo Seems like a great thing for Square too. They're always doing R&D. Why wait 5-7 years to see some ROI?
 
longdi said:
The best thing about pascal is how much power it gave laptops. Next year this time, you will get 15" laptops that are as fast as 1080ti, add to the 6cores mobile Intel cpu, laptops are going to overpower next gen consoles before they are even born(assuming they are sticking with AMD apu)
Click to expand...

Yeah but you're talking about a $2500 to $3000 laptop against a $400 to $500 console....
 
The Volta-based (or whatever the hell it'll be called for mobile chips) Tegra is gonna be a fucking beast. This bodes well for the Switch 2.

Still, Volta desktop GPUs are gonna be expensive for a while. Hoo boy.
 
longdi said:
The best thing about pascal is how much power it gave laptops. Next year this time, you will get 15" laptops that are as fast as 1080ti, add to the 6cores mobile Intel cpu, laptops are going to overpower next gen consoles before they are even born(assuming they are sticking with AMD apu)
Click to expand...

Not really. A 6-core 1080 Ti laptop is going to cost about 3 times a console's price in 2018.

There was a leak for an engineering sample for AMD's next-gen APU - 4-zen cores with Vega graphics cores coming 1H 2018. PS5 will likely have the successor to this APU and heavily customized to boot. There's ample power available through AMD's APU roadmap to more than compete with an inefficient and power hungry Intel + Nvidia Pascal laptop.
 
ThoseDeafMutes said:
They're never going to stick with a consistent scheme, tech companies are allergic to it.

They're never going to include year in there because then if they release in december or don't refresh for a while they start to look bad. They're hesitant to include the architecture name because there's no clear way for consumers to tell which one is better than another (Is Hawaii better than Fiji? Is Maxwell better than Kepler???), and if there was, people would get annoyed by rebadges and low end cards using old archs. They don't want to stick with a consistent numbering scheme because they run out fast and have to compete with competitor numbering schemes that may "look" superficially better.

IMO since consumer GPUs are an enthusiast market it wouldn't matter for any of those reasons, but I think their marketing believes that the ignorant people are a sufficiently large chunk of their market that they probably want to keep trucking with the "let's just fucking throw shit at the wall and maximize confusion" scheme they all use.
Click to expand...

Well, AMD used Vega codename in the final card's naming. I also don't think that going 11xx or 20xx route is very forward looking. What's next after that? 1360? Or 5050? They already use the architecture's first letter in Teslas (P100, V100) and some Quadros (K/M/P series), now would be a good time to use it in GeForces as well.
 
Coulomb_Barrier said:
Not really. A 6-core 1080 Ti laptop is going to cost about 3 times a console's price in 2018.

There was a leak for an engineering sample for AMD's next-gen APU - 4-zen cores with Vega graphics cores coming 1H 2018. PS5 will likely have the successor to this APU and heavily customized to boot. There's ample power available through AMD's APU roadmap to more than compete with an inefficient and power hungry Intel + Nvidia Pascal laptop.
Click to expand...

Because a 1080 Ti Laptop is competing with consoles? What?

Consumer APUs don't even compete with mid range laptops, let alone high end ones. The Raven Ridge APU has like what 1000 gpu cores at 800 mhz, or was it 700 cores@800mhz?

I don't get what point you're even trying to make? How is a console that won't release for another 2 years relevant for someone who's looking to buy a high end laptop next year? Your post is just completely illogical.
 
Newboi said:
In the end, it's up to you. Pascal is going to fulfill the need you have right now (especially at resolutions below 4K). Volta is the shiny looking object in the distance, but the time consumer grade graphics cards come out using the new architecture, there will already be talks of the next architecture that will be so much better than Volta.

This is generally the issue when you wait until near the end of a GPU generation to then decide to get a GPU. The next generation will look much more enticing even if you don't need it.

My general rule of thumb is that if there is a card available that fulfills the needs you have, go ahead an buy it. There will always be a next best thing, especially when it comes to GPUs.
Click to expand...

I've never read wiser words on GAF.
 
Newboi said:
In the end, it's up to you. Pascal is going to fulfill the need you have right now (especially at resolutions below 4K). Volta is the shiny looking object in the distance, but the time consumer grade graphics cards come out using the new architecture, there will already be talks of the next architecture that will be so much better than Volta.

This is generally the issue when you wait until near the end of a GPU generation to then decide to get a GPU. The next generation will look much more enticing even if you don't need it.

My general rule of thumb is that if there is a card available that fulfills the needs you have, go ahead an buy it. There will always be a next best thing, especially when it comes to GPUs.
Click to expand...

Not to mention people seem to forget that games continue to become more demanding. Thats how they get caught in a never-ending chase for 4K 60fps. By the time Volta is released, games will require more to run at those levels, and not long after that Volta will be insufficient and people will start talking about how the next card will crush all those games at 4K 60fps.
 
I wonder if maybe, eventually, the top 1 or 2 consumer Volta graphics cards (Titan, xx80 Ti) will actually surpass 15 TFlops (fp32),
using higher clockspeeds with the GV102 chip.

Most of the fp64, Tensor units would be removed, the interposer that HBM2 requires would be gone (use GDDR6 instead) NVLink 2 gone.

Still retaining most of the 5,376 cores.

Doesn't GP102 for Titan Xp already hit 12+ TF?
 
AmyS said:
I wonder if maybe, eventually, the top 1 or 2 consumer Volta graphics cards (Titan, xx80 Ti) will actually surpass 15 TFlops (fp32),
using higher clockspeeds with the GV102 chip.

Most of the fp64, Tensor units would be removed, the interposer that HBM2 requires would be gone (use GDDR6 instead) NVLink 2 gone.

Still retaining most of the 5,376 cores.

Doesn't GP102 for Titan Xp already hit 12+ TF?
Click to expand...
Gp102 at 2ghz is 14 tflops. I think more than half of all gp102 chips can do 1.9 ghz which is still over 13tflops
 
Renekton said:
Google's own TPU

https://cloud.google.com/blog/big-d...k-at-googles-first-tensor-processing-unit-tpu
Click to expand...
Interesting!

Google's 92 matrix multiplication teraops/s falls a little short of V100's 120 teraops/s -- on the other hand...

TPU: 2300 GOPS/W
V100: 400 GOPS/W
=factor of 5.8

TPU: 7.7 GOPS/W/mm^2
V100: 0.49 GOPS/W/mm^2
=factor of 16

TPU: 7.7x GOPS/$/W/mm^2
V100: ~0.025x GOPS/$/W/mm^2
=factor of 310

I could be off by quite a bit, based on the little information that was available.
 
So probably early 2018 consumer Voltas will be out?


Hmmm just got a 1070 so ill be sleeping on the graphics game for a while, need a new processor and motherboard before focusing on graphics again.
 
Black_Stride said:
So probably early 2018 consumer Voltas will be out?


Hmmm just got a 1070 so ill be sleeping on the graphics game for a while, need a new processor and motherboard before focusing on graphics again.
Click to expand...

Consumer Volta cards are expected in 3rd quarter this year, 4th quarter at most. There's no reason for NV to wait for "early 2018" (which is a dead season sales wise) to launch them. Next GPU process update at TSMC (7FF) will happen in 2nd quarter of 2018 which is too far away all things considered and will likely be the basis for a Volta shrink around a year from it's consumer launch.
 
dr_rus said:
Consumer Volta cards are expected in 3rd quarter this year, 4th quarter at most. There's no reason for NV to wait for "early 2018" (which is a dead season sales wise) to launch them. Next GPU process update at TSMC (7FF) will happen in 2nd quarter of 2018 which is too far away all things considered and will likely be the basis for a Volta shrink around a year from it's consumer launch.
Click to expand...
Granting that the first Volta cards will be out in H2/17, I highly doubt that we will see a 7nm shrink only a year from then. The volume ramp will probably commence by Q2/18, but the first product, i.e., iPhone, historically won't debut until late Q3. This is what happened with 16FF+, with the first GPU (GTX 1080) launching three quarters later in May.
 
E-Cat said:
Granting that the first Volta cards will be out in H2/17, I highly doubt that we will see a 7nm shrink only a year from then. The volume ramp will probably commence by Q2/18, but the first product, i.e., iPhone, historically won't debut until late Q3. This is what happened with 16FF+, with the first GPU (GTX 1080) launching three quarters later in May.
Click to expand...
May 2019 is probably the date for the sucessor of Volta... about 1.5 year after.
 
ethomaz said:
May 2019 is probably the date for the sucessor of Volta... about 1.5 year after.
Click to expand...
Even without having a shrink before Volta's successor, NVIDIA could just make a bigger chip -- assuming the first high-end consumer parts will be <500 mm^2 like GP102. The 12FFN process seems particularly well-suited for big die, anyway.

Or, maybe 7nm will have more aggressive timing than 16FF+.
 
E-Cat said:
Granting that the first Volta cards will be out in H2/17, I highly doubt that we will see a 7nm shrink only a year from then. The volume ramp will probably commence by Q2/18, but the first product, i.e., iPhone, historically won't debut until late Q3. This is what happened with 16FF+, with the first GPU (GTX 1080) launching three quarters later in May.
Click to expand...

Well, yeah, what TSMC states as a moment when a process is ready for mass production is usually some 6-12 months away from the moment when it will actually be ready for a big GPU production. So if Volta launches this Autumn on 16FF++ (aka 12FFN) then Volta 2 will likely launch 12-18 months from that moment (Autumn 2018 - Spring 2019) which is a usual NV's schedule and aligns quite nicely with 7FF moment of readiness according to TSMC (Spring 2018) plus said 6-12 months to get to the actual readiness.

E-Cat said:
Even without having a shrink before Volta's successor, NVIDIA could just make a bigger chip -- assuming the first high-end consumer parts will be <500 mm^2 like GP102. The 12FFN process seems particularly well-suited for big die, anyway.

Or, maybe 7nm will have more aggressive timing than 16FF+.
Click to expand...

I think that Volta will be as big as they'll be willing to get on 16FF+. +50% performance in the same power envelope means +50% die sizes to the relative Pascal chips which puts GV102 at ~600mm^2 already and GV104 at ~450mm^2. We know that this isn't the reticle limit on TSMC's 16FF (that being roughly 800mm^2 as seen on GV100) but anything bigger probably won't be economical to produce for a mass market. Thus what comes after Volta@12FFN will most definitely have to use a more advanced process.
 
You must log in or register to reply here.

Similar threads

NVIDIA secures 94% of AIB GPU market
2 3
108
1K
LavitzSlambert replied
KeplerL2: PS6 Won’t Use Full RDNA 5 GPU Architecture, It’s Claimed
Opinion Rumor Hardware  6 7 8 9 10
497
4K
Bojji replied
Full path tracing might not be such a GPU hog in the future, as Nvidia claims over 2x performance improvements in new research
Hardware Platform 
41
317
S0ULZB0URNE replied
Nvidia Warns of 'Very Tight' GPU Supplies for at Least Next 2 Quarters
News Hardware 
5
262
Cumulus replied
NVIDIA and Samsung are making new GeForce RTX 3060 GPUs
News Hardware 
20
558
asdasdasdbb replied
Top Bottom