• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

NVIDIA Volta Rumor Thread

Page 8 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Konan

Senior member
Jul 28, 2017
360
291
106
I was checking the Volta whitepaper over. > https://images.nvidia.com/content/volta-architecture/pdf/Volta-Architecture-Whitepaper-v1.0.pdf
Noticed that Volta moves to Four from Six clock cycles with Pascal - That is a pretty big and interesting change for Volta SM

Simultaneous Execution of FP32 and INT32 Operations Unlike Pascal GPUs, which could not execute FP32 and INT32 instructions simultaneously, the Volta GV100 SM includes separate FP32 and INT32 cores, allowing simultaneous execution of FP32 and INT32 operations at full throughput, while also increasing instruction issue throughput. Dependent instruction issue latency is also reduced for core FMA math operations, requiring only four clock cycles on Volta, compared to six cycles on Pascal.
Many applications have inner loops that perform pointer arithmetic (integer memory address calculations) combined with floating-point computations that will benefit from simultaneous execution of FP32 and INT32 instructions. Each iteration of a pipelined loop can update addresses (INT32 pointer arithmetic) and load data for the next iteration while simultaneously processing the current iteration in FP32.
 

Bouowmx

Golden Member
Nov 13, 2016
1,133
545
146
GeForce Volta is likely dependent on GDDR6 availability, 2018 Q1. So, can't bring it to market sooner, even if possible.

2018 March is fine, in the range of realistic estimates. I've been operating on 2018 Q1 for a while.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
106
They don't have any real reason to bring them to market any sooner do they?
It doesn't much matter what market conditions are, you bring a product to market when it's ready.

AMD couldn't bring Vega to market against GTX 1080, just because it would have been convenient to do so 15 months ago, because that product was not ready.

On the flip-side, you don't shelve a ready product, just because your competitor hasn't caught you.

Products have massive planning, and detailed scheduling that drives towards goal targets, often with incentives for hitting milestones on time. These plans are years long. You don't follow the competition and speed up and slow down to mimic them.

Managers don't say "Hey, AMD screwed up Vega, everyone take the next 4 months off!".

Volta arrives when it's finished and it will be pushing hard to meet internal milestones until it is finished.

When Volta arrives has absolutely nothing to do with AMD.
 

Puffnstuff

Lifer
Mar 9, 2005
15,415
4,146
136
You are correct in that the PDLC goes on yet if your current product doesn't have any market pressure there is little reason to rush the next one and that is a business reality. On the other hand if you are the company that is behind then you put in double duty to get your products to market as soon as possible to compete. If you know that your competition is about to release a new product then you drop yours if its ready to maintain your lead. What Nvidia or any other company does has everything to do with their competitors plans and that is good business.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
106
You are correct in that the PDLC goes on yet if your current product doesn't have any market pressure there is little reason to rush the next one and that is a business reality. On the other hand if you are the company that is behind then you put in double duty to get your products to market as soon as possible to compete. If you know that your competition is about to release a new product then you drop yours if its ready to maintain your lead. What Nvidia or any other company does has everything to do with their competitors plans and that is good business.
The competition only factors tangentially into the beginning of product cycle, where PLMs gather a view of the competitive landscape and create a feature wishlist and timetables (which will get tempered by realities from design). Once the product is rolling in design, PLM are largely out of the picture and development staff pays ZERO mind to what the competition is doing, and is fixed on their plan. Naturally there will be roadblocks to adjust a plan, or a feature change, but significant attempts for PLM to change the initial plan is a massive negative disruption.

Attempt to rush, or ease off, are both poor management.

NVidia GPU design is run like a well oiled machine at this point. They run to to their own timetables, and they deliver when ready, not before, and not after. They won't rush, they won't delay.

The only short term reaction they will have is marketing moves (ie Fluff). Nothing substantial will change.
 

Glo.

Diamond Member
Apr 25, 2015
4,841
3,461
136
Guys, what is the chance that Nvidia will use Actual Volta architecture in consumer products, instead of repurposing, and reusing GP100 architecture?

Volta has massive changes and advantages compared to GP100. Improved Scheduling, separate INT and FP32, for higher bandwidth, better cache layout, and this is just the tip of the iceberg, of Volta changes.

I have read some time ago, that according to sources consumer Volta is bigger change than Maxwell was for Kepler.

Is it anywhere rumored/confirmed also that the GPUs will be made on 12/16 nm FFN process, from TSMC, not usual 16 nm FF+?
 

Konan

Senior member
Jul 28, 2017
360
291
106
Is it anywhere rumored/confirmed also that the GPUs will be made on 12/16 nm FFN process, from TSMC, not usual 16 nm FF+?
It says in the whitepaper I posted top of page that GV100
It is fabricated on a new TSMC 12 nm FFN (FinFET NVIDIA)
so who knows what consumer variant will be.

This week, TSMC, Nvidia’s silicon partner, announced that it will be kicking off production of its 12nm FinFET chips later this year, making Volta-based GeForce graphics cards possible as soon as early 2018.

Given that there is a Volta-based GPU out already, this won’t be the first time TSMC has produced 12nm FinFET. However, starting in Q4 volume will be ramped up in order to appease the consumer market.
With SK Hynix confirmed they're entering volume production of GDDR6 ram for a high end graphics card early 2018. 12 nm FinFET is scheduled to hit mass production before the year is over.
Both those together suggests an Q1 2018 launch for Volta unless nvidia sits on it. It could be that TSMC's "new 12nm node" is just a rehash of their 16nm node not necessarily a die shrink. Then again, it could very well be 12nm is taped out early Q1 2018, with decent supplies starting at Q2 2018
 

Glo.

Diamond Member
Apr 25, 2015
4,841
3,461
136
It says in the whitepaper I posted top of page that GV100 so who knows what consumer variant will be.



With SK Hynix confirmed they're entering volume production of GDDR6 ram for a high end graphics card early 2018. 12 nm FinFET is scheduled to hit mass production before the year is over.
Both those together suggests an Q1 2018 launch for Volta unless nvidia sits on it. It could be that TSMC's "new 12nm node" is just a rehash of their 16nm node not necessarily a die shrink. Then again, it could very well be 12nm is taped out early Q1 2018, with decent supplies starting at Q2 2018
So this basically means, that every GPU that will not clock over 2.4 GHz, on this process, will receive 20% die shrink. I was theorizing about possibilities for GV107 chips, and whether it would be possible to pack 1024 Volta CUDA cores, with 192 bit memory bus, for this market in 75W TDP. This may confirm that it is possible design.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,139
131
These are the latest Volta rumors from China. Usual grain of salt required, but since this is a rumor thread...

- If not postponed, the first card could be released right after the Chinese New Year 2018 (February 15-17)
- GV104 comes first, the flagship card is meant to replace the GeForce GTX 1080 Ti
- GV104 scores as high as 16.000 to 17.000 points in Fire Strike Extreme GPU
- The die size of GV104 is similar to GM204 (398 mm²), therefore larger than GP104, but power consumption is closer to the latter
- NVIDIA could adopt GDDR6 across all products, including low end
 

Glo.

Diamond Member
Apr 25, 2015
4,841
3,461
136
These are the latest Volta rumors from China. Usual grain of salt required, but since this is a rumor thread...

- NVIDIA could adopt GDDR6 across all products, including low end
256 GB/s on 128 bit memory bus.

Not bad, I would love to see however 192 bit on GM107 :(.
 

Bouowmx

Golden Member
Nov 13, 2016
1,133
545
146
GV104 being about 400 mm² is believable. If you use GV100's core density improvement over GP100 to estimate GV104's area of 3584 cores (2/3 of GV100) from GP104: 420 mm²
Code:
3584/((2560/314 mm²) * ((5376/815 mm²)/(3840/610 mm²))) = 420 mm²
(GV104 core count)/((GP104 core density) * ((GV100 core density)/(GP100 core density))
GV104 and 102 flagship performance would probably be chart-topping impressive, but 107 is also interesting to me because GeForce GTX 1050 Ti's gain over the 950 (which was based on GM206, not an *07) was looked unimpressive to me.
 

Dayman1225

Golden Member
Aug 14, 2017
1,046
695
146
- GV104 scores as high as 16.000 to 17.000 points in Fire Strike Extreme GPU
So GV104 about 24%~ above 1080Ti IF this is true. So
1180/2080 > 1080Ti
1170/2070 = 1080Ti
Similar jump as Maxwell -> Pascal ? Interesting!
 

Ajay

Diamond Member
Jan 8, 2001
9,779
4,164
136
GeForce Volta is likely dependent on GDDR6 availability, 2018 Q1. So, can't bring it to market sooner, even if possible.
2018 March is fine, in the range of realistic estimates. I've been operating on 2018 Q1 for a while.
This is probably why Volta will be more expensive than Pascal. I've read that newer Graphics Ram chips are going to be much more expensive than the previous gen. Hmm, minor consolation for buying a GTX 1070 ;)
 
  • Like
Reactions: Pick2

french toast

Senior member
Feb 22, 2017
988
824
136
Looks good for Volta, hopefully this is a full dx12 implementation with asynchronous shaders and other goodies.

Personally I'm not interested in buying an nvidia GPU, but I would love nvidia to bring out another android focused tegra, either Denver 3 cores or high clocked cortex a75s, Volta mobile GPU- all in another Shield tablet with 1440p Samsung amoled, new stereo speakers, and a much bigger battery.
 
  • Like
Reactions: Lodix

Tup3x

Senior member
Dec 31, 2016
540
400
136
Guys, what is the chance that Nvidia will use Actual Volta architecture in consumer products, instead of repurposing, and reusing GP100 architecture?

Volta has massive changes and advantages compared to GP100. Improved Scheduling, separate INT and FP32, for higher bandwidth, better cache layout, and this is just the tip of the iceberg, of Volta changes.

I have read some time ago, that according to sources consumer Volta is bigger change than Maxwell was for Kepler.

Is it anywhere rumored/confirmed also that the GPUs will be made on 12/16 nm FFN process, from TSMC, not usual 16 nm FF+?
They'll take out all the tensor cores (since those are not needed and just waste space for gaming cards) but other than that I don't see why they wouldn't.
 
  • Like
Reactions: TheF34RChannel

ASK THE COMMUNITY