NVIDIA Pascal GP104 and GP 106 GPU Pictures Leaked - to feature GDDR5 memory

[Mr_Troll]

NVIDIA Pascal GP104 GPU Pictures Leaked – Built To Power 8 GB GeForce Cards, GP106 Also Shot Up Close
 

Quote

A few hours ago, we reported that NVIDIA is expected to launch their consumer grade GeForce graphics cards at Computex. The new graphics cards which will be allegedly known as the GeForce GTX 1080 and GTX 1070 will use NVIDIA’s latest Pascal GPU architecture which will power a range of platforms. The latest leak has just hit on the web, revealing the first shot if NVIDIA’s GP104 GPU

 

NVIDIA Pascal GP104 GPU Leaked – Die Measures Around 300mm2, Features 8 Gbps GDDR5 Memory

Quote

The NVIDIA GP104 GPU will be featured on a range of graphics cards that NVIDIA is expected to launch very soon. In our previous report, we mentioned that NVIDIA plans to unveil their GeForce GTX 1080 and GTX 1070 graphics cards at Computex. The GP104 GPU is currently under full production and will be ready till Computex for NVIDIA’s AIB partners to display their new custom solutions that utilize the Pascal GP104 graphics core.

 

Quote

 

Looking at the leaked picture, we can note that the GP104 GPU measure around 300mm2. This puts the GP104 around 100mm2 smaller than the GM204 GPU which was featured on the GeForce GTX 980 and GTX 970 cards. The GM204 GPU measured 398mm2 and featured 5.20 Billion transistors. The GP104 GPU is expected to house over 7 Billion transistors which will be really close to the current flagship GPU, the GM200 which has 8.00 billion transistors under its hood.

 

The Pascal SMP is a Maxwell SMM cut into halve with even higher instruction throughput!

Quote

The leaked picture further reveals that NVIDIA has opted for a more rectangular design and we have seen the same trend with their GP106 and GP100 GPUs which were shown at GTC 2016. The rectangular design doesn’t mean a lot but NVIDIA has mostly outed square shaped GPUs in their previous lineup with the exception of the GM206, GF114 and a few other GPUs.

NVIDIA GP104 Uses Denser and Faster GDDR5 Memory

Both GP104 and GP106 samples have featured the K4G80325FB-HC25 memory from Samsung.

Quote

Another interesting thing about this leak is that the GPU pictured is most probably an early engineering sample since the board is green. The board uses the latest Samsung K4G80325FB-HC25 chips which use the 8 Gb DRAM and run at 8 Gbps speeds. This will allow a decent boost over the current 7 Gbps memory which GPU makers are currently using. The higher density chips will allow NVIDIA to offer beefier VRAM solution so the 8 GB VRAM rumors might be true. NVIDIA used the same memory on their Pascal GP106 GPU which was showcased on the Drive PX 2 module at Computex so it’s confirmed that NVIDIA will stick to GDDR5 solutions.

NVIDIA GP106 With Pascal Architecture Featured – FinFET Based, Efficient GPU For The Masses

Quote

While we were working on our post, our friends at Videocardz posted an interesting story on the GP106 GPU. They have provided an interesting comparison of the GP106 GPU with GM206 which reveals that NVIDIA’s entry level chip will feature a die size less than 200mm2 (170-200 m2) compared to 227mm2 of the GM206 chip. The GPU is shipping with Drive PX 2 modules in Fall 2016 but since it will be available in MXM form factors too and already seen to do that, we could see some mobility parts earlier.

Quote

 

The NVIDIA and AMD solutions will be designed to deliver lots of horse power to gamers while being efficient. It seems like that 8 GB VRAM will become the standard for high performance graphics cards like the GeForce GTX 1080 and 1070. These graphics cards, fused with a range of new features will be aim to provide better graphics performance at high resolutions and in games that are built with DirectX 12 and Vulkan APIs.

Not liking pascal so far. 
 

Source:http://wccftech.com/nvidia-pascal-gp104-gpu-leaked/

http://videocardz.com/58973/nvidia-pascal-gp106-gpu-pictured-up-close

 

[don_svetlio]

GDDR5? GP106 die is as small as GM206 - not good.

[NumLock21]

That was expected. HBM2 will cost the gtx1080 to sky rocket in price.  Imo gddr5x not gddr5.

[Delicieuxz]

AMD and Nvidia have been dragging their feet on improved graphics performance for years, milking the market and making things graphical relatively stagnant. This really has to come to an end. I hope the upcoming GPUs kick the S out of the previous generation - and I mean in performance, not power efficiency.

[Alexzz_]

I dont know if i can trust this, look at his name

 

[SparkyRain]

BUT GRADE WHAT IT'S GONNA LOOK LIKE..

 

Is it this time again? What it's gonna be. A gpu, 20% more fps and 30% less fps per dollar. I don't know what you people thought what else it's gonna be. 

 

Nvidia has zero need to rush anything since the market is so milkable.

 

[MadG4mer]

And the Wcc is not trustworthy pagle either

[Overkilled]

Is this good or bad?

[Notional]

1 minute ago, Delicieuxz said:

AMD and Nvidia have been dragging their feet on improved graphics performance for years, milking the market and making things graphical relatively stagnant. This really has to come to an end. I hope the upcoming GPUs kick the S out of the previous generation.

To be fair, it's not really eithers fault that TSMC dropped the ball with 20nm. Fiji was supposed to be 20nm after all. 14/16nm will surely make a huge difference in power efficiency. However things will be different between AMD and NVidia:

• NVidia will include more compute capabilities, that they removed from Kepler to Maxwell (causing Maxwell to be more efficient). This will likely add a lot more heat and power drain to Pascal compared to Maxwell, but will be somewhat negated by going from 28nm to 16nm FF+. (However p100 being a 300w tdp chip is quite shocking).
• Since AMD already has really good compute capabilities, they should have a much larger jump in efficiency compared to NVidia (doesn't mean AMD will be more efficient though). But 2.5x in Polaris is quite a feat none the less.

With NVidia seemingly only changing Maxwell to Pascal, but AMD having a completely new architecture, things might look very good for AMD. Their smaller node MIGHT make a difference, but since it's 16nm FF+ not just 16nm FF, the difference should be miniscule.

 

Either way, things should be very interesting this year. I just think AMD will be impressive and Pascal will be disappointing.

[MadG4mer]

14 minutes ago, Notional said:

To be fair, it's not really eithers fault that TSMC dropped the ball with 20nm. Fiji was supposed to be 20nm after all. 14/16nm will surely make a huge difference in power efficiency. However things will be different between AMD and NVidia:

• NVidia will include more compute capabilities, that they removed from Kepler to Maxwell (causing Maxwell to be more efficient). This will likely add a lot more heat and power drain to Pascal compared to Maxwell, but will be somewhat negated by going from 28nm to 16nm FF+. (However p100 being a 300w tdp chip is quite shocking).
• Since AMD already has really good compute capabilities, they should have a much larger jump in efficiency compared to NVidia (doesn't mean AMD will be more efficient though). But 2.5x in Polaris is quite a feat none the less.

With NVidia seemingly only changing Maxwell to Pascal, but AMD having a completely new architecture, things might look very good for AMD. Their smaller node MIGHT make a difference, but since it's 16nm FF+ not just 16nm FF, the difference should be miniscule.

 

Either way, things should be very interesting this year. I just think AMD will be impressive and Pascal will be disappointing.

Hopefully they will just publish both cards and customers can buy the better one ;-)

[awesomeness10120]

30 minutes ago, NumLock21 said:

That was expected. HBM2 will cost the gtx1080 to sky rocket in price.  Imo gddr5x not gddr5.

GDDR5X isn't going to be available until late this year. So expect GDDR5 on these cards.

[Notional]

3 minutes ago, MadG4mer said:

Hopefully they will just publish both cards and customers can buy the better one ;-)

They're not going to buy the better one, if they are already stuck in a vendor lock in by buying a gsync monitor or a shield tv thing. But other than that, yeah.

[-BirdiE-]

56 minutes ago, don_svetlio said:

GDDR5? GP106 die is as small as GM206 - not good.

Is it not good that it's the same size as the GM206 since the 16nm chip can fit nearly double the transistors in the same space?

[226477_1454181668]

1 hour ago, NumLock21 said:

That was expected. HBM2 will cost the gtx1080 to sky rocket in price.  Imo gddr5x not gddr5.

And its not even the cost. Its how many 1080's nvidia will sell. It will probably be a very hot selling card, and there simply will not be enough HBM for a card like the 1080.

[NumLock21]

37 minutes ago, awesomeness10120 said:

GDDR5X isn't going to be available until late this year. So expect GDDR5 on these cards.

Hope my hd5850 will hold until hbm2 or at least gddr5x. Don't run games too high anymore, need to keep my gpu running for as long as possible. Haha

 

6 minutes ago, 226477_1454181668 said:

And its not even the cost. Its how many 1080's nvidia will sell. It will probably be a very hot selling card, and there simply will not be enough HBM for a card like the 1080.

Imo the vram limit is a factor too. Hbm tops out at 4gb.

[226477_1454181668]

8 minutes ago, NumLock21 said:

Hope my hd5850 will hold until hbm2 or at least gddr5x. Don't run games too high anymore, need to keep my gpu running for as long as possible. Haha

 

Imo the vram limit is a factor too. Hbm tops out at 4gb.

HBM1, HBM2 is up to 32gb.

[don_svetlio]

1 minute ago, 226477_1454181668 said:

HBM1, HBM2 is up to 32gb.

8*

16GB is still being developed
No mention of 32GB

[NumLock21]

8 minutes ago, 226477_1454181668 said:

HBM1, HBM2 is up to 32gb.

Max i see is 4gb.

[GoodBytes]

As I mentioned many many times, and ignore, is that Pascal is NOT a new architecture. It is still Maxwell, just tweaked, and added features enough to have a new name, but still very close to Maxwell. Volta is Nvidia next architecture and that is expected for 2017-2018.

 

Now, like always, there might be a Titan version with HBM2 memory, as those are Tesla chips that doesn't meet Nvidia specifications/requirement to meet Tesla model. But ignoring that, as this is a high end and very expensive model, there is not much to get excited really about. We may have GDDR5X though, which will help things.

 

[awesomeness10120]

13 minutes ago, GoodBytes said:

As I mentioned many many times, and ignore, is that Pascal is NOT a new architecture. It is still Maxwell, just tweaked, and added features enough to have a new name, but still very close to Maxwell. Volta is Nvidia next architecture and that is expected for 2017-2018.

 

Now, like always, there might be a Titan version with HBM2 memory, as those are Tesla chips that doesn't meet Nvidia specifications/requirement to meet Tesla model. But ignore that, as this is a high end and very expensive model, there is not much to get excited really about.

 

Yeah, the cost of HBM2 will probably limit the availability of any consumer GP100 GPU at all this year.

[Trixanity]

Well, no surprise that any launch in the summer would ship with regular GDDR5. Both HBM2 and GDDR5X aren't ready yet and will both cost more. 

[DocSwag]

Not surprised at no hbm2 anywhere. I'm not expecting it until early 2017 tbh. AMD made it pretty clear at Capsaicin that Polaris would be gddr5 and we wouldn't get hbm2 until vega at the beginning of 2017. Guessing it will be somewhat similar with pascal.

 

And no, don't expect gddr5x on them either. Micron only recently began sampling of it, not full fledged production. Their scheduled to begin mass production  in the summer (and you don't begin shipping it right after mass production begins), which means any GPUs with it would be getting announced in September/October ish most likely. And by that time you might as well wait for hbm2. I'd be pretty surprised if we got any gddr5x high end cards. By the time it's available hbm2 will be there, do it'll probably just become a cheaper slightly weaker alternative to hbm2.

[daniellearmouth]

3 hours ago, Notional said:

They're not going to buy the better one, if they are already stuck in a vendor lock in by buying a gsync monitor or a shield tv thing. But other than that, yeah.

And hence why I now have a 980 Ti.

[Notional]

2 minutes ago, daniellearmouth said:

And hence why I now have a 980 Ti.

Indeed. I really like EVGA, but I despise NVidia's business strategies, as they are very anti consumer. Some people confuses this with me being an AMD fanboy, which is missing the point completely.

 

On a side note, I remember a UK retailer called black star with a star in it  .. imported some anime and stuff from there to Denmark back in the days.

[daniellearmouth]

1 minute ago, Notional said:

Indeed. I really like EVGA, but I despise NVidia's business strategies, as they are very anti consumer. Some people confuses this with me being an AMD fanboy, which is missing the point completely.

 

On a side note, I remember a UK retailer called black star with a star in it  .. imported some anime and stuff from there to Denmark back in the days.

I can understand why one may not agree with NVIDIA's stuff. Though, personally, I'm more about the straight-up performances and utility of the card, and whilst AMD does have some better stuff, I'm more partial to the team in green for the features they offer.

Also, pretty neat!