[Rumour] Nvidia's Pascal on track for 2016

[Bouzoo]

Time to shift the focus from AMD to Nvidia (since 980Ti is out, we need new stuff right). Apparently, someone with confidential information in the semiconductor industry revealed on Bexond 3Dforum that Nvidia has already taped out it's next-generation GP100 GPU. That would mean Nvidia is in the final stage of designing the integrated circuit (microchip). If this is true, we might be getting Pascal quite shortly after Maxwell 2.0. I'll go out on a limb and say that we are gonna (obviously) get a mix of Maxwell/Pascal (I hope they leave out Kepler). If the rumour is true. 

 

 

Details:

The world’s No. 1 producer of discrete graphics processors will reportedly use one of Taiwan Semiconductor Manufacturing Co.’s 16nm FinFET fabrication technology to make its “Big Pascal” GPU. Given the timeframe of the tape-out, it is highly likely that Nvidia uses TSMC’s advanced 16nm FinFET+ (CLN16FF+) manufacturing technology. According to the post, the BP100 is Nvidia’s first 16nm FinFET chip and the company has changed its approach to roll-out of new architectures. Instead of starting from simple GPUs and introducing biggest processors quarters after the initial chips, Nvidia will begin to roll-out “Pascal” with the largest chip in the family.

Nvidia’s “Pascal” architecture represents a big leap for the company. Thanks to all-new architecture, the Nvidia’s next-gen GPUs will support many new features introduced by DirectX 12+, Vulkan and OpenCL application programming interfaces. The 16nm FinFET process technology will let Nvidia engineers to integrate considerably more stream processors and other execution units compared to today’s GPUs, significantly increasing overall performance. In addition, next-generation graphics processing units from Nvidia will support second-generation stacked high-bandwidth memory (HBM2). The HBM2 will let Nvidia and its partners build graphics boards with 16GB – 32GB of onboard memory and 820GB/s – 1TB/s of bandwidth. For high-performance computing (HPC) applications, the “Big Pascal” chip will integrate NVLink interconnection tech with 80GB/s or higher bandwidth, which will significantly increase performance of “Pascal”-based Tesla accelerators in supercomputers. Moreover, NVLink could bring major improvements to multi-GPU technologies thanks to massive bandwidth for inter-GPU communications. According to Nvidia’s estimates, graphics adapters based on “Pascal” architecture should deliver two to ten times higher performance than comparable graphics processors today in peak scenarios.

Nvidia needs GP100 chip in order to build next-generation Tesla accelerators for supercomputers. Since “Maxwell” architecture (even the GM200) lacks native support for double precision (DP) FP64 computing, it cannot be used for Tesla cards. As a result, Nvidia currently offers Tesla accelerators featuring GK110 and GK210 chips, which are basically three years old. The release of the “Big Pascal” will help Nvidia to boost sales of Tesla cards for HPC applications.

The accuracy of predictions of the Beyond3D forum member could not be verified, but some of his previous posts indicate that he has access to information that is not yet public. The post in the forum on Friday was republished by 3DCenter, a web-site known for predictions in the field of graphics processing units.

Nvidia did not comment on the news-story.

 

How true could this be?

The accuracy of predictions of the Beyond3D forum member could not be verified, but some of his previous posts indicate that he has access to information that is not yet public. The post in the forum on Friday was republished by 3DCenter, a web-site known for predictions in the field of graphics processing units.

Nvidia did not comment on the news-story.

 

I find it pretty funny how the article says that Nvidia needs this to better compete against AMD. Right. Since they have 50/50 market share because AMD is that good nowadays.

 

Source

[SurvivorNVL]

[Enderman]

NO! you just got your 980ti so BE PATIENT!

[SurvivorNVL]

NO! you just got your 980ti so BE PATIENT!

Man, at the rate I am going with plans--I'll have four new computers in the next 12 months due to impatience.  So.. I shall re-iterate.

[VagabondWraith]

My 980 Ti is already outdated.

[Enderman]

Man, at the rate I am going with plans--I'll have four new computers in the next 12 months due to impatience.  So.. I shall re-iterate.

NEIN NEIN NEIN NEIN NEIN

[Bouzoo]

NEIN NEIN NEIN NEIN NEIN

 

What did you except was gonna happen?

[Victorious Secret]

HBM2 UBER ALLES 

[Notional]

Sounds interesting 14/16nm finfet is very welcome and interesting. However, I am worried about HBM. Looking at what AMD went through with interposers, and connecting both HBM and gpu to this interposer, I really have to question, if NVidia can pull that off without a hitch. Especially with HBM 2. Further more, will the yields be high enough if AMD has first right to these chips?

 

Either way, 2016 will be very interesting in the processor world. Yay for the consumers!

 

 

Can I have one of them then? :lol:

[patrickjp93]

Everyone hold their horses. TSMC isn't even shipping 16nm FF parts yet, and I've heard nothing from the industry saying their yields are yet up and volume production is either in swing or getting ready. Furthermore, Nvidia said Pascal will be 16nm FF+, or another quarter node behind 16nmFF. Even further, we have no news that HBM 2.0 is ready to enter production, and Nvidia's promising 8 and 16GB cards this time around. It's a clickbait article and nothing more.

 

My 980 Ti is already outdated.

 

Pascal is still looking to be a Q2/Q3 affair in 2016. You're not remotely outdated yet.

[BiG StroOnZ]

I think everyone is missing the point of this article, the point of this article is not that Pascal is due for 2016. It's the fact that they are eliminating the small die and going directly to the big die. So just to refresh your memory, with recent events, Maxwell launched with GM204, which is a small die ("mid range"). Then GM200 launched as the big die. The big die typically being 20-30% faster than the small die.

 

Here in this article they are saying that NVIDIA is completely skipping the small die and going directly to the big die for Pascal launch. So there will be no mid-range graphics card launched as a premier product like in the past (eg. GTX 680 GK104, GTX 980 GM204). Instead they will be launching their GP100 right from the get go (big die Pascal). For those who are wondering what GP100 is, it will be the equivalent of GM200 (Titan X / 980 Ti) or GK110 [originally called GK100] (Titan Black / 780 Ti); except with the Pascal architecture. As opposed to launching GP104 (eg. GM204) first, then launching GP100 (eg. GM200) later. 

[TroubleKlef]

By then I can "afford" a flagship card. Yay for Pascal.

[patrickjp93]

I think everyone is missing the point of this article, the point of this article is not that Pascal is due for 2016. It's the fact that they are eliminating the small die and going directly to the big die. So just to refresh your memory, with recent events, Maxwell launched with GM204, which is a small die ("mid range"). Then GM200 launched as the big die. The big die typically being 20-30% faster than the small die.

 

Here in this article they are saying that NVIDIA is completely skipping the small die and going directly to the big die for Pascal launch. So there will be no mid-range graphics card launched as a premier product like in the past (eg. GTX 680 GK104, GTX 980 GM204). Instead they will be launching their GP100 right from the get go (big die Pascal). For those who are wondering what GP100 is, it will be the equivalent of GM200 (Titan X / 980 Ti) or GK110 (originally called GK100) Titan Black / 780 Ti); except with the Pascal architecture. 

BWAHAHAHAHA! They're trusting TSMC that much?! I know they need to be ready with IBM to fight off Knight's Landing, but are they serious?! This is some sick joke. If TSMC falls apart at 16nmFF+ I'll practically die laughing. I mean, it'll be the end of semiconductor competition as we know it, but still... That's an all-in bet on levels I haven't seen since Intel came out with Core 2 after the Netburst fiasco.

[Accursed Entity]

Aaand this is why I didn't buy a 900 series card... My 780s can hold for another year just fine. 

[BiG StroOnZ]

BWAHAHAHAHA! They're trusting TSMC that much?! I know they need to be ready with IBM to fight off Knight's Landing, but are they serious?! This is some sick joke. If TSMC falls apart at 16nmFF+ I'll practically die laughing. I mean, it'll be the end of semiconductor competition as we know it, but still... That's an all-in bet on levels I haven't seen since Intel came out with Core 2 after the Netburst fiasco.

 

This is where the most interesting part about this article comes from (and yes this is KitGuru doing this), their source for this information (releasing big die first and skipping small die) is a forum where a forum poster claims they have inside information. 

[patrickjp93]

This is where the most interesting part about this article comes from (and yes this is KitGuru doing this), their source for this information (releasing big die first and skipping small die) is a forum where a forum poster claims they have inside information. 

ROFLCOPTER&SAUCE!!! MY SIDES!!! :lol: :lol: :lol: :lol: :rolleyes: <_<

 

But seriously? I know Nvidia has money, but good lord...

 

Oh, I know him. Yeah, he's full of crap. He was one of the wingnuts swearing AMD had a 3072-core Hawaii chip ready for the 780TI.

[GoodBytes]

The problem with these kind of rumors is that it is not the reality. Often times Nvidia and AMD waits for the other manufacture card to get released before releasing there, to know if they can tweak their card (say: overclock) to be more competitive, and not destroyed. Sometimes it can go for months.

So for example, if AMD has super hyped out rumors about some insane new architecture, you get bet that this early 2016... wont' be. If AMD releases a new GPU that beats the 900 series, and affects sales, you can be sure that the Pascal will be release much sooner.

Keep in mind that Pascal is already done, as it is based on Maxwell. Volta is the full new architecture.

[Guest Kloaked]

I think everyone is missing the point of this article, the point of this article is not that Pascal is due for 2016. It's the fact that they are eliminating the small die and going directly to the big die. So just to refresh your memory, with recent events, Maxwell launched with GM204, which is a small die ("mid range"). Then GM200 launched as the big die. The big die typically being 20-30% faster than the small die.

 

Here in this article they are saying that NVIDIA is completely skipping the small die and going directly to the big die for Pascal launch. So there will be no mid-range graphics card launched as a premier product like in the past (eg. GTX 680 GK104, GTX 980 GM204). Instead they will be launching their GP100 right from the get go (big die Pascal). For those who are wondering what GP100 is, it will be the equivalent of GM200 (Titan X / 980 Ti) or GK110 [originally called GK100] (Titan Black / 780 Ti); except with the Pascal architecture. As opposed to launching GP104 (eg. GM204) first, then launching GP100 (eg. GM200) later. 

 

That will be the card that I buy.

 

4k gaming next year is going to be amaze.

[BiG StroOnZ]

That will be the card that I buy.

 

4k gaming next year is going to be amaze.

 

People excited about 4K resolution, and I can't wait to get a 2560x1600 monitor. 

[Guest Kloaked]

People excited about 4K resolution, and I can't wait to get a 2560x1600 monitor. 

 

Ha. I'm getting a G-Sync one too.

 

Actually going to build a whole new system. Gonna be osim

[Nexxus]

Serious this. All of this. I came here just to post this gif im waiting for pascal to get rid of the 680s

[SurvivorNVL]

Serious, this all of this I came here to post this im waiting for this to get rid of the 680s

I'll probably get a pair of Kingpin 980ti's to hold me over until Pascal.     That should make 3440x1440 beautiful *_*

[Qwertious]

TSMC 16nm vs Samsung/GoFlow 14nm... who wins?

[Opcode]

Sounds interesting 14/16nm finfet is very welcome and interesting. However, I am worried about HBM. Looking at what AMD went through with interposers, and connecting both HBM and gpu to this interposer, I really have to question, if NVidia can pull that off without a hitch. Especially with HBM 2. Further more, will the yields be high enough if AMD has first right to these chips?

 

Either way, 2016 will be very interesting in the processor world. Yay for the consumers!

AMD opened up all of their innovation in making HBM functional with GPUs. Nvidia will undoubtedly leverage it to make it happen for their own GPUs saving them the work. Nvidia tried a long time ago at getting HBM onto their graphics cards and it apparently did not work out for them as the original TITAN was suppose to have HBM.

[patrickjp93]

AMD opened up all of their innovation in making HBM functional with GPUs. Nvidia will undoubtedly leverage it to make it happen for their own GPUs saving them the work. Nvidia tried a long time ago at getting HBM onto their graphics cards and it apparently did not work out for them as the original TITAN was suppose to have HBM.

Nvidia bet on HMC first, and Micron decided to not compromise by going 2.5D first. It's just as well. Intel's ordered enough of the stuff to keep all of Micron's foundries going full tilt through the end of 2016. 80 million units is nuts even by Intel's standards.