Pascal: What do we know? Discussion, Latest News & Updates: 1000M Series GPU's

NVIDIA 16nm Pascal Powered GeForce GTX 1080 Launching in May (2016) – Uses Full GP104 GPU With 2x Performance Per Watt and 8 GB Memory -Source


Now that Maxwell has hit the market, we find ourselves, once again, on the horizon of new technology. What do we know about Pascal? NVIDIA Updates GPU Roadmap; Announces Pascal | NVIDIA Blog


It would appear that Pascal will be "re-branded" until 2018 when Volta is launched (presumably in 2H 2018).

• • 3D Memory: Stacks DRAM chips into dense modules with wide interfaces, and brings them inside the same package as the GPU. This lets GPUs get data from memory more quickly – boosting throughput and efficiency – allowing us to build more compact GPUs that put more power into smaller devices. The result: several times greater bandwidth, more than twice the memory capacity and quadrupled energy efficiency.
  • Unified Memory: This will make building applications that take advantage of what both GPUs and CPUs can do quicker and easier by allowing the CPU to access the GPU’s memory, and the GPU to access the CPU’s memory, so developers don’t have to allocate resources between the two.
  • NVLink: Today’s computers are constrained by the speed at which data can move between the CPU and GPU. NVLink puts a fatter pipe between the CPU and GPU, allowing data to flow at more than 80GB per second, compared to the 16GB per second available now.
  • Pascal Module: NVIDIA has designed a module to house Pascal GPUs with NVLink. At one-third the size of the standard boards used today, they’ll put the power of GPUs into more compact form factors than ever before.

Supporting Sources:

• NVIDIA's next-generation GPU is called Pascal, and it's smaller, faster and more efficient
• Nvidia announces Pascal GPU, $3K Titan Z graphics card | PCWorld
• Nvidia unveils next-gen Pascal GPU for 2016 release - CNET

Spoiler: More images.

This thread is for discussion about Pascal. Feel free to share your opinions, feedback, etc. below.

 

Pascal discussion thread, eh? And so it begins... The technology train can be delayed, it can be diverted, but it can never be stopped.

I think the 3D memory is the greatest leap there, though I do wonder what impact the unified memory will have on console ports from PS4/XBO.

 

God do not let this thread get any steam! otherwise it will become a 10000000000 page thread by the time Pascal actually comes out

 

Better than having that many separate threads, lol. We all know that would happen.

@Cake: I agree. But I also think pretty much everything about Pascal is a huge leap forward. It's so much more than Maxwell.

 

God have mercy on those who want to torture themselves with this

 

Two... years... later...

"Pascal has been unofficially announced!"

Oh no. It's already going off-topic. HELP MODS!

 

I'm waiting for Pascal, not sold on all this Maxwell hooey!

 

If Maxwell is cheaper, then I may upgrade. But yeah, I'm still on the Pascal party boat.

 

Pascal is 2 years away if you trust their roadmaps. Who knows what will happen by then.
One thing for certain.. that's hell long to wait, and hell long to be speculating.

 

I hope you have time to actually play while speculating future GPU's specs.

 

It's still not really unified memory with the cpu. High bandwidth ram helps with AA (You can add even more ROPs and feed them) and/or 4k rendering/Dual 1080p rendering (occulas rift)

 

Come on guys .we all know there's no way your going to last waiting for pascal. That's like an alcoholic telling someone he don't drink anymore.lol

 

I'm going to wait until some good drivers are released and a few dozen game-play videos hit YouTube before deciding if I want Maxwell or not. So far, it looks good, but not $1k good. Anyway, this is about Pascal! Don't ruin it within the first two pages.

 

I think Pascal will be released Q4 2015 or Q1 2016, just like Maxwell is this year. That's about 12-16 months away. So, it's not that far off. They've updated the roadmap two times this year, and I'm sure they'll do so at least once next year. I think this because Intel has moved up Skylake to Q3/Q4 2015.

Intel's next-next-gen Skylake chips set for 2015 launch despite Broadwell delay | PCWorld

 

There is no reason you could not do an MXM version of it and it would be ideal for them to do so. You could have an auxiliary connection cable for NVlink with the board space saved with memory/SLI connector.

 

I hope it's not different.

 

Right. Maxwell is here now, and here through 2016 for the most part, so best jump on that for now and make most of it.

 

Something "new" to share:

Nvidia is ‘excited’ about ‘Pascal’ and next-gen process technologies | KitGuru

A lot of old information is also within this article.

 

Yeah Pascal is at least 12-18 months from production and in our little hands, so if you have the gall to wait until 2016 no matter how you slice it, all the best. I hope you have a decent GPU, because there ain't nothing a whole lot better coming in 2015.

 

I still violently believe this thread should be burned down and closed... Pascal won't be out till 2016 and we don't need a 100000000 post thread by then

 

I agree. We probably won't even see any hint of its physical existence for another year. /see you in 2016/

 

Q4 2015, fellas. See you next year. Even if cards don't ship until 2016, look at all of the Maxwell threads - discussion began far before.

This thread won't gain that much traction, Tom. You are the one going off-topic, which leads to more off-topic discussion. This thread was created to prevent dozens of other threads from being created.

 

You have no control over this.

 

Really? Odd... Thanks for the update.

 

Is NVLink supposed to go hand-in-hand with HBM? Because with the cards we have now they can barely saturate PCIe 2.0 x8, which is like 1/4 the bandwidth of PCIe 3.0 x16, so not sure what we'd need "5 to 12x PCIe 3.0" for exactly. Or maybe it's a purely for GPU compute.

 

Yes, Nvidia is rolling out NVLink and HBM at the same time with Pascal.

Nvidia has been pretty mum about it, but everything I've seen seems to indicate that NVLink is an alternative (keyword) to PCIe for HPC, while consumer will continue to stay with PCIe. Way too early to know what will happen in the long run.

AnandTech | NVIDIA Updates GPU Roadmap; Unveils Pascal Architecture For 2016

So:
Tesla = NVLink
Quadro and GeForce = PCI Express

Also, PCIe 4.0 is on the horizon. As always, it doubles bandwidth over PCIe 3.0.

 

Nvidia is pushing NVLink big time for supercomputers:

What Is NVLink? Making the World's Fastest Computers Possible (Video) | NVIDIA Blog
How NVLink Will Enable Faster, Easier Multi-GPU Computing | Parallel Forall

"Summit" supercomputer is supposed to come in 2017 (available for end-users in 2018), with Volta GPUs:

https://www.olcf.ornl.gov/summit/

 

InB4 J.Dre undercover mod.

 

Ok so NVLink would be mostly geared towards thoses who do GPU compute, makes total sense. For your average gamer it's a complete waste of technology (and $$$ most likely).

 

So whatever launch estimate was for Pascal (I think we calculated it to about early 2016), it just got delayed by few months. Troubles in TSMC 16 nm land:

TSMC delays mass production using 16nm FinFET process to Q3 2015 | KitGuru

Also their 20 nm process is supposed to be problematic, allegedly causing Qualcomm to switch to Samsung foundry:

TSMC to face tough questions on Qualcomm orders
TSMC sales likely to disappoint in 2Q15

 

Damn, 16nm just keeps on getting delayed....

 

Yeah, delays everywhere. But I'm fine with delays as long as the performance of new mobile cards matches previous generation desktop cards (e.g. 1080M = GTX 980). The 980M is close to the GTX 780, which is impressive in itself. Now that everything is soldered, it seems like "playing the waiting game" may be worth it depending on where you sit financially, and what level of performance you currently have.*

*If you're wealthy enough to upgrade each year and love spending money (or love being in debt) this does not pertain to you.

 

And it seems Nvidia finally may dump TSMC too. They are going to use Samsung foundry for 14 nm:

http://www.tomshardware.com/news/nvidia-samsung-14nm-apple-qualcomm,28493.html

I wonder if Samsung's 14 nm process will be suitable also for Pascal GPUs, not just Tegra SoC.

 

It's official! We have a wiki page!

http://en.wikipedia.org/wiki/GeForce_1000_series

With all the lawsuits NVIDIA is dealing with, I wouldn't be surprised if there were delays in announcements, lol.

 

so huuuuuuuge boost in 4k gaming performance?

 

Yawn... I don't expect anything from Pascal until at least Q3 2016. I bet the 900 series successors (and will it really be called 1000 series?) will still be Maxwell possibly on a smaller node. Maxwell still has room to grow and Nvidia will milk it for all its worth.

Those reference articles on that Wikipedia page are a year old (March 2014). I wouldn't really call it "official".

 

@HTWingNut: I was mocking Wikipedia.

On another note:

http://www.kitguru.net/components/g...-to-10-times-higher-performance-than-maxwell/

 

Yeah there's a good reason Universities don't allow Wikipedia referencing.

 

Is it going to murder MXM module of the notebook GPUs? That is all I care about right now

 

Not sure I understand what you're asking...

It is assumed that a single (1) Pascal core will offer the same performance as two (2) Maxwell cores. Translating that to performance, a good estimate of single GPU performance: 980M SLI = 1080M. This may seem ridiculous but we saw similar gains before going from 580M to 680M. NVIDIA has just been milking the architecture ever since.

BGA (soldered components) is seemingly inevitable.

 

Wow, if that's true then it would be good to upgrade my GPU to Pascal. Only if...

 

You completely missed the elephant in the room: 680M through 980M have all been 28nm. There's been no fab shrinks since 2012; that's the reason why the performance gains have slowed down in recent years.

 

I didn't miss anything. Read the last sentence. I know that.

 

OK, so you were using 'architecture' as a synonym for die shrinks. Fair enough.

 

We haven't always had a process and architectural change at the same time. Just look at Kepler to Maxwell, both on 28nm. In the good old days we even had die shrinks on the same architecture.

 

Aye, that was the point I was trying to make

 

Lol so true.. learnt that the hard way when I refrenced wikipedia in one of my lab reports 2-3 years ago... Although the info was correct, I was told not to reference wikipedia ever but go to the source where wikipedia was referenced from and make my reference from there.... So much more work lol..

 

http://www.slashgear.com/nvidia-gpu-pascal-detailed-for-2016-release-17373989/

This article claims the CEO of NVIDIA said Pascal will offer up to 10x the graphics performance of Maxwell. Am I reading this wrong or is this some kind of miracle?

 

That is "CEO Math", aka BS. Probably will be what you said above, 2x the performance of Maxwell.

 

I don't know, Pascal is a new architecture and a die shrink... Look at the performance increase nVidia got from Kepler to Maxwell while staying at 28nm!