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

maybe they just calculate the percentage of the resolution indicators horizontally and vertically: 1920 * 200% = 3840 / 1080 * 200% = 2160, thus 3840*2160

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

NVIDIA Pascal will make its first appearance at GTC in April and will launch at Computex! They are returning to the 700 series launch schedule. The NVIDIA GTX 780 launched on May 23rd, and the 780M launched one week later. We can expect a similar schedule this year for Pascal.

Here's a leaked picture of the new shrouds:



"The leaked images which we have today are of two graphics cards: the GeForce GTX 1080 and the GeForce GTX 1070. Now first thing first, these images aren’t modified but the two shrouds are actually real. The entire cooler shroud appears to be made of a metal, we aren’t really sure about that, but NVIDIA uses Magnesium Alloy and Aluminum on their current reference NVTTM coolers. The other thing is that both shrouds feature a new design scheme that looks like an updated version of the current NVTTM cooler." - Source

 

What makes you think they're fake HTWingNut, did you notice any giveaways?

 

I think most people think it's fake simply because they're expecting Nvidia to change the nomenclature to something other than 1080.
I'm thinking they are real but wouldn't be surprised either way.

 

While they look good, I'd think that CNC milling, that we can see in the background, is a very expensive process, especially for a decorative part of the card. I think such parts are either molded or stamped.

 

Ah, this might be the giveaway that the pic is a fake then, unless CNC milling is used to create prototypes early on in the development process.

 

Yes that's possible. There other thing is that GTX 1080 is a very odd name, I think it would confuse customers. Also it's bad for SEO, if you search for GTX 1080 in Amazon you'll get many laptops with GTX cards and 1080p display.

 

Yep, I really don't think it will be 1080. Those do look like CNC made parts and not a stamped or molded part like it would be for a production part.

 

Have you considered they will have two editions of the cards: a special edition and regular edition. The special edition having this unique shroud.

I don't think it's fake. Why would someone spend millions producing fakes (plural)? An industrial size machine is approximately $250,000. If it's fake, some billionaire must be very bored, indeed. Hiring skilled workers, designers, etc. and leasing space to put the machine, all to make a fake shroud seems unrealistic.

You are all so pessimistic.

 

I mean, some people will go lengths to produce fakes to generate some attention. Fakes are getting more and more crafty.
That said, they may not invest 6 digits for fakes.

 

I'd say it's skeptism rather pessimism. A high end 3D Printer could probably pull off a pretty good fake.
I have a feeling they will infact be called 1080. I don't believe for a second anyone buying a GPU to install in their machine is going to have a brain fart and purchase a 1080p TV instead.

 

They need to release the freaking benchmarks fasters.

I could not care less for anything like shape, name, price, cores number, power, technology used. Those are useless.

I want dat benchmark. ASAP. If they can go over the max oc'd benced titan x with a stock 980 relative gpu, then it is time to oc that, and time to be actually enthusiased about next titan.
​

Talk about excitement.

 

Yeah, I'm sure. But as it is very clear the machinery used in this (seen in the background), it's unlikely a fake.

That's true. A $10000 - $20000 3D printer could probably make a nice fake. However, we see the machinery in the background. It's clearly an industrial size CNC machine, which isn't available to the general public and is quite large. You need special licensing to even own one of these.

If I had to speculate as to what these were, I'd say NVIDIA plans to make a limited edition (or launch edition) of the new GPU's and they will feature special shrouds. It's not unheard of for new product launches. NVIDIA holds 67%~ market share in a multi-billion dollar industry. A limited edition product would cost them pennies on the dollar to make - deals having been made with manufacturers, etc.

 

People use work equipment for personal use all the time. It's not like they bought the milling machine just for this.

Sent from my VS980 4G using Tapatalk

 

If that's the case, he's a talented fellow. I think it's a great shroud.

My hope is that the rest will be magnesium alloy, glass, etc. all very high-quality pieces. Painting that shroud would also be quite fun. Imagine having a gold GPU with a gloss finish. Would be so nice to have.

 

No news at all and then suddenly full release in less than 2 months?

Sent from my SM-A500FU using Tapatalk

 

So exactly like the old titan shroud and the exact same shroud on every new high end GPU.....

I am more interested in a better AIB made PCB then some fancy shroud.

Here is some facts:
http://www.anandtech.com/show/10017/micron-reports-on-gddr5x-progress
GDDR5X production wont start until June 2016, which means whatever card this is, it cant be using GDDR5X. So its either GDDR5 or HBM2.

Something tells me its probably not going to be a gaming HBM2 on a GP104 more its going to a Tesla GP100 on HBM2.

Or even a GM106 on GDDR5 is likely

But whatever we are getting, its not going to be based on GDDR5X.

 

Lol exactly what I was thinking, slight facepalm at the same time

Sent from my Nexus 6P using Tapatalk

 

HBM2 has been in mass production since January, so I'd expect it to be that. - Source

GDDR5X will likely be for mobile cards (e.g. 1060M) or Quadro. Would be a waste to manufacture HBM2 on low-end mobile graphics that can't utilize it, and Quadro that doesn't need it. GDDR5X almost doubles bandwidth and halves bus width.

My hope is that they launch the Titan and announce the rest at GTC, which would put them on track for a Computex launch for desktop, and then mobile soon thereafter. However, it wouldn't surprise me to see HBM2 only offered in the Titan until next year.

 

I wouldnt be so certain about GP100 right out of the gate on a brand new process.

The only time nvidia did that backfired horribly on them.

 

so if the biggest thing that matters according to some people on here is brute gpu power, then whats the point of HBM? why do we need faster memory?

 

Its part of the "brute GPU power". For one, you need to feed the GPU.

 

Sources to Sweclockers claim the GPUs that´ll launch at Computex will be the successors to 970 and 980 and possibly 980Ti. They will use GDDR5X. Titan X will still be available.

https://translate.google.se/transla...av-pascal-med-gddr5x-till-computex&edit-text=

 

It's April fools day. Be warned!

 

Anything coming up in the news today is 100% inaccurate! Ignore it

 

http://www.geforce.com/whats-new/articles/nvidia-announces-new-titan-mini

"Titan Mini isn’t simply smaller – its better in every way. Smaller, yet dramatically faster. More powerful, but remarkably power efficient. With a rugged titanium surface, it’s a one size fits all hardware experience without compromise to performance or color, creating a new generation of GeForce GTX-powered PCs that are smaller than ever before.

2x the performance
2x the power efficiency
5x the color options
.5x the size

Check our store regularly for availability."

April 1st Announcement, Release Date...Half-past Never

 

Aw, I want one though.

 

Such April, Much fool, very colors, wow

 

Although, thinking about it, they'll be able to do this with Pascal, if they ever launched such a thing then they could half joke during the launch that their April fools wasn't that far off!

 

Is it really so far fetched that Nvidia will only show/announce a few chips rather than the full fledged series?
Or that they will have GDDR5X, which according to other sources won't go into mass production until summer, pushing actual availability for Pascal to fall/late 2016?

 

Am I the only person who thought that the Titan Mini looks like cameras? (but yes it's just an April Fool's joke)

Hadriel

 

Gddr5x provides more than enough bandwidth for a pascal based mobile card. Who knows they might make bandwidth usage more efficient too.

Sent from my SM-A500FU using Tapatalk

 

You should read my post at the top of the previous page. They're not waiting for GDDR5X for the 1070, 1080 or Titan.

 

I'm still trying to understand where GDDR5 was not enough. I mean, the main bottleneck was noever vRAM AFAIK.

 

Some exciting benchmarks for the new Pascal flagship:

 

I'll miss April fools day.

Not really.

 

ive heard that gddr5 hurts 4k and beyond performance

 

Please tell me I will not get warns for posting these shocking videos with nVidia's best engineer and possibly next CEO talking about the new problems faced by the great company

In this video the engineer speaks about nVidia's position on enabling free G-sync on non compiling monitors:





In this video, the same engineer tried to apologize and explain what the situation was actually with the GTX970 having only 3.5vRam






In this video, the same engineer speaks his mind about nVidia's position on the new madbooks, and why nvidia thinks that they should had included discrete GPUs






In this video, the same engineer speaks about the reason behind creating GTX960 and GTX960TI







Now, the engineer seems to also have been working for Dota2 back in 2015, and gives a speech about dota2 problems back then






Now, if you are curious how this great engineer started his life, you should watch this video, where he talks about his wonderful work as a Sennheiser engineer for headphones, (HD800S is an audiophile open back headphone that costs around 2000$)





Now, NBR, please don't ban me, but it was april fools

 

Nobody is a big fan of the greatest engineer who ever made a speech for the entire world? XD


EDIT::: All of the videos are just so funny for me

 

GDDR5X is an alternative to GDDR5 but is still obsolete in regard to the growth of GPU technology, which is why companies will not rely heavily on it and have focused primarily on the development of HBM (High-Bandwidth Memory). This document quite clearly explains the benefits of HBM over GDDR5. This should answer any skepticism, doubts, questions or concerns. It would be a mistake for NVIDIA to use GDDR5X across their entire lineup.

Spoiler: HBM v GDDR5

The above image is a primary reason for the use of HBM technology. HBM2 is in mass production and has been since January. They'll most likely use HBM2.

 

My understanding is that HBM is stacked, which means memory also goes in vertical plain. This might also mean higher operating temperature. This means that new GPU needs stronger cooling, and life expectancy might be shorter. But if they pull it off without these drawbacks, we should all rejoice.

 

HBM consumes less power than GDDR5, roughly half if memory serves, and with it being right next to the die itself, cooling it shouldn't be an issue.

 

I see you make predictions based on production and availability of memory and what would make most sense for Nvidia. This might very well turn out true, but it isn't exactly an official fact.

My prediction is this first iteration of Pascal will have GDDR5X and replace 970 and 980. HBM/2 will come with the big guns next iteration/generation.

 

If they use it, I think it'll be for lower-end GPU's, mobile GPU's, or Quadro and not the flagship desktop cards.

HBM2 is next generation technology. GDDR5X seems more like a worthy placeholder for MXM 3.0b. It seems likely to me that they'll either go full-pledged HBM across the board or tease us with HBM on the upper crust of GPU's.

 

I just read a few more reviews. All of it made me ask myself a few things.
First, with GTX960 / 670, 670 seems better with a few games and benchmarks.
Second, AMD already bought up HBM and it did not make a difference for them
Third, AMD bought more Cuda cores and it did not made a huge difference.
Lastly, Intel did the shrink thingy and again, it did not bring a major difference.
Adding all of these, if each of the changes would add 5-7% improvements (realistic expectations), it would mean that Pascal 80 series will be around 30% - 50% faster than 980, in the light nvidia might fix DX12, pull gameworks out and maybe raise the power consumption a bit.

I can't wait for the release of Pascal officially, so that I will know if to get one or not.

In the meantime, this serious nvidia engineer has given a touching word on DX12 and nvidia's support for it.





Now seriously speaking, nobody made a big fuss yet over them not supporting DX12? Is it even possible to add it by software, or do they need another hardware ?

 

You cant really look at things like this. Die shrink only improves performance only if the die size is consistent.
Not sure what you mean by this or that
"Second, AMD already bought up HBM and it did not make a difference for them"
"Third, AMD bought more Cuda cores and it did not made a huge difference."

AMD doesnt have CUDA cores and HBM does help quite a bit at higher resolutions. GM200 can be bottlenecked at 4k+ memory wise. HBM also cut down the area needed for large memory bus which does take up quite a bit of valuable die space.

I am not sure what you mean by Pascal not supporting DX12. I believe Maxwell and Pascal should support it. Now if you are talking about a specific feature in DX12, its a different matter.

 

I know that AMD does not use cuda cores, but they have their own version of cores, or pipelines, from what I know, and they have lots of those, yet it made little difference in real life performance.

Also, 980 does not fully support DX12, and this is what I was curious of. I have no idea what exactly it does not support from DX12, but I know I read complaints, and I want to know more.

I did not know that GDDR5 can be a bottleneck at 4K. On the other hand, at 4K, the max refresh rate is going to be 60-65 Hz for laptops (this is NBR afterall )