Nvidia Pascal Announcement

Pascal Secrets: What Makes Nvidia GeForce GTX 1080 so Fast?
http://vrworld.com/2016/05/10/pascal-secrets-nvidia-geforce-gtx-1080/
In our initial talks with Nvidia and their partners, we learned that the GeForce GTX 1080 is coming to market in several shapes:

  • GeForce GTX 1080 8GB
  • GeForce GTX 1080 Founders Edition
  • GeForce GTX 1080 Air Overclocked Edition
  • GeForce GTX 1080 Liquid Cooled Edition
Stock GTX 1080 is clocked at 1.66 GHz, with Turbo Boost lifting it to 1.73 GHz. Founders Edition includes overclocking-friendly BIOS to raise the clocks to at least 2 GHz, and the presentation showed the chip running at 2.1 GHz. The main limiting factor for the overclocking beyond 2.2 GHz is 225 Watts, which is how much the board can officially pull from the power circuitry: 75 Watts from the motherboard and 150 W through 8-pin PEG connector. However, there are power supply manufacturers which provide more juice per rail, and we’ve seen single 8-pin connector delivering 225 W on its own. Still, partners such as ASUS, Colorful, EVGA, Galax, GigaByte, MSI are preparing custom boards with 2-3 8-pin connectors. According to our sources, reaching 2.5 GHz using a liquid cooling setup such as Corsair H115i or EK Waterblocks should not be too much of a hassle.
...
Search for performance lead the company to remove as much legacy options as possible, and you can no longer connect the GTX 1080 with an analog display. D-SUB15 is now firmly in the past, and you cannot make the connection work even if you use a 3rd party adapter. The rest of connectors include a 144Hz-capable DVI, three DisplayPort 1.4 and a single HDMI 2.0B connector.

In the search for absolute performance per transistor, Nvidia revised the way how their Streaming Multiprocessor works. When we compare GM200 versus GP100 in clock-per-clock, Pascal (slightly) lags behind Maxwell. This change to a more granulated architecture was done in order to deliver higher clocks and more performance. Splitting the single Maxwell SM into two, doubling the amount of shared memory, warps and registers enabled the FP32 and FP64 cores to operate with yet unseen efficiency. For GP104, Nvidia disabled/removed the FP64 units – reducing the double-precision compute performance to a meaningless number, just like its predecessors.

  • GP100: 15.3 billion transistors, 3840 cores, 60 SM, 4096-bit memory, 1328 MHz GPU clock
  • GP104: 7.2 billion transistors, 2560 cores, 40 SM, 256-bit memory, 1660 MHz GPU clock
What is there is single-precision (FP32) performance, which stands at 9 TFLOPS. While the GP100 chip needs a Turbo Boost to 1.48 GHz in order to deliver 10.6 TFLOPS, GP104 clocks up to 1.73 GHz and that’s not the end. If you clock the GTX 1080 to 2.1 GHz, which is achievable on air – you will speed go past the GP100. We can already see the developers and scientists that need single-precision performance placing orders for air and liquid cooled GTX 1080s.

For DirectX 12 and VR, the term Asynchronous Compute was thrown around, especially since AMD Radeon-based cards were beating Nvidia GeForce cards in DirectX 12 titles such as Ashes of The Singularity and Rise of the Tomb Raider. We were told that the Pascal architecture doesn’t have Asynchronous Compute, but that there are some aspects of this feature which qualified the card for ‘direct12_1’ feature set.
...
However, DX12 titles face another battle altogether, and that is delivering a great gaming experience. This is something where titles such as Gears of War Ultimate Edition or Quantum Break failed entirely, as Microsoft ‘screwed the pooch’ with disastrous conversions and limitations set forth by the Windows Store. Tim Sweeney event wrote an in-depth column on The Guardian stating what’s wrong with Microsoft. These days, game developers work hand in hand with both AMD and Nvidia in order to extract as much performance out of DirectX 12 as possible, which is needed for challenging VR environments.
...
Year and a half ago, after seeing that HBM1 is limited in capacity and that HBM2 memory won’t be available in real volume before 2017, Nvidia started to work with Micron’s team in Germany on building the ultimate performance GDDR5.
Manufactured in 20nm process, GDDR5X memory showed being overclocking-friendly even with the initial silicon. As the roadmap shows, the target was to hit the 10 Gpbs i.e. 2.5 GHz QDR. Given that the memory actually moves four times per cycle, it should be called Quad Data Rate, but the name GDDR SGRAM (Graphics Double Data Rate Synchronous Graphics Random Access Memory) was kept for continuity.

GeForce GTX 1080 has the memory clocked at 2.5 GHz but we do expect some of the samples clocking at 2.75-3.5 GHz (11-14 Gbps). That would raise the available bandwidth from 320GB/s to 352-448 GB/s and we do expect to see extreme overclockers pushing the memory even more. If Micron adopts 10nm process for GDDR5X, we’ll get to 4 GHz clock / 16 Gbps rather sooner than later.
http://vrworld.com/2016/05/10/pascal-secrets-nvidia-geforce-gtx-1080/
 
Last edited by a moderator:
interesting read not sure if they sacrificed IPC to increase frequency though ;)

IPC has more do with occupancy, utilization etc. They used an example of Intel CPU's, not a good example for how GPU's work.....
 
Last edited:
It will sell for the same reason tons of people bought a GTX980 when it came out: it's the fastest GPU out there and it's expected to remain that way for quite a while. ]

I'm not so sure, in other forums I see a lot of more interest for the 1070, and also complaints for lack of information about it's specs.

It's not a mid-range card. It's the highest-end you can buy. It will only become mid-end when there'll be something better.

As far as I can remember the new mid-range always eclipsed the previous high end.
But in the past, often the high end would launch first.
In a sence this also has happened now as the P100 is the new high end,
But since Tesla has surpassed the importance of high end graphics cards, we now have to wait.
 
Pascal Secrets: What Makes Nvidia GeForce GTX 1080 so Fast?
http://vrworld.com/2016/05/10/pascal-secrets-nvidia-geforce-gtx-1080/
vrworld said:
For DirectX 12 and VR, the term Asynchronous Compute was thrown around, especially since AMD Radeon-based cards were beating Nvidia GeForce cards in DirectX 12 titles such as Ashes of The Singularity and Rise of the Tomb Raider. We were told that the Pascal architecture doesn’t have Asynchronous Compute,but that there are some aspects of this feature which qualified the card for ‘direct12_1’ feature set.
I wonder what that's supposed to mean. AC is a „soft“ feature.
 
I wonder what that's supposed to mean. AC is a „soft“ feature.
This is a head scratch,
because all of the posts related to this subject that only ever so briefly mentioned in a Deep Dive presentation (not at the event) had to be wiped and yet this guy says he was told and looks ok to publish that (even though seems very likely under NDA otherwise we would hear more about that Deep Dive)....
It is fair to say it is not async compute in the context of GCN2, but there was a critical limitation in Maxwell according to that presentation that seems to be under NDA.

Also WTH is he going on about the close relationship between NVIDIA and Micron....
AMD also has the 10Gb/s GDDR5X which was shown back in February I think on one of their cards in a config-monitor app.
He specifically says:
, Nvidia started to work with Micron’s team in Germany on building the ultimate performance GDDR5
Well some exageration there, as it is equal to both companies.
Cheers
 
Last edited:
Still wondering who will buy this 1080.
The reference design is priced 150$ more as the 980 it replaces., 700$ vs 550$ (at launch).
People buying mid range don't have that kind of money to spend.
Most people buying high end already have a 980ti and will wait for the next new high end card.
So why the high price ? Low yields, lack of competition ?

The GTX 1080 reference design costs $599, not $700. The "mid-range" card is the 1070 and will cost $379 msrp. Don't expect either card to cost MSRP for very long
 
Last edited:
The GTX 1080 reference design costs $599, not $700. The "mid-range" card is the 1070 and will cost $379 msrp. Don't expect either card to cost MSRP for very long

Im sorry Andrew, but the Funders edition is the reference design 1080. Theres no other Nvidia reference design. AIB partner can if they want put a cheap pcb and cheap VRM for propose a 1080 at the MSRP floor, but this will surely not much happend and this will not been the reference design.

In fact, this MSRP dont mean much.. As AIB will propose any other 1080 gpu''s than the funders with their own "more efficient cooling" ( Asus DC thing, Triple X, etc ) with higher clock gpu's.

This reference/funders gpu are intended to live all the lifespan of the series and are allready annunced today by every board partner:

Funders edition explained:
http://www.guru3d.com/news-story/geforce-gtx-1070-1080-founders-edition-explained.html

Board partners announce GTX1080 Founders edition:
http://www.guru3d.com/news-story/board-partners-announce-gtx-1080-founders-edition.html

Custom edition ((overclocked and with home brand cooling ) are in general what launch the brand, as this is how there is competition and where they can differentiate from other brand, and this time again, this will no make a doubt that nearly every gpu's will be highly customised gpu's:

Gygabtyte tease their xtreme custom gaming 1080
http://www.guru3d.com/news-story/gigabyte-teases-their-custom-geforce-gtx-1080-xtreme-gaming.html

I dont know, but for now, i got the sad feeling that this MSRP is just a marketing thing. this price will stay on spec cheet, outside maybe some exception, im not quite sure we will see much of this MSRP on real shop between the customized high clocked partner version and funders reference gpu's.
Its allready confusing on some articles.

No doubt why we have read all and nothing after the launch about the funders edition on forums, some saying: its a special overclocked version at 2.1ghz, other, it have a vapor chamber ( every reference gpus have got vapor chamber for both AMD and nvidia since several generations. ).
And again then, It will be branded Nvidia only ( today, all partners release one, not different of the stickers on cooler we have got before ) etc etc.
 
Last edited:
That stated, let's talk about the cheaper MSRP models.

This is, in our analysis of the situation, nVidia's way of avoiding competing with its own partners in the space. The Founder's Edition will cost $700. The MSRP is $600 – so vendors like MSI, EVGA, ASUS, et al. can enter market with cards cheaper than nVidia's own, throw their own coolers on them, and overclock them differently. The vendors will exercise similar control and design/engineering over their versions of the GTX 1000 series as with previous generations. Also as with previous generations, nVidia's version of the card (now "Founder's Edition") uses heavy materials that can run-up the cost. That metal shroud will run-up the BOM more than a plastic shroud from an AIB.

NVidia wanted to land at the center of the stack, providing room for vendors to undercut nVidia reference – err, “Founder's Edition” – prices, but also allowing room for higher-end cards >$700. Drawing parallels to the GTX 900 series, a higher-priced card might be something like EVGA's GTX 980 Ti Hybrid, which ran ~$750-$770 at first launch. This was a marked increase against the MSRP of $650, but offered features which helped carve its own price bracket.
http://www.tweaktown.com/news/52056/rumor-amd-partners-showing-polaris-cards-computex/index.html

Wondering what will reviewers use as the base reference card for reviews --- the higher priced version or lowest possible price card.
 
Last edited by a moderator:
[...] a vapor chamber ( every reference gpus have got vapor chamber for both AMD and nvidia since several generations. ).
That's not quite true. 980 for example came without one, Titan/X/Black had one, 780Ti/980Ti as well, 780 not - IIRC.
 
That's not quite true. 980 for example came without one, Titan/X/Black had one, 780Ti/980Ti as well, 780 not - IIRC.

You are right i should have say, most of nvidia gpu's reference... The photos of the cooler have make me think it was effectively a VC one:
 
Last edited:
I'm not so sure, in other forums I see a lot of more interest for the 1070, and also complaints for lack of information about it's specs.
Of course, the market is big enough of a place for their to be room for both.

As far as I can remember the new mid-range always eclipsed the previous high end.
The 1070 is supposed to be about the same performance as a Titan X for non-VR workloads. Should be good enough for most at that price.

But in the past, often the high end would launch first.
Yes, that was the case with GK104 and GM204, which were the high-end at the time. And it's the same today with GP104, since P100 is currently MIA. ;)

In a sense this also has happened now as the P100 is the new high end,
But since Tesla has surpassed the importance of high end graphics cards, we now have to wait.
Tesla hasn't surpassed anything. It's a $100M/quarter business compared to $800M for GeForce. But the FP64 side of the Tesla business was getting very long in the tooth with K80, and this class of chips probably requires a lot more validation, so it made sense to give move that up. Tactical considerations don't have to mean strategic changes.
 
So, Micron posts a blog saying that GDDR5X is already in mass production. (https://www.micron.com/about/blogs/2016/may/nvidia-launches-gtx1080-with-micron-gddr5x)

This shouldn't be a surprise for anyone who bothered to pay attention, but it's funny to think about the cumulative amount of time that was spent worldwide arguing on forums that Pascal wouldn't be available until Q3 or Q4 because of GDDR5X unavailability. ;)

(Of course, rhat's nothing compared to the chatter about Polaris where showing demos behind a black curtain was ultimate proof of AMD being months ahead...)
 
So, Micron posts a blog saying that GDDR5X is already in mass production. (https://www.micron.com/about/blogs/2016/may/nvidia-launches-gtx1080-with-micron-gddr5x)

This shouldn't be a surprise for anyone who bothered to pay attention, but it's funny to think about the cumulative amount of time that was spent worldwide arguing on forums that Pascal wouldn't be available until Q3 or Q4 because of GDDR5X unavailability. ;)

"Available" can have a fluid definition when discussing these kinds of releases.

If you're willing to pony up the cash, then you could get a founder's edition 1080 before June, but something tells me that meaningful availability won't be achieved until a couple months later. I mean, the whole up-priced founder's edition wouldn't make as much sense unless Nvidia knew that supply would be tight ahead of time.

And there's recent historical precedent with the 680, another G*#04-based halo card that debuted a new process. It had pretty terrible supply issues for months. We all remember the outrage when the 690 was announced - a card that required, not one but, two previous GK104 GPUs.

I would compare this launch to the hypothetical October Vega launch that's being discussed in this thread. For both a May 1080 and an October Vega, I don't think anyone seriously thinks that meaningful supply would be available upon the initial launch.

In both cases, the launch might technically not be a paper launch, but it would be effectively that for most buyers.
 
Tesla hasn't surpassed anything. It's a $100M/quarter business compared to $800M for GeForce.
I have a feeling that this is going upwards quite a bit over the next few quarters. For 3 reasons. First: Kepler installations being due for an update, general uptake of accelerator usage in supercomputing compared to 3 years ago and the growing data mining business (and I'm counting everything AI-related here too), which could be massive due to Facebook, Google, Microsoft and others wanting a part of the pie.
 
I have a feeling that this is going upwards quite a bit over the next few quarters. For 3 reasons. First: Kepler installations being due for an update, general uptake of accelerator usage in supercomputing compared to 3 years ago and the growing data mining business (and I'm counting everything AI-related here too), which could be massive due to Facebook, Google, Microsoft and others wanting a part of the pie.
I think it will go up too. But it's not going to challenge the GeForce business in any way.
 
I mean, the whole up-priced founder's edition wouldn't make as much sense unless Nvidia knew that supply would be tight ahead of time.
The whole discussion about supply and availability is ultimately very unproductive: our only instantaneous data point is store availability, but that doesn't say anything about supply.

I do agree that supply will be tight, but only because demand will be incredibly high as well. Even if Nvidia has 100k units available worldwide on day one, it will still sell out, and websites will list out-of-stock, and people will say "see! I told you they have production issues."

And there's recent historical precedent with the 680, another G*#04-based halo card that debuted a new process. It had pretty terrible supply issues for months.
Yes, there was a very high demand for that GPU. And it shot up the steam ranks very quickly as well...

We don't know a thing about volume, yet that matters much more than supply constraints. What we do know is that Nvidia was willing to nuke it's gm200 and gm204 based product line. Nobody would do this if they didn't have an alternative available to sell.
 
Yes, that was the case with GK104 and GM204, which were the high-end at the time. And it's the same today with GP104, since P100 is currently MIA. ;).

Personally I see high-end as the biggest possible GPU die for a certain GPU generation and process.
You define it as the fastest GPU at a certain time. Any suggestion for making this distinction clear welcome.

Tesla hasn't surpassed anything. It's a $100M/quarter business compared to $800M for GeForce. But the FP64 side of the Tesla business was getting very long in the tooth with K80, and this class of chips probably requires a lot more validation, so it made sense to give move that up. Tactical considerations don't have to mean strategic changes.

Revenue may be lower, but margins are much higher. The first P100s will only ship in the DGX-1 box, packing 8, for a 129K $, so figure at least 10000$ a piece.
We may need to prepare for 15000$ Titans at which point I'll be interested in what AMD has on the table.
 
Back
Top