NVIDIA to Unveil "Pascal" at the 2016 Computex

[FordGT90Concept]

AMD has always been horrible at marketing and branding. AMD has also repeatedly made very bad decisions (like the one to aquire ATI in the first place when their position in the CPU market had a very grim outlook). A lot of it does fall on AMD itself and their desire to not change the status quo. At least Zen brings some hope that the culture of AMD is changing...but that remains to be seen.

 

[vega22]

i think the earthquake that hit tsmc shows that god loves amd and that nvidia is the work of the devil :lol:


*runs away*

 

[HumanSmoke]

i think the earthquake that hit tsmc shows that god loves amd and that nvidia is the work of the devil :lol:
*runs away*

The earthquake supposedly affected Fab 14 which is a 20nm SoC plant making Apple A9, other ARM ICs, and FPGA's.
Fab 14(B) - the 16nmFF+ extension fab where GPUs would be manufactured was back up and running quickly.

Techpowerup covered the story but neglected to research what process the fabs actually used for their production. Maybe "TSMC Damaged by Earthquake, Could Impact AMD and Nvidia GPU Production" would get more page views than"TSMC Damaged by Earthquake, Will Impact Apple A9 Production"

"God" might just be evening the score. Visit a natural disaster on Nvidia to offset all the man-made disasters that AMD have had visited upon them and have initiated themselves. If that is the case then AMD better pray for a higher level of divine intervention I suspect.

 

[medi01]

AMD is likely to have up to 32 GB on HBM2 which translates to 8 GB per stack. NVIDIA will likely offer the same.

Dafuq would one need that much RAM for??? 8k gaming?

AMD's high power consumption more than erases their cost advantage in GPUs.

I'll talk about current gen, bar Fury (which has better perf/watt), it's 40-80w difference (+20% total system power consumption) while being 10% faster. A fair trade and more than competitive, to say the least.

Why do people still buy Nvidia products? Because they laid down the groundwork years ago...

Such as effectively bribing developers to cripple perf on competitor's product and pushing even more crap time from time, from PhysX (which at least has its uses) to G-Sync (which is a stinky piese of shit).
Groundwork, my ass...
A GPU should be not too noisy, not too power hungry. It should run games smoothly.. I've just listed ALL points average gamers care about.
I could add "oh, and it should last" but then NV definitely cannot boast that, can it? And I'm not talking about DX12.

 

[the54thvoid]

A GPU should be not too noisy, not too power hungry. It should run games smoothly.. I've just listed ALL points average gamers care about.
I could add "oh, and it should last" but then NV definitely cannot boast that, can it? And I'm not talking about DX12.

By that definition, AMD's Hawaii architecture was an abortion. Nvidia haven't messed up that criteria since the GTX480 bacon maker. Which was when I went AMD with 5850's, which were nearly the best bang for bucks (IMO) I've ever bought. It's a shame in crossfire they broke the 3rd rule you listed, run games smoothly.
So, Hawaii fixed rule 3 with PCI-e crossfire but annihilated 1& 2.
Post GTX480, Nvidia haven't really dropped any design balls, 580, 680, 780, 780ti, 980ti. What they have done is create an entirely weird (but not new) price structure with pseudo professional Titan parts. The Titan Z, I will acquiesce to as being hysterically awful.
People can berate Nvidia cards if they want to live in a red misted wonderland but Nvidia make good gaming hardware. And you can't blame Gameworks. There are many AMD titles that play fastest on Nvidia hardware and on brand agnostic games, Nvidia still performs better. Let's not delude ourselves.

As far as being OT, if Pascal is found to be lacking in Asynch (if the warp scheduling is still serialised and not parallelized) then yes, later this year we may well see a true situation where AMD will clearly win some titles and those that don't rely on Async may go Nvidia way. I just hope both brands have good enough cards to avoid game specific performance problems, even when one is far better. A situation where a top tier card doesn't perform in a game because of an architecture deficit which switches title to title would be disastrous.

 

[Ithanul]

This seems to go right along with what I've been saying for 6 months: don't expect Pascal before the last few months of this year. Which means, likely just like with Maxwell, the more affordable mainstream cards will not be released until Jan/Feb, just like 960.

It's still a long time to wait for anyone that feels they need an upgrade right now.

Indeed, reason I'm waiting out again like I did with Maxwell. Then wait out for Pascal to hit used market. I'm a cheap bugger. All my cards are 2nd owned.
At least this means I will get a good year or two out of these Tis before the next Ti shows up.

Polaris...does...kick...ass...it will probably debut before Pascal too.

I sure hope so, I would love to play around with an AMD card again. But right now they suck major for folding compared to Maxwell cards. Kind of miss folding on the 7970 I had, but that thing was very hot to run 24/7 on an air cooler. But hell, I give that thing was a tank considering it was handling running over 70-75C 24/7 with a OC on it.

Dafuq would one need that much RAM for??? 8k gaming?

No kidding, unless you doing some crazy high renderings or something. I could care less about a buttload of RAM since 4k or 8k does not interest me. Give me a good high core clock and ocing ability plus performance/watts. Then I be happy as a lark while folding the crap out of the cards.

 

[FordGT90Concept]

With that kind of VRAM, they could load all of the game textures into the VRAM and leave them there freeing up the system RAM to do whatever they want.

 

[rtwjunkie]

By that definition, AMD's Hawaii architecture was an abortion. Nvidia haven't messed up that criteria since the GTX480 bacon maker. Which was when I went AMD with 5850's, which were nearly the best bang for bucks (IMO) I've ever bought. It's a shame in crossfire they broke the 3rd rule you listed, run games smoothly.
So, Hawaii fixed rule 3 with PCI-e crossfire but annihilated 1& 2.
Post GTX480, Nvidia haven't really dropped any design balls, 580, 680, 780, 780ti, 980ti. What they have done is create an entirely weird (but not new) price structure with pseudo professional Titan parts. The Titan Z, I will acquiesce to as being hysterically awful.
People can berate Nvidia cards if they want to live in a red misted wonderland but Nvidia make good gaming hardware. And you can't blame Gameworks. There are many AMD titles that play fastest on Nvidia hardware and on brand agnostic games, Nvidia still performs better. Let's not delude ourselves.

As far as being OT, if Pascal is found to be lacking in Asynch (if the warp scheduling is still serialised and not parallelized) then yes, later this year we may well see a true situation where AMD will clearly win some titles and those that don't rely on Async may go Nvidia way. I just hope both brands have good enough cards to avoid game specific performance problems, even when one is far better. A situation where a top tier card doesn't perform in a game because of an architecture deficit which switches title to title would be disastrous.

Calm, cool, thoughtul and insightfull response. Thank you.

 

[PP Mguire]

The non-cutdown Pascal and Polaris chips will no doubt run for at least $600 USD but that's normal. AMD has a card that competes with NVIDIA at every price point except Titan-Z but that's coming with the Fury X2.


AMD is not "hanging by a thread" in the graphics department. They have 20% marketshare in the discreet card market and 100% of the console market.

The 295x2 made easy work of the Titan-Z. The Z was a really stupid card. It's why I own the 295x2 and not the Titan-Z because the Z is just a flat embarrassment.

 

[HumanSmoke]

Such as effectively bribing developers to cripple perf on competitor's product and pushing even more crap time from time, from PhysX (which at least has its uses) to G-Sync (which is a stinky piese of shit).

PhysX and game dis/optimization has no bearing on the vast majority of users, or OEMs who sign big contracts. That should have been apparent when I actually stipulated such

but it is always tough to topple a market leader if that leader basically delivers - I'm talking about delivering to the vast majority of customers - OEMs and the average user that just uses the hardware and software, not an minority of enthusiasts whose presence barely registers outside of specialized sites like this.

...and regardless of how you view G-Sync (and more than a few people don't share your view), OEMs seem quite happy to market it, customers seem quite happy to use it, and from a marketing point of view being first to market counts for a lot. Adaptive Sync/Freesync is cheaper, but like G-Sync it isn't a mainstream user priority.

Groundwork, my ass...

Yet it's fact that Nvidia's game development program raised the companies profile (especially when personal computing in general and PC gaming in particular were going through its expansion phase), so apropos of your comment, I'm going to say you are talking out of your one-eyed orifice.
I'd also ask you, if you think that the game dev software R&D has no merit, why would AMD spend resources copying the features? The ideas of GeForce Experience, frame pacing, G-Sync, Shadowplay, Optimus and a host of other features of varying merit have all been appropriated by AMD. If they don't add value to the brand do you think that AMD are just pathological imitators? Most neutral observers would probably note that the features help sell the brand and the hardware, and AMD would be foolish not to exploit any opportunity to advance both.

 

[the54thvoid]

The 295x2 made easy work of the Titan-Z. The Z was a really stupid card. It's why I own the 295x2 and not the Titan-Z because the Z is just a flat embarrassment.

Frankly the best iteration of the whole Asus uber cards was this one.

Absolutely gorgeous piece of hardware. If Nvidia had allowed the board partners to do the same to the Titan Z (and of course not be so expensive) that round would have been awesome.

 

[PP Mguire]

Frankly the best iteration of the whole Asus uber cards was this one.

Absolutely gorgeous piece of hardware. If Nvidia had allowed the board partners to do the same to the Titan Z (and of course not be so expensive) that round would have been awesome.

Yea ultimately cost and super low clock speeds made the Titan-Z a turd. It's stupid of Nvidia not to let AIB's play with the Titan class cards.

 

[mcraygsx]

Is it safe to say we will see 1080 Ti or Titan version in Summer 2017 and not before ? Mind the naming scheme.

 

[DarthBaggins]

Really can't wait for big pascal and the next line from amd, fan of both manufacturers as each have their string points. Also AMD got me to get back into the PC gaming race

 

[medi01]

By that definition, AMD's Hawaii architecture

I recall 780 Ti used to be faster than 290x? Ironic.

Nvidia make good gaming hardware

Yes. And that's good. But then there comes proprietary "only me" crap with it, which is bad for the market as a whole.

OEMs seem quite happy to market it, customers seem quite happy to use it

So they were with Prescott (all of 'em, OEMs and customers). This point is moot.

I didn't quite get exactly what you meant when talking about "delivering"... Delivering what?

Adaptive Sync/Freesync is cheaper, but like G-Sync it isn't a mainstream user priority.

It isn't cheaper, it comes free (with most scaler chips). G-Sync users pay 100$ for something, the only point of which is to ban NVs competitors. Oh, and it limits monitors to only one port. Very convenient.

Yet it's fact that Nvidia's game development program..

PhysX was BOUGHT and forcefully made exclusive. At best it is "NV bought game development program".
Then you slap another nice sum to bribe devs to use it, and, yay, it's sooo good for customers.

 

[PP Mguire]

Is it safe to say we will see 1080 Ti or Titan version in Summer 2017 and not before ? Mind the naming scheme.

My money is on this time next year, and I hope I'm wrong.

I recall 780 Ti used to be faster than 290x? Ironic.


Yes. And that's good. But then there comes proprietary "only me" crap with it, which is bad for the market as a whole.


So they were with Prescott (all of 'em, OEMs and customers). This point is moot.

I didn't quite get exactly what you meant when talking about "delivering"... Delivering what?


It isn't cheaper, it comes free (with most scaler chips). G-Sync users pay 100$ for something, the only point of which is to ban NVs competitors. Oh, and it limits monitors to only one port. Very convenient.


PhysX was BOUGHT and forcefully made exclusive. At best it is "NV bought game development program".
Then you slap another nice sum to bribe devs to use it, and, yay, it's sooo good for customers.

Gsync brought the tech to the table where Freesync was vapor at conventions. You have to be daft to not realize Gsync pushed the tech into the hands of the consumers and made Freesync come out much quicker.

Newer Gsync monitors have more than one input.

 

[HumanSmoke]

Gsync brought the tech to the table where Freesync was vapor at conventions. You have to be daft to not realize Gsync pushed the tech into the hands of the consumers and made Freesync come out much quicker.

QFT, although I suspect any reasoned argument is lost on medi01. He seems to have lost the plot of the thread he jumped on - which was about the various companies position in their respective markets and how they arrived at the present situation

PhysX was BOUGHT and forcefully made exclusive. At best it is "NV bought game development program" Then you slap another nice sum to bribe devs to use it, and, yay, it's sooo good for customers.

So what? The philosophical debate over the ethics of PhysX doesn't alter the fact that Nvidia used its gaming development program to further its brand awareness. They are two mutually exclusive arguments. Do me a favour - if you're quoting me at least make your response relevant to what is being discussed.

 

[rruff]

AI is driving development of GPUs: http://www.itworld.com/article/2898...fold-performance-jump-with-next-gpu-tech.html

"Nvidia said it will offer up to 32GB of RAM per GPU. This will allow for up to five times better performance in what Nvidia calls "deep learning applications" which are applications capable of gathering data and learning to recognize patterns or images. It's also a sign that this card will be for high performance computing, as the majority of video cards have just 2GB of memory."


http://blogs.nvidia.com/blog/2016/02/23/pratt-gtc-toyota/
"And GPUs, which are key to training a new generation of machines with superhuman capabilities, are at the center of this story (see “Accelerating AI with GPUs: A New Computing Model”)."

Someone is willing to pay a lot of money for this stuff...

 

[HumanSmoke]

AI is driving development of GPUs: http://www.itworld.com/article/2898...fold-performance-jump-with-next-gpu-tech.html

Ten times the performance is easily doable just by reinstating the 1:3 rate FP64 of GK110/210 (as opposed to Maxwell's 1:32). Pascal also has verified half-precision (FP16) support - if it also has quarter-precision (FP8) support, that would also more than do it. FP16 has at least some gaming application unlike double precision.

 

[cyneater]

So when are they going to release turbo pascal ... and delphi ... bom tish

 

[medi01]

So what?

So that world would have been better, had nVidia NOT bought it.

There was NO NEED in GSync the way it was done, there was nothing special about variable refresh rate, that stuff was already there in notebooks (that's why it didn't take AMD long to counter). The only drive (and wasted money) was to come out with some "only me, only mine!!!" shit, nothing else.

Had it been a common, open standard, that would have pushed market forward a lot. But no, we have crippled "only this company" shit now. Thanks, great progress.

It's great to have more than one competitive player in the market. It sucks when they play dirty, the way nVidia does.

Strong arm politics all over the place on all fronts: XFX, hell, ANAND BLOODY TECH. Punished, learned the lesson, next time put cherry picked overclocked fermi vs stock AMD. And that's only VISIBLE part of it, who fucking knows what's going on underneath.

 

[HumanSmoke]

So that world would have been better, had nVidia NOT bought it.

There was NO NEED in GSync the way it was done, there was nothing special about variable refresh rate, that stuff was already there in notebooks (that's why it didn't take AMD long to counter). The only drive (and wasted money) was to come out with some "only me, only mine!!!" shit, nothing else.

Had it been a common, open standard, that would have pushed market forward a lot. But no, we have crippled "only this company" shit now. Thanks, great progress.

It's great to have more than one competitive player in the market. It sucks when they play dirty, the way nVidia does.

Strong arm politics all over the place on all fronts: XFX, hell, ANAND BLOODY TECH. Punished, learned the lesson, next time put cherry picked overclocked fermi vs stock AMD. And that's only VISIBLE part of it, who fucking knows what's going on underneath.

 

[Frick]

Frankly the best iteration of the whole Asus uber cards was this one.



Absolutely gorgeous piece of hardware. If Nvidia had allowed the board partners to do the same to the Titan Z (and of course not be so expensive) that round would have been awesome.

I loved the Ares series (wasn't there another company that made some similar things btw? very high end, premium dual cards) - dual GPU cards always were a hoot to me - but the latest one really was something else.

 

[HumanSmoke]

I loved the Ares series (wasn't there another company that made some similar things btw? very high end, premium dual cards) - dual GPU cards always were a hoot to me - but the latest one really was something else.

Non-reference dual cards? They've been a staple for a while. Asus's Ares and Mars cards are just the latest iterations. Both Gigabyte and Asus offered dual GF 6000 series boards, GeCube's X1650 XT Dual, and a few EVGA models. As for premium dual waterblocked cards, a few AIB's slap a Swiftech or EK block on their products - even hybrid coolers go back a fair way

 

[Frick]

Non-reference dual cards? They've been a staple for a while. Asus's Ares and Mars cards are just the latest iterations. Both Gigabyte and Asus offered dual GF 6000 series boards, GeCube's X1650 XT Dual, and a few EVGA models. As for premium dual waterblocked cards, a few AIB's slap a Swiftech or EK block on their products - even hybrid coolers go back a fair way

Yeah I know but I meant a series of them. It might have been Mars I was thinking about.. For some reason I was thinking about Sapphire.