Caporegime
That is a genuine fear I had but I have heard on the grapevine that Nvidia will be launching big from the off.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **
- Thread starter Boomstick777
- Start date
That is a genuine fear I had but I have heard on the grapevine that Nvidia will be launching big from the off.
Man of Honour
If NVidia have the tech ready they will launch it, if they don't AMD would leapfrog them anyway.
Once tech is available it does not stay secret for long.
Let's use simple words:
Nvidia has on the market something worth 10, AMD something worth 8.
If Pascal is worth 20, while the next AMD chip 12, wouldn't Nvidia be better off putting on the marker the inferior version, valued 15, to keep the extra performance when needed?
It's nothing special, it's probably what's going on with Intel and what happened with the first Titan.
Either that or me being very sceptic AMD will ever be able to have a stronger top card than Nvidia.
Nvidia has on the market something worth 10, AMD something worth 8.
If Pascal is worth 20, while the next AMD chip 12, wouldn't Nvidia be better off putting on the marker the inferior version, valued 15, to keep the extra performance when needed?
It's nothing special, it's probably what's going on with Intel and what happened with the first Titan.
Either that or me being very sceptic AMD will ever be able to have a stronger top card than Nvidia.
Man of Honour
AMD have got the fastest card on the block now and it totally thrashes a TX, 980 Ti or Fury X.
AMD have had the top card in the past quite a bit.
Caveat to that - nVidia have commercial contracts for big powerful GPUs for compute/supercomputer use and nothing to fulfil them with the Maxwell lineup so we will probably see the bigger ones sooner this time around.
Amd and Nvidia use the same or almost the process, unlike Intel they been well in front of Amd for quite sometime by a large margin.
Most of the time Amd and Nvidia are quite close to each other and like Kaap said they both know what the other doing most of the time anyway.
Most of the time Amd and Nvidia are quite close to each other and like Kaap said they both know what the other doing most of the time anyway.
Pascal will be big jump compared to current gen cards due to massive jump in process tech, but compared to competition of 2016 a lot of people might be surprised about Pascal. nVidia badly needs compute/workstation replacement for Kepler, so I doubt we will see Pascal being stripped out maxwell like desktop beast. nVidia most likely will not cut corners on DP performance this time. Maxwell is fantastic arch for desktop/gaming and the only way achieving this was to cut everything compute out of the chip. AMD went with all around arch this time (though a lot of people don't give credit to them, on how close they managed to stay in desktop performance race).
Also there is some indications that AMD will use 14nm FF, so hopefully we will see some amazing tech from both sides.
Soldato
There is also the posibility of NvLink replacing the SLI bridges.
And then there is AMD's Coherent Data Fabric (CDF) which will link Zen cpu's to Greenland Gpu's.
But it depends if these techs come to the desktop.
I could see AMD doing it since having a unified design cuts costs. And then there is the performance benefit of having 25gb's per link between CPU - GPU, also between GPU - GPU. While being at lower latency than PCI-E. so as stated on information about CDF, a 4x setup will have 100gb's bandwith between xPU's
Even if 50gb's over 2 links is overkill. that leaves however many other links on the cpu free for other peripherals and cards in the system.
so even if you had say 40 links available on the CPU. if it communicated using CDF between AMD components using 2 links for 50gb for 1 gpu. Then that leaves you with 38 other channels for SSD's etc using PCI-E. Since the Global Memory Interface on the amd parts can fall back to PCI-E from CDF if need be.
so instead of the current situation where you end up with say 16 links used straight away by a gpu, leaving you with 24 left for other peripherals/GPU's etc. Your processor will have far more links free in the end. From this perspective i could see them using CDF on the desktop because of these advantages on high end boards.
And then there is AMD's Coherent Data Fabric (CDF) which will link Zen cpu's to Greenland Gpu's.
But it depends if these techs come to the desktop.
I could see AMD doing it since having a unified design cuts costs. And then there is the performance benefit of having 25gb's per link between CPU - GPU, also between GPU - GPU. While being at lower latency than PCI-E. so as stated on information about CDF, a 4x setup will have 100gb's bandwith between xPU's
Even if 50gb's over 2 links is overkill. that leaves however many other links on the cpu free for other peripherals and cards in the system.
so even if you had say 40 links available on the CPU. if it communicated using CDF between AMD components using 2 links for 50gb for 1 gpu. Then that leaves you with 38 other channels for SSD's etc using PCI-E. Since the Global Memory Interface on the amd parts can fall back to PCI-E from CDF if need be.
so instead of the current situation where you end up with say 16 links used straight away by a gpu, leaving you with 24 left for other peripherals/GPU's etc. Your processor will have far more links free in the end. From this perspective i could see them using CDF on the desktop because of these advantages on high end boards.
Last edited:
Yerp, Pascal should be a tremendous. Hoping to get similar performance from a mid range (Priced) card to a 980 Ti. At lower power consumption would be happy with that. It's time for a proper shakeup.
Hope AMD have something good cooking up because what Maxwell has done on 28nm is impressive, god knows how good Pascal on die shrink with HBM 2.0 will be
Last edited:
Associate
- Joined
- 8 Jul 2014
- Posts
- 117
Can we expect the 1080ti to be out during the launch of Pascal?
As I'll be looking to upgrade as soon as Pascal hits and would much rather go with a full fat GP100 chip than a GTX 1080... As I'll be gaming at 2560x1440 then.
6700k + Pascal + DX12... Just wow.
As I'll be looking to upgrade as soon as Pascal hits and would much rather go with a full fat GP100 chip than a GTX 1080... As I'll be gaming at 2560x1440 then.
6700k + Pascal + DX12... Just wow.
Associate
Like most posters I would like to see a Pascal based GPU Q1 but it will probably be more like Q2/Q3 with volume around Q4 2016 and realistic prices for the average user.
It is very unrealistic to hope for nvlink or AMD fabric to come to desktops as pcie replacement. For one motherboard manufacturer will not cater 2 different GPU connection interfaces. Customers basically would be locked in into whatever motherboard supports. And nvidia would never add support for AMD standard in their GPUs 
We can expect usual crap from nvidia. Titan 1st to milk the cows, and then if AMD shows up bring To and not let AMD card to embarass their titan.
Q1 is unrealistic, even q2 might be optimistic for smooth launch.
But we'll see
Associate
GTX 680 March 2012
GTX Titan February 2013
GTX 780 May 2013
GTX 780Ti November 2013
GTX 980 September 2014
GTX Titan X March 2015
GTX 980 Ti June 2015
There is a trend of NVIDIA making good money on the small version chips before releasing the big daddy. I don't see this changing with Pascal, but I think the best we can hope for is a 9 month wait for the mainstream big chip with Maxwell unlike the year+ wait with Kepler. There is much money to be made on the small chip cards imo.
Last edited:
It is very unrealistic to hope for nvlink or AMD fabric to come to desktops as pcie replacement. For one motherboard manufacturer will not cater 2 different GPU connection interfaces. Customers basically would be locked in into whatever motherboard supports. And nvidia would never add support for AMD standard in their GPUs
+1
Also the new process will hold them back, they won't be able to make a 600mm2 monster and clock speeds will probably suffer too. Not to mention all the other caveats, compromises and workarounds they will have working off a node tuned for mobile SOCs.
If you are expecting a 550mm2+ monster in Q1 you will be disappointed. Even in Q3.
It will be a tiny 1080 at first, GP104, likely not much bigger than 200mm2, with no uplift from 980ti just like 780ti -> 980 was. It will however use less power and NV will lie about the TDP again and release some stupid demo to brainwash all the casuals into thinking they've cured cancer or somesuch rubbish. Joe Rogan will talk about it on his podcast as if it's some amazing invention that will change the world. Then we will have another round of so-called enthusiasts proclaiming "low power is the future, I will never use a card above [insert TDP here] anymore, Nvidia rulzzzz". Blech. Of course they will forget all this when GP100 cut arrives a year or so later.
...Sorry I'm rambling, what where we talking about...
Last edited:
Agree with this - just look at the difficulties that Intel (the world leader in process technology) had. Die shrinks are not the easy ticket to lower power consumption and higher performance that they used to be, we have to reel in our anticipations.
Also remember that AMD have priority for HBM2. When you consider HBM yields, it may mean that NVIDIA cannot release their card for quite some time.
I personally hope AMD are able to be first to market with a HBM2 monster that will regain them significant market share, they really, really need a win.
Caporegime
So you will be getting one then?