Discussion: Nvidia Pascal

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
I have the Titan X and I have friends with the 980 ti... We have done testing with identical clocks...

We don't know the 3D GURU clocks however i'm assuming they are both reference cards... I doubt the clocks are the same because not all cards are created equally and clock speeds fluctuate unless you manually tune the card...
 
The rumor from benchlife menton that GP104 could end up with GDDR5X or GDDR5. buy most likely GDDR5 because they said mass production of GDDR5X won't start until Q2 or Q3 this year. but personally i don't think GDDR5 will held back the card that much. GDDR5X should provide bandwidth close to what current HBM1 cab provide. The target was to hit 16Gbps (double than current 8Ghz memory used in high end gpu) over the time but if i remember it right some early samples of GDDR5X could already hit 14Gbps. The main advantage was suppose to be much faster transition from GDDR5 to GDDR5X bacause GDDR5X retain GDDR5 compatibility (hence they were called GDDR5X not GDDR6).
 
The top gun definitely going to use HBM. but then again i heard HBM2 probably won't be ready towards end of this year. Will nvidia willing to wait that long? Those Xeon Phi is real threat to nvidia. For sure they can't afford 2017 for GP100. For nvidia GP100 must come out in 2016.
 
Well it certainly won't be named GeForce 1080, such a name would be pretty bad for a card gunning for 4k...

Also the important question is whether Pascal will be fixing all the DX12 issues Nvidia has, I have a feeling AMD may finally end up with a superior product this summer if this goes on - no HBM and (most likely) no GDDR5X is a bummer and GCN is already well equipped for DX12, so Polaris will most likely be strong there at a get go.
 
A few things to note.

1. Pascal will certainly include desktop cards as this was talked about during the conference NVIDIA held in Japan in February.

2. Pascal cards with HBM2 will be launching in May as multiple Universities and R&D companies have stated they will be taking stock of Quadro and Tesla cards based on Pascal.

3. Mid and low range desktop cards will likely be using GDDR5X instead of HBM2. This will include the "1080", the "1070", and most likely one or two lower end cards.

4. HBM2, produced for NVIDIA by Samsung, will be on the Quadro, Tesla, and Titan cards. The Titan replacement should (if NVIDIA sticks to the same release schedule as the 600, 700, and 900 series cards) be coming sometime around September.

5. The Ti cards won't be out until February of next year at the earliest, as NVIDIA will likely follow the same 9-12 month after release plan that they used for the 700 and 900 series cards.

6. Given that Micron have said that GDDR5X won't begin mass production until the end of April at the earliest, I wouldn't expect to see the "1080" become available until mid June.
 
Personally i'm not really excited with DX12. and seeing how some dx12 games perform right now some of my prediction probably already come true. Amd for their part as long as the game support DX12 they most likely not going to bother with DX11 optimization. Though their true strengh for DX12 right now lie on async compute. But still they also lack that hardware needed for other feature of DX. so when game start using that feature i don't know how that will going to affect current crop of radeon.
 


Nvidia already fixed the issue's look at the tomshardware & 3d guru benchmarks below...


http://www.tomshardware.com/reviews/ashes-of-the-singularity-beta-async-compute-multi-adapter-power-consumption,4479-2.ht

https://www.guru3d.com/articles_pages/ashes_of_singularity_directx_12_benchmark_ii_review,6.html

980 ti average fps is about the same as the fury x and the minimum fps is higher on the 980 ti.

I've never tested this game however based on recent online benchmarks it seems like the performance is similar...

Interesting to see h Rise of tombraider, and the new hitman will benefit from dx 12

 
Time to set aside the notion that AMD is better at DirectX 12. The two available benchmarks are long-standing AMD titles, one with its roots as an AMD tech demo and the other straight out of the AMD Gaming Evolved system. Very predictably, they show AMD doing better than Nvidia. Also predictably, they promote the one and only aspect of DirectX 12 where AMD has an advantage. These titles completely ignore those aspects of DirectX 12 where Nvidia holds the advantage, which include feature levels for which AMD cards are completely incapable. At this point, with only two AMD demos, the calls for AMDs superiority strongly echo those we heard when Mantle was released.

For the future, the advantage will be with whichever architecture scores the most wins in terms of game development relationships. In today's competitive landscape, it is a guarantee that any upcoming Nvidia sponsored titles will lean more heavily on raster processes and not so much on async compute. It is also a guarantee that the majority of games to be released in 2016 will favor Nvidia architecture, the architecture that 80% of gamers have in their rigs. This is already a reality to this point of the year.

Now the only caveat, and what I mean by "in this competitive landscape" is that there is also a very high likelihood that Nvidia will incorporate async compute capabilities into Pascal, and at a high level at that. At that point, then it's safe to assume that Nvidia-favoring game developers would follow suit and exploit those capabilities. Bottom line, Nvidia is best positioned to have games developed that feature the capabilities of its architecture, simply due to its dominant 80% discreet GPU market share.
 


some of this people they just look at it at one aspect and readily announced the winner. we are very well aware than current GCN lack the hardware needed for DX11.3 (referred as FL 12_1 in DX12). and yet some people insisting that AMD will dominate anything that is DX12. some even go as far DX12 = async compute when the feature itself only part of DX12 and not a mandatory even to use DX12. you know what the most funny thing is? when nvidia do better than AMD in GOW ultimate edition (even with updated driver that increase radeon performance up to 60%) they say that does not count because it is not 'true' DX12 title. and looking at hitman benchmark it seems nobody really care why 280X performance did not increased with DX12 at all while 380 have it's performance decreased in DX12. if the same pattern shown on nvidia card.....well they quickly become 'purposely gimped to sell new hardware'.
 


I was considering returning my Gsync monitor and waiting for the next generation of cards to come out, but last night I looked through AMD's annual financial report (out of curiosity, because I wasn't even thinking of it being a factor for the return). It's not even just Nvidia market share, it's also how poorly AMD has been financially for the past few years.

2011 - Net income of $491 million
2012 - Net loss of $1.2 billion
2013 - Net loss of $83 million
2014 - Net loss of $403 million
2015 - Net loss of $660 million

They've also had a decline in revenue the past few years. I read a report that it's very possible for AMD to file bankruptcy in 2019 because they have a lot of debt due in that year. If you want to read a dark and gloomy outlook of their company, read the Risk Factor section in AMD's annual report.

It's hard for me, as a consumer, to put money into an AMD card (I would imagine a business feels the same). AMD could very well have a great card this year, but it's not going to be easy to get out of the rut they're in.

This might not be news for some people, but I really had no idea they were doing this poorly. Sadly, if AMD does go under at some point, there won't be competition for Intel or Nvidia (not that AMD has been much to worry about CPU/APU wise in the past few years anyway).

This post isn't meant to bash AMD in anyway. I was just so shocked last night reading all this.

 
Hence the move recently establishing RTG (Radeon Technology Group). If the worst happenning in the future they probably will move as separate entity than AMD. Other conpany might as well going to buy them. Good for the graphic portion of AMD but i don't know about their x86. That's why some people mention that Zen are very important for AMD. If Zen end up as a flop then the company might as well going under. If AMD and RTG going their own way then that will be instant kill to APU.
 


What happens if I own an AMD card and they go under. Will there still be driver support?
 


I wouldn't worry too much yet about the all or nothing support. I was just thinking of quality of support as they struggle. I'm also not trying to scare anyone away from AMD in the future. I personally just would rather stick with the safer of the two, especially when I've had little to no issues with Nvidia cards in the past 5 years.

With the creation of the Radeon Technology Group (thanks renz496 for pointing that out), we should hopefully see how just the graphics division is doing in the next quarterly report. That should give us a better idea of what to expect in the future. In the annual report, GPU and CPU were the same entity, so it generally spoke about them together. How much revenue and how much cost were coming from each one wasn't clear.

Consolidations and company hierarchies aren't my strong points, so I can't be much help with how each business entity affects the other.

Edit: The reason I say not to worry about all or nothing support is:

1) The company can restructure so that "AMD" is basically a holding company for GPU and CPU, so if GPU is doing great and CPU is doing terrible, then CPU won't really affect the GPU entity.

2) AMD can sell their intellectual property for all their GPU related stuff and another company basically takes over (with or without support)

3) Bankruptcy doesn't necessarily mean they go out of business (depending on which chapter they file for). They can continue to operate while their debt gets restructured by a court approved committee.

4) Probably a ton of other reasons.
 
If they go under that's the worse to happen. When nvidia bought 3DFX they no longer support product that released by 3DFX. So i can understand people concern about this matter. If RTG remain as independent company they probably will still provide support for current crop of Radeon. But if they got bought by another company then it is up to their new 'boss' to support old product or not.
 
Actually we already see that since nvidia kepler and intel sandy bridge. And amd for their part did not want simply to provide competition so we can have our hardware cheaper. In fact they also want to join the premium game just like their compatitor are. Hence we got our 7970 at 550 (instead of 400) back in 2012 and Fury X at 650 (instead of 550) last year.
 


The 290x challenged the GTX 780, Original Titan & 780 ti despite the fact that it runs very hot and uses more energy...

Maxwell is the Architecture that AMD has had trouble competing with...
 
Status
Not open for further replies.