Polaris 11 could have 8.6 billion transistors.

[Imakuni]

Does it have CUDA? No?

 

Then I guess I'm still enslaved by Nvidia......

[Tedol]

I need it in not Windows Os.... I need it for Vulkan ! 

[That Norwegian Guy]

1 hour ago, Humbug said:

I'm betting that regardless of who wins the polaris vs pascal round they will be within 15% of each others general performance... Like they always are. Don't see anybody totally wiping the floor with it, based on recent GPU wars history...

I don't know, I think this time it might be different, as each manufacturer's priorities and preparedness for DX12 comes under light. Every new DirectX generation there's been "shocking" developments and this should be no different.

[That Norwegian Guy]

8 minutes ago, Imakuni said:

Does it have CUDA? No?

It does actually, they've had a CUDA license for 4 months now.

 

Nothing I can do about double-post btw, forum software doesn't let me copy a quote to an edit of a post.

[Imakuni]

Just now, That Norwegian Guy said:

It does actually, they've had a CUDA license for 4 months now.

 

Nothing I can do about double-post btw, forum software doesn't let me copy a quote to an edit of a post.

As far as I know, that's not a fully fledged CUDA license. It was only a partial license with a ton of restrictions; and if I remember it right, it was also only for server grade GPUs, not consumer ones.

[stconquest]

9 minutes ago, That Norwegian Guy said:

It does actually, they've had a CUDA license for 4 months now.

 

Nothing I can do about double-post btw, forum software doesn't let me copy a quote to an edit of a post.

I thought this as well, but it does not seem to be licensing the way you think it is.

 

I believe AMD has the license to develop a compiler for developers.  They (Devs) can use the compiler to program better OpenCL support for whatever program they are working on.

 

I could still be way off, but I just woke up... gimme a break =P

[patrickjp93]

1 hour ago, Prysin said:

They can, if they plan to pull another rebrand party next year or the one after. Simply pull another 300 series. Hue hue hue.

 

Or, they know samsungs 10nm is going to arrive fast enough that they can jump down yet another node very soon.

LOL! Yeah, 14nm came fast only because it uses 20nm BEOL. 10nm will use 14nm BEOL. 10nm is not coming soon from anyone except maybe the DRAM and NAND people were you don't have anywhere near the complexity of a GPU or CPU.

[dalekphalm]

1 hour ago, That Norwegian Guy said:

It does actually, they've had a CUDA license for 4 months now.

 

Nothing I can do about double-post btw, forum software doesn't let me copy a quote to an edit of a post.

Source?

 

As far as I'm aware, AMD does NOT have a CUDA license, and the chances of them making Polaris capable of running CUDA code (Eg: PhysX, GamesWorks effects based on CUDA, etc) is pretty much zero.

 

The only mention of a CUDA license was @patrickjp93 who was "inferring" that they had acquired one, due to the Compiler that AMD created (It takes the raw CUDA source code that a dev would have, and translates it to C++ if I recall correctly, thus allowing devs who wrote a CUDA program to quickly port it to OpenCL or whatever). There was never any actual evidence of that even meaning they had a CUDA license, let alone being allowed to execute actual compiled CUDA code.

 

If my information is incorrect or out of date, please correct me with a source.

1 hour ago, Imakuni said:

As far as I know, that's not a fully fledged CUDA license. It was only a partial license with a ton of restrictions; and if I remember it right, it was also only for server grade GPUs, not consumer ones.

As far as I'm aware, there was never any license at all. All the program does is take CUDA code, and translate it into C++, so that a dev (Who owns the rights to their own code/program) can easily port something from CUDA to C++. The original article's author speculated that it might mean AMD acquired a CUDA license, but it was pure speculation only.

1 hour ago, stconquest said:

I thought this as well, but it does not seem to be licensing the way you think it is.

 

I believe AMD has the license to develop a compiler for developers.  They (Devs) can use the compiler to program better OpenCL support for whatever program they are working on.

 

I could still be way off, but I just woke up... gimme a break =P

As far as I'm aware, no license is required for what AMD created. It's a program that takes uncompiled raw CUDA code and translates it to C++, so that it can be compiled for OpenCL. It doesn't translate it in real time or anything - the dev still needs to check the code, correct any mistakes or anything the translator missed, and then he still needs to compile the code into OpenCL.

[stconquest]

Just now, dalekphalm said:

Source?

 

As far as I'm aware, AMD does NOT have a CUDA license, and the chances of them making Polaris capable of running CUDA code (Eg: PhysX, GamesWorks effects based on CUDA, etc) is pretty much zero.

 

The only mention of a CUDA license was @patrickjp93 who was "inferring" that they had acquired one, due to the Compiler that AMD created (It takes the raw CUDA source code that a dev would have, and translates it to C++ if I recall correctly, thus allowing devs who wrote a CUDA program to quickly port it to OpenCL or whatever). There was never any actual evidence of that even meaning they had a CUDA license, let alone being allowed to execute actual compiled CUDA code.

 

If my information is incorrect or out of date, please correct me with a source.

As far as I'm aware, there was never any license at all. All the program does is take CUDA code, and translate it into C++, so that a dev (Who owns the rights to their own code/program) can easily port something from CUDA to C++. The original article's author speculated that it might mean AMD acquired a CUDA license, but it was pure speculation only.

As far as I'm aware, no license is required for what AMD created. It's a program that takes uncompiled raw CUDA code and translates it to C++, so that it can be compiled for OpenCL. It doesn't translate it in real time or anything - the dev still needs to check the code, correct any mistakes or anything the translator missed, and then he still needs to compile the code into OpenCL.

ty.

[Curufinwe_wins]

3 hours ago, Arokhantos said:

Hopefully they will including all gameworks titles, gameworks is so bad.

You know... Shit aditudes like this are the reason we get console grade graphics. 

 

People need to get off their high horses and look long term... Gameworks moves the graphics industry  forward at a pace that wouldn't be possible  without it (example,  amd wouldn't work nearly as hard on their variant if they didn't feel a need to catch up to nvidia.)  It is a thing that can at any time be turned off,  and one day down the line even if you don't have the horsepower for it yet,  you will be able to play those games with it and appreciate the additional graphics (just like crysis series today).  

 

 

[Doobeedoo]

I'm sure flagship should have significantly more transistors. Taking account new node and architectural improvements, quotes on how much efficient it can be, though making room for potential power rather than just efficiency will probably be made.

[-BirdiE-]

I'm torn.

 

I want AMD to do better than Nvidia... But I also just spent $1300 USD on a G-Sync monitor....

[dalekphalm]

43 minutes ago, -BirdiE- said:

I'm torn.

 

I want AMD to do better than Nvidia... But I also just spent $1300 USD on a G-Sync monitor....

Well you could sell it and buy an Adaptive-Sync monitor. Sooner or later NVIDIA will support them (Hell, NVIDIA could add driver support literally any time they want - Pascal will in all likelihood, be 100% hardware compatible with Adaptive-Sync monitors).

[DocSwag]

Just now, dalekphalm said:

Well you could sell it and buy an Adaptive-Sync monitor. Sooner or later NVIDIA will support them (Hell, NVIDIA could add driver support literally any time they want - Pascal will in all likelihood, be 100% hardware compatible with Adaptive-Sync monitors).

Problem is. WILL Nvidia support it? I think that the G-Sync chips are purchased from them (I could be wrong) so they wouldn't want to give business away for free.

[dalekphalm]

1 minute ago, DocSwag said:

Problem is. WILL Nvidia support it? I think that the G-Sync chips are purchased from them (I could be wrong) so they wouldn't want to give business away for free.

Yes, NVIDIA makes the chips (Or, more precisely, they pay a fab to make them, then they program them) and sells them directly to the monitor vendors.

 

But yes, I think sooner or later, NVIDIA will support Adaptive-Sync. Eventually, we'll get to true feature parity. Right now, while it's pretty damn close, G-Sync is still a more robust implementation. That gap will narrow, then disappear over time.

 

Once we get to a point where G-Sync and the industry standard (eg: Non-AMD) Adaptive-Sync are at feature parity, and perform the same, then (hopefully) we'll see enough pressure from consumers that NVIDIA will adopt the industry standard.

 

We all know that it's within their technical capabilities. All they need is a GPU with DP 1.2a or DP 1.3+. They can even keep the G-Sync branding on their drivers, just like AMD has the FreeSync branding for their Adaptive-Sync drivers.

 

Just think of the day where you can buy a $1500 monitor that is both FreeSync AND G-Sync certified! But NVIDIA wants to keep that day as far away as possible, because they LOVE vendor lock-in.

[-BirdiE-]

9 minutes ago, dalekphalm said:

Well you could sell it and buy an Adaptive-Sync monitor. Sooner or later NVIDIA will support them (Hell, NVIDIA could add driver support literally any time they want - Pascal will in all likelihood, be 100% hardware compatible with Adaptive-Sync monitors).

Yeah. I think DocSwag got it pretty right. I don't think Nvidia would do that if they're selling g-sync chips.. it would kill their sales since there would be no reason for any manufacturer to make a g-sync monitor.

 

And just to save you potential confusion in the future, the non-proprietary technology is called "Freesync" and both Freesync and G-sync are examples of "Adaptive Sync" technology.

[branden_lucero]

they need to pull a Skyline. deliver a sleeper hit power house that actually delivers more power than advertised. 

[corsairian]

8 hours ago, Starelementpoke said:

I dunno, I don´t play cod.

What game were you talking about?

[Curufinwe_wins]

5 hours ago, Arokhantos said:

How is using non universal standards good vs universal standards ?

Nvidia won't even allow early acces for amd to optimise game gameworks is purely nvidia thingy that works on ati to but aimed to bring down ati performance down, like tell me why would a brick need 10,000 polygons ? why do we need tesalated water underground not even visible, gameworks is just bad and you know it, if not you just ignorant.

Good job completely ignoring 100% of the point.

Nvidia making Gameworks and continuing to develop is literally the ONLY reason AMD pushes forward with their own standard.

 

Do you honestly believe if Nvidia didn't implement Gameworks that a studio would do actually equivalent features themselves? LOLOLOL.

 

It isn't non-universal standards vs universal standards, it was (prior to the latest iteration of GPUOPEN) have Nvidia do physx/gameworks for you or just don't bother implementing anything like it (havoc engine was the closest thing to Gameworks and it is a joke by comparison). Now (FINALLY) AMD is following in Nvidia's footsteps (GPUOPEN) and has a system that MIGHT be able to challenge Gameworks in a few months/years (it still is significantly worse from a graphics perspective).

 

BTW the whole Gameworks conspiracy theory is truly laughable these days with AMD cards shooting above par performance relative to their Nvidia (outside of Fiji which is pretty bad) in Gameworks title (and the Gameworks fiasco is one of the reasons AMD pushed the tesselation improvements to Hawaii and other non GCN1.2 200/300 series cards, so really you should thank Nvidia for giving AMD owners free preformance from here on out. Tesselation isn't going anywhere, and really it shouldn't. You have to make wire frames more complex at some point because textures have basically already caught up.)

[VagabondWraith]

Both companies help each other innovate. They push each other to "one-up" the other in tech. Without it, we'd see Intel's minuscule tick-tock 5%* performance improvement from generation to generation.

 

*Exaggeration, but still.

[Stadin6]

17 hours ago, dalekphalm said:

 port something from CUDA to C++. 

i think it was C not C++.

not sure tho

[DocSwag]

10 hours ago, Curufinwe_wins said:

Good job completely ignoring 100% of the point.

Nvidia making Gameworks and continuing to develop is literally the ONLY reason AMD pushes forward with their own standard.

 

Do you honestly believe if Nvidia didn't implement Gameworks that a studio would do actually equivalent features themselves? LOLOLOL.

 

It isn't non-universal standards vs universal standards, it was (prior to the latest iteration of GPUOPEN) have Nvidia do physx/gameworks for you or just don't bother implementing anything like it (havoc engine was the closest thing to Gameworks and it is a joke by comparison). Now (FINALLY) AMD is following in Nvidia's footsteps (GPUOPEN) and has a system that MIGHT be able to challenge Gameworks in a few months/years (it still is significantly worse from a graphics perspective).

 

BTW the whole Gameworks conspiracy theory is truly laughable these days with AMD cards shooting above par performance relative to their Nvidia (outside of Fiji which is pretty bad) in Gameworks title (and the Gameworks fiasco is one of the reasons AMD pushed the tesselation improvements to Hawaii and other non GCN1.2 200/300 series cards, so really you should thank Nvidia for giving AMD owners free preformance from here on out. Tesselation isn't going anywhere, and really it shouldn't. You have to make wire frames more complex at some point because textures have basically already caught up.)

I see your point. Gameworks was bad for GPU performance, but it pushed AMD to try harder as well as launch a GPUOpen initiative.

However, I believe even though overall it was probably good, Nvidia could have gone a different route.

For me, the consumer, I would much rather have Nvidia just trying to make better GPUs.

Instead, gameworks reduced performance for both Nvidia and AMD GPUs. AMD GPUs just got hit harder.

Now we are finally seeing its benefits. However, if instead of doing this Nvidia just pushed harder with their GPUs we might have got the same results without all this gameworks crap mixed in.

However, I do admit that GPUOpen is probably one of the good things that came from Gameworks that probably would not have otherwise.

[dalekphalm]

19 hours ago, -BirdiE- said:

Yeah. I think DocSwag got it pretty right. I don't think Nvidia would do that if they're selling g-sync chips.. it would kill their sales since there would be no reason for any manufacturer to make a g-sync monitor.

 

And just to save you potential confusion in the future, the non-proprietary technology is called "Freesync" and both Freesync and G-sync are examples of "Adaptive Sync" technology.

You're mistaken. FreeSync is a proprietary, entirely AMD owned Driver-Side software package. The FreeSync branding is used quite liberally by the monitor industry (and AMD), since currently they're the only party to officially support any products that work with Adaptive-Sync.

 

Adaptive-Sync is the DisplayPort industry standard.

See:

http://www.vesa.org/news/vesa-adds-adaptive-sync-to-popular-displayport-video-standard/

 

Adaptive-Sync is the hardware side implementation, as part of the VESA DisplayPort standard (It's an optional feature still though - so it's up to the monitor manufacturers to implement it still).

 

G-Sync is NOT Adaptive-Sync, nor does it use the same technology. G-Sync is an example of Adaptive-VSync (Note the V), which is a generic term used to describe various technologies that adapt frame-rate and refresh rate on demand. Adaptive-Sync is a very specific, VESA standard.

 

G-Sync works in a very similar fashion to Adaptive-Sync, but there are fundamental differences, including the amount of two-way communication that happens between the G-Sync module and the GPU. The entire reason a G-Sync module was required, is because at the time, Adaptive-Sync didn't even exist, so there was no Monitor hardware-side way for NVIDIA to adapt the refresh rate as needed. They filled this gap with a proprietary hardware module -> the G-Sync Module.

6 hours ago, Stadin6 said:

i think it was C not C++.

not sure tho

Could be - I'm not 100% sure whether it's C or C++ - at this point though, the difference is trivial, as my point was still well understood

[-BirdiE-]

5 minutes ago, dalekphalm said:

-snip-

I stand corrected.

[Coaxialgamer]

Wait. 

That cant be right,  since the fury x has 8.9 billion.  That would be a downgrade.  He must be refering to polaris 10.