Support NeoGAF

[Locuza]

Were GloFlo and Samsung not collaborating in terms of 14nm design and fabrication in order to speed up TTM so that they would not be left behind by Intel and TSMC? I recall reading as much last year some time

The story is a little bit different.
GloFo development their own 14xm process, which would be too late or/and too bad for the market, in which case GloFo licensed the 14nm process from Samsung to offer something solid in the upcoming timeframe.

The process between Samsung and GloFo should be very similiar but for the fabrication I would say that only GloFo will directly produce chips for AMD.

 

[jeff_rigby]

The story is a little bit different.
GloFo development their own 14xm process, which would be too late or/and too bad for the market, in which case GloFo licensed the 14nm process from Samsung to offer something solid in the upcoming timeframe.

The process between Samsung and GloFo should be very similiar but for the fabrication I would say that only GloFo will directly produce chips for AMD.

This could be possible if articles were confusing Samsung making the HBM2 for AMD with second sourcing. Or AMD is also second sourcing HBM2.

not only will Samsung manufacture AMD's next-gen GPUs, but also its upcoming "Zen" family of CPUs, at least a portion of it. AMD is looking to distribute manufacturing loads between two foundries, Samsung and GlobalFoundries, perhaps to ensure that foundry-level teething trouble doesn't throw its product launch cycle off the rails.

Samsung to Start Mass Production of HBM2 in 2016

What is new application 2018?

 

[Locuza]

Or licensed process from Samsung = Samsung produces chips for AMD.

I believe it's very unlikely that Samsung will produce chips for AMD.
I doubt they have the capacity.
Samsung needs to produce an awful lot of their own chips for mobile and also using chips from Qualcomm, where I don't know if the new Snapdragon is also produced by Samsung or also by TSMC.
Either way it shows capasity shortage on Samsungs side.
And with GloFo AMD has a multi-year contract, so it would be wise to have nearly everything on GloFo.

PS: I don't have a clue what the new applications might be, maybe they don't know themself, hence the question-mark. ^^

 

[Renekton]

The only official thing is, that AMD confirmed to have tape-out some designs at Global Foundries.
No more, no less.

This whole Samsung/TSMC/GloFo thingy, who is what producing is simply speculation at this point.
I personally wouldn't hold my breath for Samsung at all.

If it's just GF alone, AMD is toast.

GF was at least partly responsible for the Bulldozer mess. Always late to the party, they also made noise years ago about 20nm and other process tech which AMD banked future designs on but never delivered.

 

[Locuza]

Like in the past it should be two sources.
TSMC + GloFo.
Only now AMD may have more GPUs fabricated on GloFo's process.

If GloFo's process is bad for high performance chips, AMD maybe even fabricate Zen at TSMC.

This would rotate the picture we had in the past and we really don't know what AMD will do.
They have the proces parameters, they know what their plans and potential volume is, they will choose the way it makes sense.

 

[General Lee]

https://www.youtube.com/watch?time_continue=19&v=5g3eQejGJ_A

Looks like 14 nm is confirmed for GPUs. AMD seems to be very excited about their upcoming products. Gives hope that it's not going to disappoint.

 

[EndcatOmega]

https://www.youtube.com/watch?time_continue=19&v=5g3eQejGJ_A

Looks like 14 nm is confirmed for GPUs. AMD seems to be very excited about their upcoming products. Gives hope that it's not going to disappoint.

As much as I hope it's good AMD wasn't going to come along and go, 'yeah it's going to be a bit shit, really'.

 

[dr_rus]

AMD Reveals Polaris GPU Architecture: 4th Gen GCN to Arrive In Mid-2016

So Polaris isn't really a substitute for GCN, it's a "macro architecture" name which include GCN 1.3 / GCN 4 in it. This makes more sense.

Another couple of interesting tidbits from Smith:

- As I've suspected the most of 2.5x perf/watt increase is coming from the new process but there seems to be enough architectural changes to bring Polaris at least to Maxwell's power efficiency level; we'll have to wait for real products to see how this will pan out; it obviously makes little sense to compare an early Polaris chip to a production GM206 videocard as Polaris won't compete with that.

- Kinda expected but still - the first Polaris videocard to the market will be a low end one with GDDR5. I fully expect the same from Pascal although I think that NV will try to launch GP104 middle end alongside with their lower end GP106/GP108 offerings like they did with Kepler family.

2016 looks to be really exciting. With any luck we'll see Radeons back in full force and NV on the defensive which is always the best NV there is.

 

[Irobot82]

TechReport has their article up. Great read!

 

[Locuza]

Looks like 14 nm is confirmed for GPUs.

AMD confirmed two new GPUs for 2016.
The small Polaris will be manufactured by Global Foundries 14nm according to AMDs Raja Koduri.
Questions about the process for the bigger one, were not answered.
Also if more GPUs than two are to expect.

But they confirmed, that they will use both, GloFo and TSMC for production.
So TSMC is at least for one chip confirmed.
Either for the second confirmed chip or maybe for a third one, which might be announced later.

 

[tuxfool]

AMD will be on a smaller node compared to Nvidia, at least for the flagship cards (Samsung 14nm vs TSMC 16nm). It's going to be interesting to see how these new cards perform.

The node size is a marketing term. Ignore it.

What other physical characteristics of each type of process matter far more. For example Intel's 14nm is actually more dense than Samsung's, simply because the metal layers on Samsung are still at 20nm feature sizes.

 

[Renekton]

Like in the past it should be two sources.
TSMC + GloFo.
Only now AMD may have more GPUs fabricated on GloFo's process.

If GloFo's process is bad for high performance chips, AMD maybe even fabricate Zen at TSMC.

This would rotate the picture we had in the past and we really don't know what AMD will do.
They have the proces parameters, they know what their plans and potential volume is, they will choose the way it makes sense.

Hmm Anandtech said GF will manufacture the lower-end ones like the mystery 120mm² GPU shown.

(nvm beaten)

 

[Locuza]

PCGH said, that they were given a look at the card and could make a rough sketch, which would indicate something around ~150mm².

Übrigens: Da wir den Chip selbst in Händen hielten, können wir anhand einer rudimentär angefertigten Skizze eine Größe von etwa 12x12,5mm, also rund 150 mm² grob schätzen - rein optisch wird sicherlich der "Daumennagel"-Vergleich des Öfteren bemüht werden.

http://www.pcgameshardware.de/AMD-Polaris-Hardware-261587/Specials/AMD-Polaris-Daten-Release-1182286/

 

[aeolist]

PCGH said, that they were given a look at the card and could make a rough sketch, which would indicate something around ~150mm².

http://www.pcgameshardware.de/AMD-Polaris-Hardware-261587/Specials/AMD-Polaris-Daten-Release-1182286/

anandtech said their eyeball guess was >120mm^2. 150 would be huge for a lower-end GPU on a new node.

 

[Rushersauce]

I'm already saving for my new GPU. AMD better deliver or I'll go back to Nvidia

 

[hesido]

you do realize this is comparing amds new architecture against maxwell right? while i expect amds architecture to remain better suited to modern workloads compared to pascal, this comparison isnt very meaningful

Oh, I know, and you're right about the meaningfulness, although the performance per watt increase is substantial and somewhat within the ballpark of expected NVidia gains, and couple this with the fact that AMD is the only solution who can package an X86 inside a single chip which should provide better energy efficiency compared to a separate CPU-GPU combo, they are poised to bring to market the most efficient CPU+GPU chip, which will make them an easy choice for the console hardware vendors.

 

[Meh3D]

How do you know that and what exactly do you mean bei GPU compute development boards?


If Polaris changes a lot, it's nearly impossible to have zero impact on performance.


Small correction here, they both use Sea Islands (GCN 1.1/GCN Gen 2) GPUs.

Hello there Locuza. Going to try to keep it brief.


Most recently, I worked on an unpublished paper on procedural generation of space bodies. Last semester I attempted the project on my CUDA development board. However, it was Kepler based while the host system running the project was Maxwell based. Furthermore, the referenced works were Fermi based. (Just issues all around. Lack of source code, completely different architectures etc..)

Over the break I continued to work on a side project related to the paper except this time on scientifically accurate space bodies and their collision versus the code I wrote for Unreal Game Engine. The R9-390 was “heavenly.” Specifically, the amount of host memory and double precision. The new CUDA development board is expensive and even more limited than the first. Tesla boards are too expensive for what I get in return versus the Hawaii boards.

If it was just double precision, the 7970 is an f’ing champion. If it was just great compute performance in OpenCL, the Maxwell boards are awesome. However, for me, the R9-390 I got for 300 USD has just been stellar for both task I needed it for in coding.

PS. Damn Khronos and OpenCL for their cluster fuck of an API and their documentation. And Intel for their pitiful 1.5GB maximum in their Iris 6100 GPU. Kudos for AMD and Apple for their work and documentation.

 

[dr_rus]

PS. Damn Khronos and OpenCL for their cluster fuck of an API and their documentation. And Intel for their pitiful 1.5GB maximum in their Iris 6100 GPU. Kudos for AMD and Apple for their work and documentation.

Aren't these the same guys? =)

 

[Meh3D]

Aren't these the same guys? =)

Yes and no. They both are working on their own compute APIs now (Metal and Boltzmann Initiative) and most of the criticism about OpenCL from these two have been spot on. I'm quite happy about this. I original wrote "Fuck" but decided against. Don't want it pulled up in an interview in the future. XD

 

[Locuza]

@ MarsMartin

Your statement was that with the way Fury went Hawaii was the last inexpensive development board without compromises.
But to be sure, you need to know, what AMD will do in the future.
Fiji itself is "problematic" for every market segment.

The Hardware DP-Rate is 1/16 and thanks to HBM the card is limited to 4GB, which is even for Gamers not optimal.

Coming next is Polaris and without insider information, we won't know how AMD wants to handle DP-Rate and Memory-Size for consumer and professionals.

 

[Meh3D]

@ MarsMartin

Your statement was that with the way Fury went Hawaii was the last inexpensive development board without compromises.
But to be sure, you need to know, what AMD will do in the future.
Fiji itself is "problematic" for every market segment.

The Hardware DP-Rate is 1/16 and thanks to HBM the card is limited to 4GB, which is even for Gamers not optimal.

Coming next is Polaris and without insider information, we won't know how AMD wants to handle DP-Rate and Memory-Size for consumer and professionals.

Oh that statement. Nothing against Fury. The way the architecture went makes sense because of its primary market. NVidia did the same thing with Fermi -> Kepler/Maxwell. In fact, you can still buy brand new Fermi based Tesla accelerators. There is a still a demand for these accelerators (upgrading existing clusters) and you can buy 470 GTX's with the exact GPU to play with.

That statement was basically reflecting that Hawaii/Granada are the last breed of GPU's in this regard. The last Fermi's if you will.

 

[Locuza]

My question was and still is, why are they or what makes you so certain?

 

[Bankastuba]

Tremble, Nvidia.

 

[dr_rus]

Yes and no. They both are working on their own compute APIs now (Metal and Boltzmann Initiative) and most of the criticism about OpenCL from these two have been spot on. I'm quite happy about this. I original wrote "Fuck" but decided against. Don't want it pulled up in an interview in the future. XD

So I take the "yes and no" as a yes then? 8) OpenCL was always a joint Apple+AMD initiative so it's really funny to see how you praise their efforts with Metal and Boltzmann Initiative (not sure that you can call a compiler and a CUDA translation utility an "API" btw) while calling out Khronos on OpenCL support.

Oh that statement. Nothing against Fury. The way the architecture went makes sense because of its primary market. NVidia did the same thing with Fermi -> Kepler/Maxwell. In fact, you can still buy brand new Fermi based Tesla accelerators. There is a still a demand for these accelerators (upgrading existing clusters) and you can buy 470 GTX's with the exact GPU to play with.

What's Fermi have which you can't get in Kepler? Fast DPs in consumer level h/w? Anything else?

 

[wachie]

Tremble, Nvidia.

This is funny when Koduri is making statements like this.

Next year (2017) we need to target 20 percent better at this cost and this power.

http://videocardz.com/58157/amd-has-two-polaris-finfet-gpus-in-development

Nvidia will gobble up the market and AMD's market share will be 5% or less if the above comes to pass.

 

[Nostremitus]

Good. Hurry up and give me something I won't regret buying. I really want to support AMD.

I have no regrets for my 290x, it's been a trooper, it started out with decent price/perf(minus the bitcoin fiasco) and it has done nothing but get better with each driver update.

 

[Thraktor]

This is funny when Koduri is making statements like this.

http://videocardz.com/58157/amd-has-two-polaris-finfet-gpus-in-development

Nvidia will gobble up the market and AMD's market share will be 5% or less if the above comes to pass.

That's taken ludicrously out of context, he's saying that usually graphics cards target 20% more performance, but that this time they "set a completely different goal". Here's what he actually said:

When we set to design this GPU, we set a completely different goal than for the usual way the PC road maps go. Those are driven by "The benchmark score this year is X, next year we need to target 20 percent better at this cost and this power." We decided to do something exciting with this GPU. Let’s spike it so we can accomplish something we hadn’t accomplished before.

Here's the link to the original interview. (I've added quotation marks to Venturebeat's originally formatting, as they should have been included by whoever was transcribing the interview in the first place).

Not that any of these PR statements mean anything, anyway, but if you're going to try to base an argument on them at least quote them properly.

 

[dr_rus]

This is funny when Koduri is making statements like this.

http://videocardz.com/58157/amd-has-two-polaris-finfet-gpus-in-development

Nvidia will gobble up the market and AMD's market share will be 5% or less if the above comes to pass.

Not if NV will have something that will be 20% better as well.

 

[Blanquito]

Here's the link to the original interview.

Thanks for the link, that was a great read.

There are a couple of quotes that I found interesting in that article... I'll quote them here.

We can do HAVC encode and decode at 4K on this chip. It’ll be great for game streaming at high resolution, which gamers absolutely love. It takes no cycles away from games. You can record gameplay and still have an awesome frame rate. It’ll be available in mid-2016.

The writer probably meant to write HVEC. In any case, it sounds like these GPUs (and next gen consoles) will be able to stream at up to 4k if they wanted to (and their internet could handle it). It seems AMD really liked the idea of a streaming chip built into the APU (or GPU in this case) that doesn't take cycles away from the GPU/games.

What they’ve been able to achieve on consoles in the current generation, versus the current high-end PC — the current high-end PC specs are at least four to eight times faster than current consoles. The Fury X is an eight teraflop machine. The PS4 is a two teraflop machine. It’s four times more compute in that single Fury. You can build a dual Fury PC. But PC doesn’t give you that much better an experience with cutting edge content, because they can extract more performance from a console. They’re also investing a lot of IP into that architecture. They’re doing some really clever things that are not possible on the PC yet.

I wonder what things he's talking about here, in respect to things that PS4 can do that are not possible on PC yet? Is it something we already kind of know about (e.g. good async compute due to the direct links between CPU & GPU), or is it something different?

Good stuff.

 

[Kezen]

I wonder what things he's talking about here, in respect to things that PS4 can do that are not possible on PC yet? Is it something we already kind of know about (e.g. good async compute due to the direct links between CPU & GPU), or is it something different?

I'll believe it when I see it.
We have been told that hundreds of time without anything to back it up.

 

[Blanquito]

I'll believe it when I see it.
We have been told that hundreds of time without anything to back it up.

It's not like it's some random forum-poster that's saying it... this is the head of AMD's GPU division. You can't just hand-wave away someone at that level saying something like that. (Especially since the interview wasn't focused on the console business, it was focused on the PC GPU business and he was talking about how he wants to improve their discreet GPUs)

 

[Kezen]

It's not like it's some random forum-poster that's saying it... this is the head of AMD's GPU division. You can't just hand-wave away someone at that level saying something like that.

Authority arguments are not receivable.
Show me concrete evidence of a tech, a game not possible on PC with DX12.

"Clever things" is vague as hell, you can't expect me to take that at face value regardless of who is making that claim. Is he talking about specific optimization or rendering techniques ?

 

[Blanquito]

Authority arguments are not receivable.
Show me concrete evidence of a tech, a game not possible on PC with DX12.

Haha, calm yourself. I'm not trying to make an argument, I was merely wondering what it could possibly be. Ya know, trying to discuss things on a discussion board.

 

[Kezen]

Haha, calm yourself. I'm not trying to make an argument, I was merely wondering what it could possibly be. Ya know, trying to discuss things on a discussion board.

Shame the interviewer did not press him. But it's not like we have not been there, console magic etc...

 

[Blanquito]

Shame the interviewer did not press him.

Agreed.

 

[Kezen]

Agreed.

I finally found something ! HRAA, used on some Ubi games which apparently is too low level for PC (DX11).
Results are mediocre but I guess that counts for something.

But I was talking (explicitly) about DX12.

Oh I guess we will find out eventually, I mean it's not like multiplatform games don't exist, right ?
So we can look forward to things consoles skus will benefit from that will not make the cut on PC DX12.

Oh boy I can't wait.

 

[dr_rus]

Haha, calm yourself. I'm not trying to make an argument, I was merely wondering what it could possibly be. Ya know, trying to discuss things on a discussion board.

It can be and probably is absolutely nothing. There is nothing that you can do on PS4 but can't do on PC. Some stuff just runs better on PS4 because of UMA.

 

[Kezen]

It can be and probably is absolutely nothing. There is nothing that you can do on PS4 but can't do on PC. Some stuff just runs better on PS4 because of UMA.

Does specific optimization techniques count ? I mean I can see that being the case considering the architecture is different from a PC but I read his quote as much, much more than that. If the visual integrity can be maintained on PC even with a different optimization technique then it's nowhere near impressive.

I read his quote as "visual features" that could only work on consoles, that's my interpretation.

 

[Blanquito]

It can be and probably is absolutely nothing. There is nothing that you can do on PS4 but can't do on PC. Some stuff just runs better on PS4 because of UMA.

If there is nothing that you can do on PS4 that you can't do on PC, then why would the head of AMD's GPU division (in an interview about improving their discreet GPU line) say that? I mean, I would understand it coming from a Sony/MS spokesperson, or something similar. But not from a guy trying to sell you your next PC GPU. Which is why I posted the quote.

But, as was mentioned, we don't really have any more information than that, so there isn't much to talk about at the moment, sadly.

I finally found something ! HRAA, used on some Ubi games which apparently is too low level for PC (DX11).
Results are mediocre but I guess that counts for something.

But I was talking (explicitly) about DX12.

Oh I guess we will find out eventually, I mean it's not like multiplatform games don't exist, right ?
So we can look forward to things consoles skus will benefit from that will not make the cut on PC DX12.

Oh boy I can't wait.

Haha, interesting. Do you have a link where you read about this?

 

[Locuza]

So we can look forward to things consoles skus will benefit from that will not make the cut on PC DX12.

Like you said, HRAA for example, because the DX API doesn't offer programmable sample-points.
This is an API-Limitation, OGL does offer it and probably Vulkan also.
Vulkan in general could be do some things, that DX12 doesn't right now and there is an extension mechanic.

DX12 right now is lacking cross lane operations, which someone tweeted, improved some algorithm with the factor 2 on consoles.

For some more info:
http://advances.realtimerendering.com/s2015/aaltonenhaar_siggraph2015_combined_final_footer_220dpi.pdf

Page 57.

 

[jeff_rigby]

Does specific optimization techniques count ? I mean I can see that being the case considering the architecture is different from a PC but I read his quote as much, much more than that. If the visual integrity can be maintained on PC even with a different optimization technique then it's nowhere near impressive.

I read his quote as "visual features" that could only work on consoles, that's my interpretation.

I think it's more a matter of implementation.

The XB1 will be able to game stream using HEVC now while no PC until Polaris dGPUs has a HEVC encode that doesn't use GPU cycles so no PC game HEVC streaming for another year or so....

XB1 game streaming to PC using HEVC
No PC game streaming to XB1 using HEVC

Kaveri may or may not be able to decode HEVC without using the GPU but it still has a VCE h.264 encoder. Polaris apparently has both HEVC encode and decode using Xtensa DPU(s) which is what the XB1 uses.

 

[pottuvoi]

Haha, interesting. Do you have a link where you read about this?

Here is a presentation about it.

As you can see it's not a single method, but a multiple solutions for different cases of aliasing.

The analytical AA in it is really fast on GCN due the ability to use interpolator data in pixel shaders. (Sadly is not exposed in OGL/DX)
Here fun things about GCN in presentation, video version. (has interpolator things etc.)

 

[Kezen]

Haha, interesting. Do you have a link where you read about this?

Nope, I have not found any specifics but that does not mean it can't be found.
There are things like programmable MSAA blending which is not exposed by DX12 yet, but I have no idea about the graphical reach of that feature.

I guess my bewilderment stemed from my own interpretation of his words. Because technically speaking the aforementioned features are indeed exclusive to consoles but what I had in mind when reading the quote was something along the lines of a brand new, ground-breaking visual feature which DX12 could not make possible.

Like you said, HRAA for example, because the DX API doesn't offer programmable sample-points.
This is an API-Limitation, OGL does offer it and probably Vulkan also.
Vulkan in general could be do some things, that DX12 doesn't right now and there is an extension mechanic.

DX12 right now is lacking cross lane operations, which someone tweeted, improved some algorithm with the factor 2 on consoles.

For some more info:
http://advances.realtimerendering.com/s2015/aaltonenhaar_siggraph2015_combined_final_footer_220dpi.pdf

Page 57.

Interesting. What are the genuine, total practical gains in games ?

 

[Blanquito]

Like you said, HRAA for example, because the DX API doesn't offer programmable sample-points.
This is an API-Limitation, OGL does offer it and probably Vulkan also.
Vulkan in general could be do some things, that DX12 doesn't right now and there is an extension mechanic.

DX12 right now is lacking cross lane operations, which someone tweeted, improved some algorithm with the factor 2 on consoles.

For some more info:
http://advances.realtimerendering.com/s2015/aaltonenhaar_siggraph2015_combined_final_footer_220dpi.pdf

Page 57.

Thanks, nice slides.

I guess my bewilderment stemed from my own interpretation of his words. Because technically speaking the aforementioned features are indeed exclusive to consoles but what I had in mind when reading the quote was something along the lines of a brand new, ground-breaking visual feature which DX12 could not make possible.

Oh yeah, I'm not delusional, I wasn't trying to imply some brand new visual feature exclusive to consoles. I was thinking along the lines of some small performance enhancement or feature that helps eek out a bit more power here and there (like their async compute, for example). Glad we got that cleared up.

 

[Kezen]

Oh yeah, I'm not delusional, I wasn't trying to imply some brand new visual feature exclusive to consoles. I was thinking along the lines of some small performance enhancement or feature that helps eek out a bit more power here and there (like their async compute, for example). Glad we got that cleared up.

Async compute is exposed by DX12, but you are right that there may be more to squeeze on consoles than on a typical GCN GPU on PC.

The problem as always is that we have very little to base our speculation on, very few developpers have bothered shedding light on the matter and it is not quite clear to which extent DX12 and Vulkan will narrow the gap in efficiency between gaming PCs and consoles.

I apologize if I came across as abrasive, that really wasn't my intention.

Rise of the Tomb Raider (ex xbone exclusive) will be interesting to benchmark, it uses async compute on consoles so I wonder what hardware on the PC side will achieve the same quality at the same framerate.
Thus far the R7 260 did the job, but obviously the 1040 clock speed helped vs the 853 on the bone.

 

[Locuza]

The XB1 will be able to game stream using HEVC now while no PC until Polaris dGPUs has a HEVC encode that doesn't use GPU cycles so no PC game HEVC streaming for another year or so....

XB1 game streaming to PC using HEVC
No PC game streaming to XB1 using HEVC

Kaveri may or may not be able to decode HEVC without using the GPU but it still has a VCE h.264 encoder. Polaris apparently has both HEVC encode and decode using Xtensa DPU(s) which is what the XB1 uses.

Maxwell v2 supports in general HEVC(h.265) encode.
But only the GM206 chip also offers h.265 decode.
Normally it's often the other way around, but I would guess Nvidia wanted something for shadowplay.

Carrizo and Fiji support both HEVC decode (UVD), but about the encode side (VCE) I'm not sure.
I would believe no?

Intels Skylake does a lot, HEVC decode/encode and for the 10-Bit profile they use the hybrid approach with GPU utilization.

Interesting. What are the genuine, total practical gains in games ?

No idea.

Like for the MSAA-Samples, there is also an OGL extension to support cross lane operations and Timothy Lottes did one blog-post about the topic:
http://timothylottes.blogspot.de/2015/11/cross-invocation-data-sharing.html

 

[Kezen]

Like for the MSAA-Samples, there is also an OGL extension to support cross lane operations and Timothy Lottes did one blog-post about the topic:
http://timothylottes.blogspot.de/2015/11/cross-invocation-data-sharing.html

I guess it was too much to ask for DX12 to cover all the bases.
Are IHV specific DX12 extensions possible ? Nvidia have CUDA but it must be a hard sell considering it will only work on Nvidia.

 

[pottuvoi]

I guess it was too much to ask for DX12 to cover all the bases.
Are IHV specific DX12 extensions possible ? Nvidia have CUDA but it must be a hard sell considering it will only work on Nvidia.

DX11 has extensions so I would quess so.
Vulkan might be more open to deep features though, really wait the GDC as there should be great talks on subject.

 

[Kezen]

DX11 has extensions so I would quess so.
Vulkan might be more open to deep features though, really wait the GDC as there should be great talks on subject.

Are those Nvidia/AMD specific extensions used commonly ?
I mean the base API must have accumulated a number of limitations along the years so I would suspect some tech pushing devs have been using them.

 

[Dictator93]

DX11 has extensions so I would quess so.
Vulkan might be more open to deep features though, really wait the GDC as there should be great talks on subject.

Are those Nvidia/AMD specific extensions used commonly ?

Aren't all hardware exclusive effects done via specific extensions that tie into DX11 (in DX 11 titles that is)? Puttovoi?