Ask Me Anything - Official AMD Radeon Representative

[cmi86]

How far back will FreeSync support go for your graphics cards, and is this a hardware issue or a development choice? I'm mostly concerned about the HD7970 (for selfish reasons *looks at signature*).

The supporting GPUs:

Kaveri, Beema, Mullins, Temash, R9 295X2, R9 290X, R9 290, R9 285, R7 260X, R7 260.

This is a decision that reflects hardware capabilities. It's not a software/development choice.

Are you allowed to discuss anything in regard to the upcoming radeon line ? I have notice any questions regarding them have been side stepped, mine included.

 

[Computer LVR]

Will mid range GPUs such as the r9 270x get VSR support in a future driver update?

 

[fatboyslimerr]

Are you planning to stay competitive in the mobile GPU market???

 

[gogo8092201]

I have been super stoked about Mantle since I first heard about it. Now it is a switch you can flip and get "free" FPS. Awesome! I can believe adoption has been an uphill battle, is AMD still pushing Mantle, and do you have plans to continue it beyond DX12?

 

[Thracks]

Is the mantle API version 98304 effectively the gold release then?

The Mantle API is still in beta.

then what about AMD promised to release the public SDK this year?

It is still our plan to release the Mantle SDK publicly. It was born to be open.

 

[Skysnake]

Hi,

nice to see again a Q&A session.

Are there new informations, when OpenCL2.0 comes to the Radeon GPUs?

And from the perspective of a developer, what is with the "GlobalDataShare" in the GNC architecture. This feature ist named since the GCN start, but until now, there is no explanation, how it works. Only the ISA give some minor hints...
It looks like there would be a possebility for syncs over workgroup borders, what would be very very very cool! Everybody wants something like this, as a GPGPU-Developer.

And another question. There is still much rumors about Frametimes and nVidias FCAT System. Will we see a improved version from AMD, that really show the frametimes compared to the ingame timer? So that all 32Bit of the color room are used, as a timestomp code, and not only 6 or so colors?

And the last question to HPC. RDMA is a very nice thing, is it possible to use RDMA over Infiniband with AMD GPUs, or will we see this feature in the near future?

Tanks for your answers.

 

[Thracks]

I have been super stoked about Mantle since I first heard about it. Now it is a switch you can flip and get "free" FPS. Awesome! I can believe adoption has been an uphill battle, is AMD still pushing Mantle, and do you have plans to continue it beyond DX12?

Adoption has been extremely swift, with 5 engines/20+ games and now ~100 developers working with Mantle. Mantle will coexist with DX12 as a different tool in the toolbox for devs to use if it makes sense to do so. We are ready to support those devs.

But I also agree with your implication that DX12 is very exciting, and poised for broad adoption. We intend to support it on all of our GCN-based GPUs. I can't wait!

 

[Thracks]

Hi,

nice to see again a Q&A session.

Are there new informations, when OpenCL2.0 comes to the Radeon GPUs?

And from the perspective of a developer, what is with the "GlobalDataShare" in the GNC architecture. This feature ist named since the GCN start, but until now, there is no explanation, how it works. Only the ISA give some minor hints...
It looks like there would be a possebility for syncs over workgroup borders, what would be very very very cool! Everybody wants something like this, as a GPGPU-Developer.

And another question. There is still much rumors about Frametimes and nVidias FCAT System. Will we see a improved version from AMD, that really show the frametimes compared to the ingame timer? So that all 32Bit of the color room are used, as a timestomp code, and not only 6 or so colors?

And the last question to HPC. RDMA is a very nice thing, is it possible to use RDMA over Infiniband with AMD GPUs, or will we see this feature in the near future?

Tanks for your answers.

Hi, I am not a developer so I am unfortunately unable to answer many of your questions. However, as of Catalyst Omega, OpenCL 2.0 is supported on our recent products.

 

[Thracks]

How far back will FreeSync support go for your graphics cards, and is this a hardware issue or a development choice? I'm mostly concerned about the HD7970 (for selfish reasons *looks at signature*).

The supporting GPUs:

Kaveri, Beema, Mullins, Temash, R9 295X2, R9 290X, R9 290, R9 285, R7 260X, R7 260.

This is a decision that reflects hardware capabilities. It's not a software/development choice.

Are you allowed to discuss anything in regard to the upcoming radeon line ? I have notice any questions regarding them have been side stepped, mine included.

I am never allowed to discuss any future products. I have strict confidentiality agreements with my employer. I happen to enjoy my job very much and don't want to lose it!

 

[Thracks]

Will mid range GPUs such as the r9 270x get VSR support in a future driver update?

Hi, yes. We intend to add 4K VSR to all products in the R7 and R9 family, starting with the R7 260 and up. Certain products do not feature the necessary hardware to utilize VSR in its current implementation, so we must explore alternative algorithms and that will take a little more dev time. We expect that driver in Jan/Feb.

 

[Jtaylor1986]

Has AMD ever explored disabling directX support for mantle titles to force developers to do a better job optimizing? Speaking from experience Mantle performance in Battlefield 4 and Dragon Age Inquisition has been generally inferior to DX11 which is very disappointing.

 

[Thracks]

That said, FreeSync and DisplayPort Adaptive-Sync are different. DPAS just allows you to control the display. How do you do that? You need some sort of "secret sauce." That's FreeSync.

How much "secret sauce" could there possibly be behind flipping the output frame buffer once the current frame finishes rendering and triggering the frame dump to the display?

It seems to me like it should be a trivial tweak over the existing vsync page flip. (Except maybe for the extra work to signal the variable refresh over DP. I have not looked at how Adaptive Sync works at the protocol level but if many existing scalers can be firmware-upgraded with AS, I presume the signaling changes must be similarly trivial.)

My point was that DPAS is the technology that enables the solution. Just because you've connected two DPAS-aware products doesn't mean you'll also magically receive dynamic refresh rate. That's all I can say for now.

 

[Avenger762]

I know you said that CPU's aren't your department, BUT is there anything you can tell us on the CPU front. I've been using a 965BE for 5 years now and still have no compelling replacement CPU. I love AMD CPU's and am still waiting for a new CPU worthy of upgrading to.

 

[Thracks]

Has AMD ever explored disabling directX support for mantle titles to force developers to do a better job optimizing? Speaking from experience Mantle performance in Battlefield 4 and Dragon Age Inquisition has been generally inferior to DX11 which is very disappointing.

I think that would be a horrible idea. People should be able to choose whatever API they want, developers and gamers alike.

 

[ubercake]

One of the problems I see with Nvidia's release of G-sync technologies is an extreme inability to meet demand (keeping prices high). We see one monitor with a very limited supply released to market followed by another manufacturer's release of an extremely limited supply. I'm not sure why this is exactly. G-sync is great tech, but if you can't get it to everyone that wants it, there's a problem.

I have a feeling AMD can really capitalize on this and get Freesync into more homes just by manufacturers keeping supplies of Adaptive-Sync monitors available. Will Samsung and other vendors be releasing monitors with Adaptive-Sync in ample numbers? Are they making it part of their normal production process (since it is a standard?)?

I really think if people begin experiencing the benefits of the adaptive-sync technology, they won't want to ever go back to a display without it. On top of it, if it's not an expensive monitor add-on (not adding to the cost of the monitor) and many manufacturers adopt it as standard, that gives AMD a real strategic advantage.

 

[gogo8092201]

I think one of the problems is the gap between what Mantle was proposed to be, and what it is right now. To me, getting the dev's closer to the hardware just makes sense. Especially where engines are concerned. Tweak an engine here or there, fix issues/increase performance in multiple games. I am in love with parallelism, and while it doesn't solve all problems, it seems that mantle is way more optimized for top end hardware than DX11 ever could be. I guess that is why it is not intuitive that the lower end CPU's with beast GPU's seem to have the most to gain. From what you have seen, and water cooler talk, do you expect future improvements on how devs interact with Mantle could realize greater gains for high end hardware? Make a 6 or 8 core CPU more attractive? Share what you can, I understand NDA's when you can't.

 

[Skysnake]

Hi,

nice to see again a Q&A session.

Are there new informations, when OpenCL2.0 comes to the Radeon GPUs?

And from the perspective of a developer, what is with the "GlobalDataShare" in the GNC architecture. This feature ist named since the GCN start, but until now, there is no explanation, how it works. Only the ISA give some minor hints...
It looks like there would be a possebility for syncs over workgroup borders, what would be very very very cool! Everybody wants something like this, as a GPGPU-Developer.

And another question. There is still much rumors about Frametimes and nVidias FCAT System. Will we see a improved version from AMD, that really show the frametimes compared to the ingame timer? So that all 32Bit of the color room are used, as a timestomp code, and not only 6 or so colors?

And the last question to HPC. RDMA is a very nice thing, is it possible to use RDMA over Infiniband with AMD GPUs, or will we see this feature in the near future?

Tanks for your answers.

Hi, I am not a developer so I am unfortunately unable to answer many of your questions. However, as of Catalyst Omega, OpenCL 2.0 is supported on our recent products.

What a pity 🙁

I am waiting since years now for the corresponding informations. :/

 

[ubercake]

Which GPUs will have Freesync technology available upon release of the Adaptive-sync monitors?

 

[Jtaylor1986]

I don't necessarily agree as Mantle optimization is always going to be 2nd fiddle to DX optimization in developers to do list just based on user base demographics and finite development time. Poorly implemented or optimized Mantle is basically unusable though given it's a low level API so it's kind of a catch 22 no? You are shooting yourselves in the foot by putting all the onus on developers and giving them an out in DX11. We've seen how its played out already cards with less than 2GB VRAM being unusable in BF4 1 year post launch and Dragon Age Inquisition mantle performance being significantly inferior until a patch was issued 3 weeks post launch. My fear is that this API has great potential but it's future is extremely uncertain unless it reaches a critical mass given results to date and DX12 on the horizon.

 

[smeezekitty]

Will features that are claimed to be "R series" specific (such as those in Catalyst Omega) work on the 7xxx equivalents to their R series rebrand?

For example 5k monitor support

5K Monitor Support:
Unparalleled display performance for professionals and gamers who require the highest resolution available and silky-smooth, fatigue-free graphics

5120x2880 at 60Hz
14.7 million pixels, 218 pixels-per-inch
Dual DisplayPort 1.2 input
Available on AMD FirePro™ and AMD Radeon™ R series GPUs

And "Perfect Picture Additions"

Available on AMD 7000 series APUs and AMD Radeon™ R series GPUs

 

[Thracks]

Which GPUs will have Freesync technology available upon release of the Adaptive-sync monitors?

295X2, 290X, 290, 285, 260X, 260, Beema, Mullins, Temash, Kaveri.

 

[Daniel Sauvageau]

One of the problems I see with Nvidia's release of G-sync technologies is an extreme inability to meet demand (keeping prices high). We see one monitor with a very limited supply released to market followed by another manufacturer's release of an extremely limited supply. I'm not sure why this is exactly. G-sync is great tech, but if you can't get it to everyone that wants it, there's a problem.

The problem is simple: single source / single vendor.

Monitor manufacturers cannot manufacture more GSync monitors than the number of GSync scalers they are able to source from Nvidia and they cannot get them cheaper than Nvidia's asking price.

I always suspected Adaptive Sync was a trivial tweak but not to the point of being a mere firmware tweak in many cases. If scaler manufacturers do release such firmware updates for their existing chips, that should tip the scale quite significantly. Too bad the majority of displays out there that use one of those firmware-upgradable scalers likely have no hardware and software accommodations to actually do the firmware update in a user-friendly manner.

 

[Thracks]

I think one of the problems is the gap between what Mantle was proposed to be, and what it is right now. To me, getting the dev's closer to the hardware just makes sense. Especially where engines are concerned. Tweak an engine here or there, fix issues/increase performance in multiple games. I am in love with parallelism, and while it doesn't solve all problems, it seems that mantle is way more optimized for top end hardware than DX11 ever could be. I guess that is why it is not intuitive that the lower end CPU's with beast GPU's seem to have the most to gain. From what you have seen, and water cooler talk, do you expect future improvements on how devs interact with Mantle could realize greater gains for high end hardware? Make a 6 or 8 core CPU more attractive? Share what you can, I understand NDA's when you can't.

Low-overhead APIs give the most to lower end CPUs, because it's lower end CPUs that suffer the most from API overhead. In other words, those are the platforms with more to gain.

There are many techniques in the Mantle API specification that can enable robust experiences on high-end hardware. See split-frame rendering in Civ:BE as a prime example of that. But Mantle is young, and developers are still learning Mantle as they go. The emphasis is definitely on a working implementation that addresses the most number of platforms with improvements. By the numbers, those are mid-range systems.

 

[renz496]

Is the mantle API version 98304 effectively the gold release then?

The Mantle API is still in beta.

then what about AMD promised to release the public SDK this year?

It is still our plan to release the Mantle SDK publicly. It was born to be open.

any possible time frame?

 

[cmi86]

Which GPUs will have Freesync technology available upon release of the Adaptive-sync monitors?

The supporting GPUs:

Kaveri, Beema, Mullins, Temash, R9 295X2, R9 290X, R9 290, R9 285, R7 260X, R7 260.

This is a decision that reflects hardware capabilities. It's not a software/development choice.

That was his answer. He was not willing to speculate about anything future which is understandable.

However what I do not understand is that 3 of those free sync GPU's are re brands from the previous generation
that do not support free sync. I don't know if i buy that it is a hardware limited thing.. It seems more like a cash grab to get people out of still capable 7XXX cards and get them in to the R9/R7 line. The old proverbial carrot on a stick if you will.

I love AMD and have for years but at the end of the day they are still a corporation, and a semi struggling one at that.