AMD GPU14 Tech Event Sept 25 - AMD Hawiian Islands

[Jaydip]

In what benchmarks that are "compute"? The comparison of Quadro K5000 vs. FirePro isn't even relevant since most of the performance in that market segment is based on driver/optimization. If you have a program that used CUDA for 5-6 years, of course it's going to run faster on Quadro. And the discussion was for compute in games or otherwise I would have specified FirePro vs. Quadro. We should see R9 290X beat GTX780 in nearly every compute heavy title where 7970GE beat 680.

If you look at distributed computing forum, AMD cards annihilate NV in nearly every modern computing program. Folding@Home used to run faster on NV cards because it was compiled with an old inefficient code and never took advantage of GPUs with modern features. Hashing, Milkyway@home, CollatzConjecture, etc. the more modern DC programs -- NV is not even on the map. Those programs were not designed specifically to run faster on AMD hardware while NV worked for more than half a decade to optimize CUDA + developer framework for Quadro cards.

That is true for any segment.NV didn't bother to optimize drivers for their consumers line.

 

[SiliconWars]

Tessellation in Crysis 2 costs 15% on a Fermi card. And Tessellation is providing more FPS than the normal way over the CPU.

Yeah the point being the exact same visual effect could be got for 2%.

 

[GaiaHunter]

Pretty much my thought here. I have my doubts that developers will alienate potential customers by using an api that only works on specific hardware. At best we may see a few titles using this through gaming evolved while everything else runs off directx and you brute force your way to performance.

All the games will still use DX.

If indeed Mantle is the Xbox One’s low level API, then this changes the frame of reference for Mantle dramatically. No longer is Mantle just a new low level API for AMD GCN cards, whose success is defined by whether AMD can get developers to create games specifically for it, but Mantle becomes the bridge for porting over Xbox One games to the PC. Developers who make extensive use of the Xbox One low level API would be able to directly bring over large pieces of their rendering code to the PC and reuse it, and in doing so maintain the benefits of using that low-level code in the first place. Mantle will not (and cannot) preclude the need for developers to also do a proper port to Direct3D – after all AMD is currently the minority party in the discrete PC graphics space – but it does provide the option of keeping that low level code, when in the past that would never be an option.

 

[ShintaiDK]

Sound familiar? Industry desires a more efficient way to utilize hardware resources. If there is a superior alternative to DX, let it be an option. Worst case it fails, best case, DX goes the way of OpenGL and we get better looking games that run faster.

No it doesnt. You just try and twist reality :thumbsdown:

 

[raghu78]

looks like Hawaii could have some kind of hardware solution for frame compositing and pacing. :thumbsup: Tahiti and Pitcairn might still be stuck with software only fix.

http://techreport.com/news/25428/dr...inity-4k-frame-pacing-issues-coming-this-fall

"In our talk, Koduri acknowledged that the frame delivery issues with CrossFire were "unfortunate," and he expressed a desire to make things right. Koduri told me AMD is working on a fix for CrossFire and single-large-surface display configurations, like Eyefinity and tiled 4K monitors, for current Radeon graphics cards. He said AMD plans to deliver a driver with the fix this fall.

AMD's newest Radeon GPU, the "Hawaii" chip that will power the Radeon R9 290 and 290X cards announced earlier today, will of course be a top priority, as well. Although we can't yet divulge too many details, we expect Hawaii-based graphics cards to arrive with a very capable solution for CrossFire frame compositing and pacing already in place.

Some of the new Radeon cards announced today that are based on older GPUs like Tahiti and Pitcairn will presumably have to wait for the fall driver release in order to see this issue resolved."

 

[SiliconWars]

Shintai how many times have you said monopolies and lack of competition in CPU is a good thing? Double standards much?

 

[RussianSensation]

No it doesnt. You just try and twist reality :thumbsdown:

Sorry, developers almost universally disagree with you.
http://www.bit-tech.net/hardware/graphics/2011/03/16/farewell-to-directx/1

If what you are saying is true, there wouldn't be so much excitement for PS4 offering low-level access and the "wrapper" graphics API.

Knock yourself out trying to disprove this:

"In terms of rendering, there was some interesting news. Norden pointed out one of the principal weaknesses of DirectX 11 and OpenGL - they need to service a vast array of different hardware. The advantage of PlayStation 4 is that it's a fixed hardware platform, meaning that the specifics of the tech can be addressed directly. (It's worth pointing out at this point that the next-gen Xbox has hardware-specific extensions on top of the standard DX11 API.)

"We can significantly enhance performance by bypassing a lot of the artificial DirectX limitations and bottlenecks that are imposed so DirectX can work across a wide range of hardware," he revealed.

The development environment is designed to be flexible enough to get code up and running quickly, but offering the option for the more adventurous developers to get more out of the platform. To that end, PlayStation 4 has two rendering APIs.

"One of them is the absolute low-level API, you're talking directly to the hardware. It's used to draw the static RAM buffers and feed them directly to the GPU," Norden shared. "It's much, much lower level than you're used to with DirectX or OpenGL but it's not quite at the driver level. It's very similar if you've programmed PS3 or PS Vita, very similar to those graphics libraries."

http://www.eurogamer.net/articles/digitalfoundry-inside-playstation-4

You are so off the mark on this one, you might as well save your time until you get professional sources that show that DX API is more efficient than lower level API access to the hardware. Since GCN is a fixed architecture for years to come, AMD can finally unveil this lower level API for their products. I think you should up the term key competitive advantage and core competency. AMD is leveraging its product differentiation strategy NOT by locking its direct competitor but by virtue of Intelligent use of new technology that specifically benefits their own product.

If NV is still using the abacus for its API and AMD rolled out a calculator for API, no one is stopping NV from creating its own calculator for its own products. Similarly, you can't state that because NV can create 550-560mm2 die chips as it can afford to finance them with Quadro/Tesla, that it's suddenly unfair to AMD. NV's ability to release such chip is its own core competency! NV is not locking out AMD from releasing large die GPUs either.

 

[Final8ty]

Then by your own logic how can you call it progress ?

The performance part is regardless if its exclusive, the added gfx for only one brand because if its locket out is not, hence why its very unlike to add more gfx just for one brand, there has been no added gfx with AMD involvement that cant run on others hardware so again progress as all can run it.

 

[ShintaiDK]

Sorry, developers almost universally disagree with you.
http://www.bit-tech.net/hardware/graphics/2011/03/16/farewell-to-directx/1

You keep twisting my post and reality. Try read what I wrote again instead.

 

[ShintaiDK]

Shintai how many times have you said monopolies and lack of competition in CPU is a good thing? Double standards much?

Considering that you dont understand the economics of hardware, I can see why you got confused. But its an entirely different economic model for software.

 

[sontin]

You keep twisting my post and reality. Try read what I wrote again instead.

What do you expect? AMD cant compete on a hardware level. Now they switching to the software level and keep competition out. They did it with OpenCL (Winzip 16.5) and now they have an own low level api.

 

[GaiaHunter]

What do you expect? AMD cant compete on a hardware level. Now they switching to the software level and keep competition out. They did it with OpenCL (Winzip 16.5) and now they have an own low level api.

Just like Intel has their own compiler and NVIDIA has their own CUDA language, because they couldn't compete hardware wise?

Many people said that AMD problem is that they create the hardware and then they hope for developers to produce software to work for it and that they should be more aggressive pushing software to utilize their hardware.

Apparently that is what they are doing.

 

[flopper]

Just like Intel has their own compiler and NVIDIA has their own CUDA language, because they couldn't compete hardware wise?

Exactly.
always funny how people try to wrap their own beliefs into make sense evidence proposition.

 

[Rikard]

I should have saved my popcorn for this thread instead of wasting it all on Raja.

 

[RussianSensation]

What do you expect? AMD cant compete on a hardware level. Now they switching to the software level and keep competition out. They did it with OpenCL (Winzip 16.5) and now they have an own low level api.

Amazing logic. So when NV worked for a decade to optimize games to run faster on GeForce cards in their TWIMTBP program, that was because they couldn't compete on the hardware level? And when NV exposed tessellation advantage in C2-C3 and other games, esp. when 5870 got destroyed by GTX480, that was also unfair? You can't make up your mind, except in cases where AMD brings some advantage to the table you constantly attack it and downplay it. But you already said you don't care if AMD released a $500 card that matches your 1000 EUR Titan. What are you so upset about? If anything, it will force NV to release even faster cards to compete.

You might want to use the forum's search function where many of us said that NV's Fermi was more future proof for DX11 games because of its architectural tessellation advantage. It forced AMD to innovate and continue to improve its geometry performance over the next 2-3 generations. If AMD can provide lower level API to improve gaming performance, why can't NV? It's not as if AMD is locking out DX11/OpenGL game code.

I bet if NV got there first, you'd hail them as innovators for 'unlocking' Kepler's full potential.

 

[imaheadcase]

Nothing was impressive about that presentation. Mantel is not going to be supported like most things, the audio is not going to be supported other than a handful of games no doubt.

Everything they talked about was a wasted effort. All and all this further fractures the community up, if BF4 players get a huge advantage vs nvidia, BF4 lost sales. So i don't think that will be the case.

The ONLY thing that matters is performance with this card.

Review sites better have benchmarks before that BF4 preorder starts, or lots of ticked off people.

 

[DooKey]

Amazing logic. So when NV worked for a decade to optimize games to run faster on GeForce cards in their TWIMTBP program, that was because they couldn't compete on the hardware level? And when NV exposed tessellation advantage in C2-C3 and other games, esp. when 5870 got destroyed by GTX480, that was also unfair? You can't make up your mind, except in cases where AMD brings some advantage to the table you constantly attack it and downplay it. But you already said you don't care if AMD released a $500 card that matches your 1000 EUR Titan. What are you so upset about? If anything, it will force NV to release even faster cards to compete.

You might want to use the forum's search function where many of us said that NV's Fermi was more future proof for DX11 games because of its architectural tessellation advantage. It forced AMD to innovate and continue to improve its geometry performance over the next 2-3 generations. If AMD can provide lower level API to improve gaming performance, why can't NV? It's not as if AMD is locking out DX11/OpenGL game code.

I bet if NV got there first, you'd hail them as innovators for 'unlocking' Kepler's full potential.

My main issue with this is it most likely will cause NV to bring out their own version of mantle. Once that happens then it will really pollute the gaming environment. Some games will support mantle, some will support nv-mantle. Now what? Glide had the same issues. Gamers who had non-3dfx cards hated games that supported glide because their card didn't perform as well in DX as a 3dfx card did with glide. Developers read the writing on the wall and quit supporting glide unless they were paid (which as far as I know that's the only way 3dfx got them to code for it) to support it. A common API (DX, even though developers want it to be better) is better for the gamer (who isn't a video card company groupie) and isn't that what WE want?

 

[GaiaHunter]

My main issue with this is it most likely will cause NV to bring out their own version of mantle. Once that happens then it will really pollute the gaming environment. Some games will support mantle, some will support nv-mantle. Now what? Glide had the same issues. Gamers who had non-3dfx cards hated games that supported glide because their card didn't perform as well in DX as a 3dfx card did with glide. Developers read the writing on the wall and quit supporting glide unless they were paid (which as far as I know that's the only way 3dfx got them to code for it) to support it. A common API (DX, even though developers want it to be better) is better for the gamer (who isn't a video card company groupie) and isn't that what WE want?

How many videocard makers were around then?
How many games shared the same engine?
How many developers coded for multiplatform?

 

[wand3r3r]

My main issue with this is it most likely will cause NV to bring out their own version of mantle. Once that happens then it will really pollute the gaming environment. Some games will support mantle, some will support nv-mantle. Now what? Glide had the same issues. Gamers who had non-3dfx cards hated games that supported glide because their card didn't perform as well in DX as a 3dfx card did with glide. Developers read the writing on the wall and quit supporting glide unless they were paid (which as far as I know that's the only way 3dfx got them to code for it) to support it. A common API (DX, even though developers want it to be better) is better for the gamer (who isn't a video card company groupie) and isn't that what WE want?

I agree with your points for the most part.

It's going to divide the market initially, but if it catches on I guess NV and AMD will both hop on board and it won't matter as much.

 

[SiliconWars]

A common API (DX, even though developers want it to be better) is better for the gamer (who isn't a video card company groupie) and isn't that what WE want?

Uh what, you got that pretty backwards. If a gamer isn't a video card company groupie they'll get ALL the benefits by going with an AMD card. It's ONLY Nvidia groupies who refuse to switch who will lose out.

 

[sontin]

Just like Intel has their own compiler and NVIDIA has their own CUDA language, because they couldn't compete hardware wise?

nVidia supports OpenCL and DirectCompute. Cuda is there that they can leverage their feature set much faster. For graphics they use OpenGL in the same way with their own extensions.
Cuda would be same if we compare Cuda to x86 and saying that nVidia is much faster with GPGPU than AMD GPUs while using x86...
Mantle is here that developers dont use DX/OpenGL anymore to give AMD an advantage. If their hardware would be so much better they would not need an own API.
And Cuda is only usefull for the professional market. People want to solve problems. Cuda has no feature in the consumer market in which developers want to sell their software to as many people as possible.

Many people said that AMD problem is that they create the hardware and then they hope for developers to produce software to work for it and that they should be more aggressive pushing software to utilize their hardware.

And how ist Mantle resolving this problem? :hmm:

 

[96Firebird]

Uh what, you got that pretty backwards. If a gamer isn't a video card company groupie they'll get ALL the benefits by going with an AMD card. It's ONLY Nvidia groupies who refuse to switch who will lose out.

I want you to re-read this to yourself so you can see how stupid it sounds.

 

[wand3r3r]

Uh what, you got that pretty backwards. If a gamer isn't a video card company groupie they'll get ALL the benefits by going with an AMD card. It's ONLY Nvidia groupies who refuse to switch who will lose out.

Just like cude/phsyx. I don't know if fighting fire with fire is the best option, but it's funny to see the nvidians squirm and switch goalposts anyway.

 

[SiliconWars]

I want you to re-read this to yourself so you can see how stupid it sounds.

Nope, seems completely fine to me. Mantle is exactly what PC gamers have been crying out for - better graphics and performance than consoles (and less reliance on CPU). This will lead to much cheaper systems being viable gaming machines.

Oh, if you're an Nvidia fan? You got two options, either switch to red or get in the DX11 slow lane. Your choice, I'm benefiting from Mantle and you can as well. This is the crux of the issue - Nvidia never had an overwhelming product that made people switch from ATI. CUDA or PhysX, we didn't care.

Do you care enough about Mantle to switch? If so, AMD has won. Only the most lost of Nvidia fans will stay loyal to green in the face of overwhelming fps superiority.

Or you know what, maybe Nvidia could start selling cards for $100 cheaper than their AMD counterparts instead of $200 more? Imagine that!

 

[Arkadrel]

Why Mantle is differnt from Cuda/physx:

Physx cripples people that use it on the CPU.

Mantle will boost people that use it (and wont harm people that dont use it, they ll have same performance with directx as they would always have had)


This is just a feature that ll benefit AMD users.
They could end up seeing FPS in certain games being much higher than nvidia users.