D3D12 is Coming! AMD Presentation

[Despoiler]

https://www.youtube.com/watch?v=H1L4iLIU9xU

Take aways
-Ashes of the Singularity can easily max out 16 core machines.
-Break up the work to hit the specific parts of the GPU architecture. Combine work that is going to the same GPU resources. This keeps all the units busy.
-Transition uber shader to compute.
-Compute, compute, compute!
-Do any work that can be done in AMDs massive SP arrays.
-General theme. Console guys/gals are already successfully using all of these techniques. PC devs need to follow their lead to reap the same benefits.

I'm pretty positive that AMD will be in a really good performance position with Fury once these new APIs are being utilized. It's obvious to me that Fury is ahead of it's time. It's time is coming though. Their console wins are allowing them to more or less dictate what the API to GPU landscape is going to look like.

 

[.vodka]

Every day that passes I regret more and more not buying the 2600k when I built my rig back then, and went with a 2500k.

Games built from the ground up for DX12 are going to blow minds. I also believe that Fiji and Hawaii will have interesting results under such games.

 

[Black Octagon]

Every day that passes I regret more and more not buying the 2600k when I built my rig back then, and went with a 2500k.

Games built from the ground up for DX12 are going to blow minds. I also believe that Fiji and Hawaii will have interesting results under such games.

I made the exact same error. Oh well, Skylake-E it is, I guess

 

[Rickyyy369]

Every day that passes I regret more and more not buying the 2600k when I built my rig back then, and went with a 2500k.

Games built from the ground up for DX12 are going to blow minds. I also believe that Fiji and Hawaii will have interesting results under such games.

You think you regret your decision? I had the chance to go with the 4770K when I built the rig below.

 

[Despoiler]

I made the exact same error. Oh well, Skylake-E it is, I guess

AMD Zen should be out too. Either way it's looking like "moar cores" might finally come to fruition.

 

[lehtv]

Pretty sure game devs know that the vast majority of their target audience has only 4 CPU threads at their disposal. Whatever optimizations they can do for 6+ threaded processors, they better not let that get in the way of a perfectly playable experience on an i5. Otherwise, almost no one's going to buy their games.

 

[AtenRa]

Pretty sure game devs know that the vast majority of their target audience has only 4 CPU threads at their disposal. Whatever optimizations they can do for 6+ threaded processors, they better not let that get in the way of a perfectly playable experience on an i5. Otherwise, almost no one's going to buy their games.

I still remember that we gamers upgraded our hardware for a single game, DX-12 games will do just that once again

 

[spat55]

Every day that passes I regret more and more not buying the 2600k when I built my rig back then, and went with a 2500k.

Games built from the ground up for DX12 are going to blow minds. I also believe that Fiji and Hawaii will have interesting results under such games.

Not fussed because my 3570k is now over 3 years old coming into 4 years and the amount of time used and money saved will go nicely when it needs replacing, if Zen is as good as what they say it will smash this setup.

 

[james1701]

Pretty sure game devs know that the vast majority of their target audience has only 4 CPU threads at their disposal. Whatever optimizations they can do for 6+ threaded processors, they better not let that get in the way of a perfectly playable experience on an i5. Otherwise, almost no one's going to buy their games.

That's a console mentality there. If you want the industry to grow, you can't always build to the lowest common denominator. What grew the gaming desktop, Crysis is what.

 

[lehtv]

I'm not saying what developers should do, or what the industry should do. I'm not even saying what I want to happen. I'm saying what almost all developers will most certainly do - ensure that their games are smoothly playable on the current hardware of their target audience. That is the only business decision that makes sense. If a dev releases a game that requires 8+ threads for smooth gameplay and is actually amazing enough of a game that people will upgrade their rigs en masse, then so be it but that's more of a wild card, you shouldn't expect that to happen.

 

[Phanuel]

That's a console mentality there. If you want the industry to grow, you can't always build to the lowest common denominator. What grew the gaming desktop, Crysis is what.

Then Intel would need to start releasing $200 8c+ processors.

 

[freeskier93]

Then Intel would need to start releasing $200 8c+ processors.

I payed $220 for my 4C/8T Xeon E3 at Microcenter :thumbsup: Sure not 8 real cores but I'm liking my decision more an more everyday, bring on DX12!

Or just wait an see what AMD releases for Zen cuz you better bet they will have 8+ core offerings. If Zen is as good as AMD is making it out to be an 8 core Zen might blow a 4 core i5 out of the water with DX12 games, and likely be close to your $200 price point.

 

[Silverforce11]

Pretty sure game devs know that the vast majority of their target audience has only 4 CPU threads at their disposal. Whatever optimizations they can do for 6+ threaded processors, they better not let that get in the way of a perfectly playable experience on an i5. Otherwise, almost no one's going to buy their games.

Thats why games will have a options, those on lower thread/core count will have to turn down some settings.

 

[atticus14]

Then Intel would need to start releasing $200 8c+ processors.

Assuming AMD Zen brings something to the table, we'll quickly be in a core race again. If Intel has truly hit a performance wall, seeing as we get very small improvements, I can see AMD closing the performance gap and hitting that same wall relatively soon. If you want to think about it in the simplest terms, all AMD has to do is release a Sandy Bridge like Chip @ 4+Ghz with updated codecs and instructions and i'm not sure they would even be behind anymore, especially in a DX12 world and I'm sure they would be more willing to bring 6-8 core to the mainstream.

 

[3DVagabond]

Assuming AMD Zen brings something to the table, we'll quickly be in a core race again. If Intel has truly hit a performance wall, seeing as we get very small improvements, I can see AMD closing the performance gap and hitting that same wall relatively soon. If you want to think about it in the simplest terms, all AMD has to do is release a Sandy Bridge like Chip @ 4+Ghz with updated codecs and instructions and i'm not sure they would even be behind anymore, especially in a DX12 world and I'm sure they would be more willing to bring 6-8 core to the mainstream.

Hopefully Zen can shake things up and bring competition back to CPU's. It's a wait and see though. AMD hasn't shown they can compete in a very long time.

I'm holding out hope that much of it is software not taking advantage of multiple cores and that changes. Maybe even better instruction sets can take advantage of it in hardware if the software side continues to lag?

 

[Silverforce11]

Assuming AMD Zen brings something to the table, we'll quickly be in a core race again. If Intel has truly hit a performance wall, seeing as we get very small improvements, I can see AMD closing the performance gap and hitting that same wall relatively soon. If you want to think about it in the simplest terms, all AMD has to do is release a Sandy Bridge like Chip @ 4+Ghz with updated codecs and instructions and i'm not sure they would even be behind anymore, especially in a DX12 world and I'm sure they would be more willing to bring 6-8 core to the mainstream.

Intel hasn't hit the performance wall. They aren't even bothering to try. Look at the tiny die size of their Skylake. It's because they don't need to make anything better, they got the market monopoly already. So they sell you the smallest die for the maximum possible price.

The reason we're getting 600mm2 GPUs is purely because of the competitive nature in that market.

 

[RU482]

Intel hasn't hit the performance wall. They aren't even bothering to try. Look at the tiny die size of their Skylake. It's because they don't need to make anything better, they got the market monopoly already. So they sell you the smallest die for the maximum possible price.

The reason we're getting 600mm2 GPUs is purely because of the competitive nature in that market.

whats their sales ratio of U-sku vs K-sku CPUs?

 

[atticus14]

Intel hasn't hit the performance wall. They aren't even bothering to try. Look at the tiny die size of their Skylake. It's because they don't need to make anything better, they got the market monopoly already. So they sell you the smallest die for the maximum possible price.

The reason we're getting 600mm2 GPUs is purely because of the competitive nature in that market.

I'm sure Intel is holding back a little bit, but their main customers are server providers which are power oriented which value efficiency more than anything and Intel is delivering, if they want more performance just add cores. While I would like to think Intel could drop a performance bomb at anytime, I'm not entirely sure they can, outside of simply making their chips bigger. I think what the average consumer is getting screwed out of the most is lower prices and/or more cores for the same money. We are still paying 320 for a 4 core i7.

 

[Azix]

Intel hasn't hit the performance wall. They aren't even bothering to try. Look at the tiny die size of their Skylake. It's because they don't need to make anything better, they got the market monopoly already. So they sell you the smallest die for the maximum possible price.

The reason we're getting 600mm2 GPUs is purely because of the competitive nature in that market.

and because it's actually needed more or less. and because we were stuck on 28nm while intel has gone to 14nm.

I had wanted to say that when consoles came out people were thinking it would be easier to port to PC and things would get better because the consoles were on x86 CPUs and had higher RAM. I realize now that all that was not going to happen at the time because PC lacked the API to make it happen.

 

[Silverforce11]

and because it's actually needed more or less. and because we were stuck on 28nm while intel has gone to 14nm.

Even before, NV made very large GPUs, >500mm2. Because it had to. It was being pushed by a very competitive AMD.

Nobody is pushing Intel in the x86 space. So they just maximize their margins by selling you the cheapest thing (smallest die size) they make at the maximum price.

 

[lehtv]

Thats why games will have a options, those on lower thread/core count will have to turn down some settings.

We'll see how they implement that. Typically in-game settings are graphics settings and have little if anything to do with CPU load, much less with use of CPU multithreading.

 

[NTMBK]

We'll see how they implement that. Typically in-game settings are graphics settings and have little if anything to do with CPU load, much less with use of CPU multithreading.

Reduced detail means lower poly counts, which means lower setup costs for the CPU.

 

[AtenRa]

Reduced detail means lower poly counts, which means lower setup costs for the CPU.

Not only that, lower settings will provide lower unit count(NPCs), lower AI and lower physics. All those are heavily CPU bound.

 

[Sunaiac]

Reduced detail means lower poly counts, which means lower setup costs for the CPU.

If only geomety could reside on the GPU and then we would just have to upload some transformations matrices.

Oh, wait, it's not 1999 anymore !

 

[NTMBK]

If only geomety could reside on the GPU and then we would just have to upload some transformations matrices.

Oh, wait, it's not 1999 anymore !

I meant loading and unpacking resources, and streaming them to the GPU. I'm aware that hardware T&L has been around for a while now