direct x 12 is very promising...for AMD

[monstercameron]

Recap
DirectX 12 save more than 50% of CPU power compared to DirectX 11.

When allowed to use equivalent power to DirectX 11, DirectX 12 can also yield 50% better FPS performance.

DirectX 12 can run and show these improvements on a wide variety of devices including Microsoft’s flagship device the Surface Pro 3.

There’s more to come; more demos, more power gains, more performance gains; sign up for the DirectX 12 Early Access Program to keep up to date!

source: http://blogs.msdn.com/b/directx/arc...-high-performance-and-high-power-savings.aspx

opinion: If intel was to get such a decent boost in this early state, then in the future AMD apus would get a big boost just by lowering the cpu pressure and allowing its gpu to stay in a higher power state [all this is assuming a mobile platform]. now we can see why intel wanted access to experiment with mantle.

 

[bystander36]

I'm not sure APU's really gain any more from this than high end systems. I'd bet the 120hz/SLI/CF crowd gains the most.

While APU's have slow CPU's, it is matched with a slow GPU, so the CPU really hasn't been that big of a bottleneck. 120hz/SLI/CF users are the ones who experience the most CPU bottlenecks.

I'd say that AMD normal CPU's are the ones that will gain the most, as they are more likely to be paired with faster GPU's. And everyone really wins, as some games simply are CPU hogs, like RTS games.

 

[alcoholbob]

Since we've been stuck on 28nm for 4 years and counting, and likely 5 years+ before we move on to a new node, future progress will have to come from software optimizations because the metal is progressing at a snail's pace.

 

[Silverforce11]

I'm not sure APU's really gain any more from this than high end systems. I'd bet the 120hz/SLI/CF crowd gains the most.

While APU's have slow CPU's, it is matched with a slow GPU, so the CPU really hasn't been that big of a bottleneck. 120hz/SLI/CF users are the ones who experience the most CPU bottlenecks.

I'd say that AMD normal CPU's are the ones that will gain the most, as they are more likely to be paired with faster GPU's. And everyone really wins, as some games simply are CPU hogs, like RTS games.

I don't think you understand.

If it reduces CPU overhead and thus, require less CPU computing, the CPU can downclock to save power, allowing the GPU to boost higher, the entire SOC/APU will still be within its constrained TDP (for mobiles very important due to cooling), but overall the performance is much higher.

This tech is great for Intel & AMD on mobiles, but greater for Intel on desktops and high-end rigs with multi-cards, because even now some games are bottlenecked by OC i7 in Crossfire setups.

 

[RussianSensation]

Since we've been stuck on 28nm for 4 years and counting, and likely 5 years+ before we move on to a new node, future progress will have to come from software optimizations because the metal is progressing at a snail's pace.

The first 28nm product was HD7970 that launched in December 2011. It's thus been slightly less than 3 years on 28nm node, not 4 years. For NV it's 2.5 years since GTX680 launched. If 20nm launches late next year, it'll be 4 years since 28nm node.

---

DX12 should benefit not just AMD but NV systems too. Not sure why in the title only AMD is mentioned. Those results look promising. If DX12 fulfills on its promise, it could create another wave of upgrades which is healthy for PC hardware.

 

[sandorski]

The first 28nm product was HD7970 that launched in December 2011. It's thus been slightly less than 3 years on 28nm node, not 4 years. For NV it's 2.5 years since GTX680 launched. If 20nm launches late next year, it'll be 4 years since 28nm node.

---

DX12 should benefit not just AMD but NV systems too. Not sure why in the title only AMD is mentioned. Those results look promising. If DX12 fulfills on its promise, it could create another wave of upgrades which is healthy for PC hardware.

I think he was thinking on the CPU side of things.

 

[TechFan1]

I thought it was interesting that it was a surface pro 3 rendering what equated to 50,000 draw calls per frame. Hopefully that bodes very well for game developers being able to do many more draw calls on full fledged pcs.

 

[ShintaiDK]

The reduced CPU overhead will simply be used for other things. The period of lower CPU overhead will only be temporary. Its just a matter of time before the first DX12 only game have used any savings on new things like physics, AI, higher complexity etc. And we are back to the same spot as today.

 

[dacostafilipe]

And we are back to the same spot as today.

:hmm:

More drawcalls, more objects, better multithreading ... but still the same.

Makes total sense.

 

[frozentundra123456]

So far, we have seen only small gains with mantle on apus using the igp, because they are not CPU limited. This is obviously a demo designed to load up the cpu and show the most improvement, not a real world game. Whatever benefits are seen will obviously benefit both Intel and AMD.

 

[Keysplayr]

n/m. got my mind reversed.

 

[Wild Thing]

The reduced CPU overhead will simply be used for other things. The period of lower CPU overhead will only be temporary. Its just a matter of time before the first DX12 only game have used any savings on new things like physics, AI, higher complexity etc. And we are back to the same spot as today.

The reduction in CPU overhead and thermals means it can be used not for "other things" but "more things".
It's an admirable aim or outcome.:thumbsup:

 

[bystander36]

I don't think you understand.

If it reduces CPU overhead and thus, require less CPU computing, the CPU can downclock to save power, allowing the GPU to boost higher, the entire SOC/APU will still be within its constrained TDP (for mobiles very important due to cooling), but overall the performance is much higher.

Hasn't it already been shown that Mantle doesn't improve APU performance much at all, unless paired with a discrete GPU? DX12 isn't looking to do anything more than Mantle in lowering CPU overhead.

This tech is great for Intel & AMD on mobiles, but greater for Intel on desktops and high-end rigs with multi-cards, because even now some games are bottlenecked by OC i7 in Crossfire setups.

And that is what I said.

 

[ShintaiDK]

The reduction in CPU overhead and thermals means it can be used not for "other things" but "more things".
It's an admirable aim or outcome.:thumbsup:

Oh I fully agree. The point was that any lower CPU utilization would quickly be used and not left untouched.

 

[biostud]

The reduced CPU overhead will simply be used for other things. The period of lower CPU overhead will only be temporary. Its just a matter of time before the first DX12 only game have used any savings on new things like physics, AI, higher complexity etc. And we are back to the same spot as today.

But we will have better physics, AI and complexity.

 

[Silverforce11]

Hasn't it already been shown that Mantle doesn't improve APU performance much at all, unless paired with a discrete GPU? DX12 isn't looking to do anything more than Mantle in lowering CPU overhead.

Whenever you have 30W limit to play with, if you only need your CPU to run at 5W instead of its normal 15W half-half split (you get the point), it will downclock the CPU the and overclock the GPU part.

Current gen APUs don't benefit much because their iGPU have a set limit boost. They need to have better TDP management between CPU/GPU based on load monitoring as well.

So its pretty much next gen stuff that will take advantage of this feature, probably just in time for DX12 release!

ps. Hawaii already dynamicall adjust its clock speed based on load. If its not heavy loaded it will downclock to 600-700mhz etc.

 

[Erenhardt]

So far, we have seen only small gains with mantle on apus using the igp, because they are not CPU limited. This is obviously a demo designed to load up the cpu and show the most improvement, not a real world game. Whatever benefits are seen will obviously benefit both Intel and AMD.

I guess it depends on the workload. Here are some benches from latest mantle title: PvZ Garden Warfare

http://pclab.pl/art58492-3.html

Im not sure what they did there, but in some cases that is 30% boost in avg. FPS and knowing how mantle worked for other games, better minimums and frame times aswell.

Lowering power requirements for the CPU leaves more power margin for IGP. This is very good especially for AMD mobile systems due to higher power requirements of their CPUs.

So in TDP limited mobile environment both performance improvements stack with each other making it a game-changer

 

[SPBHM]

with the 7850K and Mantle is more of the typical benefit of just reducing the CPU bottleneck for higher GPU usage (the IGP is not going to clock any higher with Mantle I think), the interesting thing for the future is something else, like for Intel, their CPU and IGP are fighting for the the TDP (specially 15W CPUs and lower), with less CPU power wasted you can run the IGP at a higher clock and achieve better performance... so yes, lower overhead API is extremely interesting even for the slow IGPs from Intel.

 

[ShintaiDK]

I guess it depends on the workload. Here are some benches from latest mantle title: PvZ Garden Warfare

http://pclab.pl/art58492-3.html

Im not sure what they did there, but in some cases that is 30% boost in avg. FPS and knowing how mantle worked for other games, better minimums and frame times aswell.

Lowering power requirements for the CPU leaves more power margin for IGP. This is very good especially for AMD mobile systems due to higher power requirements of their CPUs.

So in TDP limited mobile environment both performance improvements stack with each other making it a game-changer

That simply looks CPU limited. The 6800K is the giveaway.

 

[Lonyo]

Would be more interesting to see power use in the Mantle vs non-Mantle PvZ benchmarks rather than FPS, to see if there is indeed an impact on power (which is kind of what this thread was about anyway).

Anyone here have an APU and power monitor and want to test it out?

 

[monstercameron]

Would be more interesting to see power use in the Mantle vs non-Mantle PvZ benchmarks rather than FPS, to see if there is indeed an impact on power (which is kind of what this thread was about anyway).

Anyone here have an APU and power monitor and want to test it out?

the power reductions were from the cpu side, overall package power reductions might not happen.

 

[Lonyo]

the power reductions were from the cpu side, overall package power reductions might not happen.

That's why you test to find out. Does overall power change or stay the same. Even if it stays the same, that's with higher fps.
Can you set Windows/AMD power management to cap speed and reduce power but have the same FPS as non-Mantle etc.

 

[Hitman928]

I think AtenRa ran power numbers when benching DX11 vs Mantle in Thief and found lower power usage with Mantle. Can't remember for sure though.

 

[frozentundra123456]

I guess it depends on the workload. Here are some benches from latest mantle title: PvZ Garden Warfare

http://pclab.pl/art58492-3.html

Im not sure what they did there, but in some cases that is 30% boost in avg. FPS and knowing how mantle worked for other games, better minimums and frame times aswell.

Lowering power requirements for the CPU leaves more power margin for IGP. This is very good especially for AMD mobile systems due to higher power requirements of their CPUs.

So in TDP limited mobile environment both performance improvements stack with each other making it a game-changer

On the other hand, a recent article in Tom's hardware showed only a 10% increase in Thief and essentially no increase in BF4. So you can pick whatever benchmark you want. Point is, it is not a foregone conclusion that Mantle (or DX 12) is going to make some kind of magic transformation in an APU. We need more information at different settings in a lot more games to come to some kind of conclusion. And as I stated, that demo cited by the OP is obviously programmed to load up the cpu with draw calls and show DX12 in the best light.

 

[witeken]

Promising for AMD? You know that this is a demo from Intel, right?