AT has dx12 drivers?

[96Firebird]

I dont recall this anticipation of massive increase in drawcall when discussing mantle one and a half year ago

Refresh your memory.

 

[geoxile]

I really doubt most devs are going to take advantage of the extra CPU time. IIRC the Metro devs mentioned the PS4 already had a very low-level API and the X1 was getting one in pieces - this was a year or so ago - and considering most games these days are cross-platformers I can't see the devs straying far from the console versions.

 

[Silverforce11]

After watching the video again, the graphics look different, the 980 looks to me missing certain things compared to the 290x. On the 980 the ships go from having a blue trail to having none, to having blueish glow again. Also when there are explosions at first you will see things coming away leaving a red trail, I don't see that on the 980. There are also ships that shoot out bright red things that are grains of rice shaped. On the 980 there is no color to them at all and you can barely tell the ships are firing.

This is on DX12. I am sure NVidia will get that sort of thing figured out.

Edit: also will account for some of the performance difference.

Hard to notice with so much going on in the scene.

 

[bystander36]

People use to keep saying Mantle only benefits weaker CPUs, absolutely not true. We've seen in BF4 MP with Crossfire, the performance gains and smoothness is huge.

Mantle or DX12 alleviates CPU bottlenecks, which occur on weaker CPUs OR Multi-GPU systems.

In situations where its not CPU limited, making the CPU work less and reducing power use also benefits Intel, because their CPUs are already much stronger than AMD. AMD's CPU will still have to work harder and thus consuming more power for the same level of performance.

Nothing truly helps AMD's CPU besides them making a good CPU architecture, period.

You seem to be contradictory. You seem fully aware that Mantle, and DX reduces CPU bottlenecks, yet are arguing that DX12 and Mantle will have no effect on AMD and Intel's comparable value when it comes to gaming.

With low overhead DX and Mantle API's, the huge lead that Intel has is reduced. Sure, there will be times when Intel does better, especially for those looking to play at 300 FPS, but for all practical purposes, the gap is reduced, making AMD a more reasonable choice.

 

[bystander36]

I dont recall this anticipation of massive increase in drawcall when discussing mantle one and a half year ago fair enough its not rts all over where its typically very much needed.
But anyway a thin api is far more than that.

Sure they did. Didn't you follow the Star Citizen demo? That was all about draw calls. The reason it tanks on DX11 is draw call limitations and that is what they were all excited about.

 

[Shehriazad]

@ShintaiDK

Looking at 290x DX12 performance. Just daaaaamn. Hope 3xx is better.

For the fact that the 290X is an entire year OLDER than the 980...it holds up insanely well...I don't even know what you're expecting from a card that's from 2013.

In DX12 the 980 that came out a year later seems to be doing 55% better(first graph)...but it also pretty much exactly costs 55% more (at least over here in Germany)...so from a price to performance view Nvidia seems far from beating a card that is from 2013.... just sayin'.

 

[bystander36]

Here is a thought; what if DX12 has a major effect on the performance parity of Nvidia and AMD? What is one or the other is held back a lot more, due to drivers, and with a thin layer, some cards may become a lot more powerful than previously seen.

I do, doubt it would be much of a difference, but we really don't know.

 

[Paul98]

As I mentioned above, the 980 looks to be not be rendering certain parts which could very well cause it to have better performance. Plus this is not a real benchmark as in it's not the same every run and doesn't go a fixed length. So it really doesn't tell you much other than the CPU overhead is greatly reduced.

 

[Silverforce11]

As I mentioned above, the 980 looks to be not be rendering certain parts which could very well cause it to have better performance. Plus this is not a real benchmark as in it's not the same every run and doesn't go a fixed length. So it really doesn't tell you much other than the CPU overhead is greatly reduced.

Can you take some screenshots to compare the differences?

 

[Paul98]

Can you take some screenshots to compare the differences?

Look at 15, 22, and 56 on the 290x DX12

Then look at 10, 14, 37 on the 980 DX12

First one look the middle top left you will see red streaks behind bright glowing pieces to the left of the explosions on the 290x.

Second looks at the ships that just spawned at the middle bottom, you will see blueish jets coming out the back, on the 980 just the ships.

Last you will see bright red shots coming from the ship on the bottom on the 290, and barely see anything from the shots leaving the same ship on the 980.

 

[Silverforce11]

Look at 15, 22, and 56 on the 290x DX12

Then look at 10, 14, 37 on the 980 DX12

First one look the middle top left you will see red streaks behind bright glowing pieces to the left of the explosions on the 290x.

Second looks at the ships that just spawned at the middle bottom, you will see blueish jets coming out the back, on the 980 just the ships.

Last you will see bright red shots coming from the ship on the bottom on the 290, and barely see anything from the shots leaving the same ship on the 980.

Damn, you're right. Nice spot.

Definitely they render the scene differently.

 

[Shehriazad]

So the 900 cards aren't just missing Vram and ROPs...they are also missing some scenery *trolls away* XDDDD

 

[garagisti]

Interesting and great to see DX12 can give a huge boost in FPS to anything that doesn't support mantle.

Micro-screw?

 

[Carfax83]

As I mentioned above, the 980 looks to be not be rendering certain parts which could very well cause it to have better performance. Plus this is not a real benchmark as in it's not the same every run and doesn't go a fixed length. So it really doesn't tell you much other than the CPU overhead is greatly reduced.

You gave the answer yourself to your own question.

Starswarm is not a benchmark. It's more of a stress test, and each run is different..

So implying that the 980 is receiving some kind of unfair advantage is moot..

 

[Silverforce11]

You gave the answer yourself to your own question.

Starswarm is not a benchmark. It's more of a stress test, and each run is different..

So implying that the 980 is receiving some kind of unfair advantage is moot..

Well if its not rendering some special effects, explosion sprites, ship trails... must be beta drivers on beta windows. :/

 

[Paul98]

You gave the answer yourself to your own question.

Starswarm is not a benchmark. It's more of a stress test, and each run is different..

So implying that the 980 is receiving some kind of unfair advantage is moot..

Each run is a little different, but it is actually not rendering things that it should be, as the objects are there in all just the 980 isn't rendering them in DX12. That said I am sure that will be fixed with drivers, just it will also lower performance.

Thus as I said you can't draw any real comparison here except that DX12 is similar to mantle in performance, and has much lower overhead.

 

[Silverforce11]

No, it is actually not rendering things that it should be.

Or the reverse is true, the R290X is rendering extra things it shouldn't be. :/

 

[Paul98]

Or the reverse is true, the R290X is rendering extra things it shouldn't be. :/

Not for things like that, as it's only not rendered on 980 dx12, while being rendered on dx11.

 

[Silverforce11]

Not for things like that, as it's only not rendered on 980 dx12, while being rendered on dx11.

Now that's just plain sneaky... maybe AT's Ryan Smith care to shed some inputs?

 

[Carfax83]

You guys are making a mountain out of a molehill. I just looked at those frames you pointed out, and I saw nothing amiss. There are differences of course, because as I pointed out earlier, every run is different. Different numbers of units, and likely even different types of units.. The A.I is calculated in real time, so every battle is almost unique in some manner.

Thats one of the reasons why I never liked Starswarm. I used it when it first released last year, and I wasn't too impressed by it's variable nature. But then to be fair, the developers never intended for it to be a deterministic benchmark.

Which is why on the Steam download page, it says "Stress Test," and not benchmark.

 

[AnandThenMan]

If it's not a benchmark it's sure being presented as one.

It should be noted that while Star Swarm itself is a synthetic benchmark, the underlying Nitrous engine is relevant and is being used in multiple upcoming games. Stardock is using the Nitrous engine for their forthcoming Star Control game, and Oxide is using the engine for their own, yet-to-be-announced game. So although Star Swarm is still a best case scenario, many of its lessons will be applicable to these future games.

As for the benchmark itself, we should also note that Star Swarm is a non-deterministic simulation. The benchmark is based on having two AI fleets fight each other, and as a result the outcome can differ from run to run. The good news is that although it’s not a deterministic benchmark, the benchmark’s RTS mode is reliable enough to keep the run-to-run variation low enough to produce reasonably consistent results. Among individual runs we’ll still see some fluctuations, while the benchmark will reliably demonstrate larger performance trends.

 

[Alatar]

Well if its not rendering some special effects, explosion sprites, ship trails... must be beta drivers on beta windows. :/

Each run is a little different, but it is actually not rendering things that it should be, as the objects are there in all just the 980 isn't rendering them in DX12. That said I am sure that will be fixed with drivers, just it will also lower performance.

Thus as I said you can't draw any real comparison here except that DX12 is similar to mantle in performance, and has much lower overhead.

Or the reverse is true, the R290X is rendering extra things it shouldn't be. :/

I've reproduced the situation on mantle before. Last year actually.

280X DX11 vs. mantle

(open in a new tab)

I have no idea why it happened but I was never able to see any performance differences due to the trails (or whatever the blue things are) being less noticeable.

Either way the real performance killer in star swarm is motion blur, nothing else will give you extremely big performance differences. Motion blur will drop your perf in half.

 

[Abwx]

All this is well but was it mentioned that DX12 will be available only with Windows 10..?.

http://www.hardware.fr/news/14073/directx-12-benchmarks-exclusivite-windows-10.html

 

[Carfax83]

If it's not a benchmark it's sure being presented as one.

Star Swarm stress test

The most accurate description I think would be a real time non deterministic simulation.

 

[Silverforce11]

I've reproduced the situation on mantle before. Last year actually.

So its not even rendering the same scene.. lots of ships have no trails. Then explosions have no particles.