AMD Driver overhead tested (PCLAB.pl)

[App4that]

4 minutes ago, Prysin said:

learn to read, then come back here.

 

 

I was offering my assistance 0.o

[SteveGrabowski0]

Great thread, @Majestic. It's frustrating reading all the AMD has the best GPU at every price point other than $650 over and over on this forum. Seeing the R9 380 recommended over an over with i3s and even Pentiums even though Digital Foundry has been showing problems with AMD gpus and weak cpus since November 2014. I can't believe how strong brand loyalty to AMD can be here. If everyone bought a 6700k or a 5820k then all the AMD love here would be completely justified, but it makes a lot less sense when everyone here loves the i5-4460.

[App4that]

1 minute ago, SteveGrabowski0 said:

Great thread, @Majestic. It's frustrating reading all the AMD has the best GPU at every price point other than $650 over and over on this forum. Seeing the R9 380 recommended over an over with i3s and even Pentiums even though Digital Foundry has been showing problems with AMD gpus and weak cpus since November 2014. I can't believe how strong brand loyalty to AMD can be here.

I share in your frustration. The irony is that this forum proves Nvidia is not the company the AMD fans make it out to be. If Nvidia was half as bad as their reputation on this forum is, they'd pull support from Linus. You don't think Nvidia is aware of this forum? They are. Yet they continue their support.

 

 

[Prysin]

@Majestic
Looking through PClabs.pl tests on The Witcher 3, i think i know the area they are testing "Nilfgardian Connection" and "Swamp", and if it is (i will log in and go there myself), the differences we see MAY actually be Nvidia effects playing part. There are some places in TW3 where my 295x2 starts to take a dump for reasons i cannot explain. With a 4790k @ 4.4Ghz (too lazy to bother with more atm) and 32GB of 1866MHz RAM (middle of the road speed). I should have NO bottlenecks short of Drivers and GPU.
CF works flawlessly in TW3, it is fine, stuttering is nearly non existent (happens in battles with lots of NPCs and fast movements).

 

So why do i get issues? I think it may be HBAO+... which is a gorgeous effect, and i refuse to turn it off, even if it is a Nvidia effect. I use it in FC4 too. Gorgeous, wont turn it off.

But it does, in some places give me weird dips in not minimum, but maximum FPS....

For the Forest section of the PClabs.PL, we see AMDs shader array coming into play. All those trees need shading, lots of shadows to apply, light to warp and process. AMD destroys Nvidia in those things. So this is why AMD is ahead in Forest. As for which forest it is, i dunno. It prolly is the same map as the two other tests. Which is a shame, because there is more detailed and heavier tesselated (nvidia bonus) forsests in other maps.

 

NPC density is low on the second map you get to, which is where it seems they did this test. Which means i am sad to say, but take any tests between FX CPUs and Intel CPUs with a grain of salt, because if i am correct. The NPC density (aka CPU load) is NOT very high in this map, because both friendly and hostile NPC count is extremely low compared to other sections of the game.

IF i am correct though. IF.... They may be testing somewhere else, but i suspect not. This seems like it is done within the first hour of gameplay in terms of mission names.

[NuclearPeace]

This probably has to be one of the most frustrating things about my 380x. My i3 4370 may not be ideal, but its certainly not a slouch. Unfortunately, due to the way GCN is designed and how AMD doesn't have the resources to create a multithreaded DX11 driver, oftentimes my core that is doing all of the heavy rendering also has to sort through all of the overhead, which hurts FPS. I heard Polaris was supposed to fix that issue on the hardware side but knowing AMD they probably won't bother trying to improve their DX11 drivers significantly anymore.

[Prysin]

AMD went balls deep with DX12 back with mantle. They wont be spending too much time with their drivers anytime soon.

That being said, in September 2015, they released a patch that increased their total drawcalls by 400k, Nvidia is still ahead by quite a margin, but you know they did something right when every AMD card there is got almost flat 7 FPS boost in almost every title there was....

[App4that]

9 minutes ago, Prysin said:

AMD went balls deep with DX12 back with mantle. They wont be spending too much time with their drivers anytime soon.

That being said, in September 2015, they released a patch that increased their total drawcalls by 400k, Nvidia is still ahead by quite a margin, but you know they did something right when every AMD card there is got almost flat 7 FPS boost in almost every title there was....

Funny, I saw no such change. Though to be fair at the time I was knocking the bottom out of every game I had, so it would have been tough to see

[Prysin]

5 minutes ago, App4that said:

Funny, I saw no such change. Though to be fair at the time I was knocking the bottom out of every game I had, so it would have been tough to see

that is because you are unhappy of your previous product and does not wish upon it to fare well. a fairly normal behavior on those that feel wronged

[App4that]

Just now, Prysin said:

that is because you are unhappy of your previous product and does not wish upon it to fare well. a fairly normal behavior on those that feel wronged

*sigh*

Please don't tell me how I feel. I'm pretty good at interpreting my own emotional state.

 

As I've seen some results of AMD+Nvidia test with DX12, I fully plan on picking up a 290x once everyone loses their minds on the new cards.

[Prysin]

12 minutes ago, App4that said:

*sigh*

Please don't tell me how I feel. I'm pretty good at interpreting my own emotional state.

 

As I've seen some results of AMD+Nvidia test with DX12, I fully plan on picking up a 290x once everyone loses their minds on the new cards.

And with that, i say goodnight to you

[Energycore]

Is this happening because Nvidia cards handle PhysX workloads on their own while AMD's solutions relegate that to the CPU?

[Majestic]

11 minutes ago, Energycore said:

Is this happening because Nvidia cards handle PhysX workloads on their own while AMD's solutions relegate that to the CPU?

No, it's indifferent of gameworks features. Look at the driver overhead tests from futuremark or ashes of the singularity. It's intrinsic.

It's an inefficiency to their drivers.

[Nena Trinity]

33 minutes ago, Majestic said:

No, it's indifferent of gameworks features. Look at the driver overhead tests from futuremark or ashes of the singularity. It's intrinsic.

It's an inefficiency to their drivers.

Will they make some more DX12 tests soon? (FX-6300 is a great combo with R9 380 but that depends on the game if you ask me in some games with GTX 960 will be better...)

[Festive]

Can someone come and help me create a search-party for @App4that? He's been missing too long and I cry.

 

This thread was a spicy one, the AMD fanboys were in full terror mode. But I guess this does clear a few things up @Majestic so thank you for this thread.

[Kinda Bottlenecked]

4 minutes ago, Festive said:

Can someone come and help me create a search-party for @App4that? He's been missing too long and I cry.

 

This thread was a spicy one, the AMD fanboys were in full terror mode. But I guess this does clear a few things up @Majestic so thank you for this thread.

 

Won't deny that I miss that dude. No one else could come up with a better argument.  

[DXMember]

so what's the conclusion?

is there overhead or is thre not?

[don_svetlio]

2 minutes ago, DXMember said:

so what's the conclusion?

is there overhead or is thre not?

There is but it's not nearly as bad as people make it out to be

[Majestic]

30 minutes ago, don_svetlio said:

There is but it's not nearly as bad as people make it out to be

 

At some points the difference is 20 vs.50 fps. That's some nasty overhead on their DX11 drivers. You don't see that difference with Nvidia. Nor do they drop to 20fps in taxing scenes.

[DXMember]

2 hours ago, Majestic said:

 

At some points the difference is 20 vs.50 fps. That's some nasty overhead on their DX11 drivers. You don't see that difference with Nvidia. Nor do they drop to 20fps in taxing scenes.

you can't compare it on different API's

where did you learn science? who taught you experiment integrity?

you can't make such conclusions without hooking into AMD drivers, DX11 library and Windows WDDM to measure the times and no currently publicly available tool offers that and neither has anyone done such tests that are public

[Majestic]

36 minutes ago, DXMember said:

you can't compare it on different API's

where did you learn science? who taught you experiment integrity?

you can't make such conclusions without hooking into AMD drivers, DX11 library and Windows WDDM to measure the times and no currently publicly available tool offers that and neither has anyone done such tests that are public

Who taught you that buzzwords alone make an argument? And the only thing that changes is the GPU's. And it's the same for lower-end and higher-end GPU's aswell.

So the only factor is AMD's drivers. And the fact they can't seem to multithread on DX11.

[DXMember]

36 minutes ago, Majestic said:

Who taught you that buzzwords alone make an argument? And the only thing that changes is the GPU's. And it's the same for lower-end and higher-end GPU's aswell.

So the only factor is AMD's drivers. And the fact they can't seem to multithread on DX11.

I'm not saying there's no "overhead", I'm saying you can't measure it by comparing different APIs as one example - DX12 in AotS utilizes ASync

even more so you can't compare nVidia to AMD in AotS as they have completely different code paths, for one nVidia doesn't use ASync in DX12, both hardware vendors have different application specific shader code optimizations in their driver - they're not even doing the same workload - so you can't compare them on "driver overhead" at all

all you can do is give both GPUs a game and measure the framerate, if AotS with DX12 does have specific timers to measure "driver overhead" and "CPU bound frames" - that's great, but DX11 doesn't

 

P.S. "buzzwords" - what?

[lipefuark]

On 5/3/2016 at 9:59 AM, Majestic said:

Who taught you that buzzwords alone make an argument? And the only thing that changes is the GPU's. And it's the same for lower-end and higher-end GPU's aswell.

So the only factor is AMD's drivers. And the fact they can't seem to multithread on DX11.

 

Brother, do you know if there is any benchmarks comparison between 1070 and RX 480, i know 1070 is way expensive and better, but i'm concerd about a lot of people with older configs having reports regarding LOW FPS with 1070...And i live in Brazil and here 1070 is 2250 BRLS, rx 480 reference is 1450 (because amd cards are overall sightly higher price here), i have 2000 brls already, and i just want to know which one have less CPU OVERHEAD between 1070 , also future 1060 and AIB rx 480...

 

Because i just have a i5 4690k that CAN'T be OC because of my h97m-d3h mb, so that's the main reason for this concern

 

Thank you bro

[Prysin]

46 minutes ago, lipefuark said:

 

Brother, do you know if there is any benchmarks comparison between 1070 and RX 480, i know 1070 is way expensive and better, but i'm concerd about a lot of people with older configs having reports regarding LOW FPS with 1070...And i live in Brazil and here 1070 is 2250 BRLS, rx 480 reference is 1450 (because amd cards are overall sightly higher price here), i have 2000 brls already, and i just want to know which one have less CPU OVERHEAD between 1070 , also future 1060 and AIB rx 480...

 

Because i just have a i5 4690k that CAN'T be OC because of my h97m-d3h mb, so that's the main reason for this concern

 

Thank you bro

the 1070 should have less driver overhead in DX11. However in DX12, Nvidia has a tiny bit MORE overhead then AMD.

AMD seems to focus more on DX12 then DX11 driver efficiency. Now, the overhead for Nvidia in DX12 is based on HARDWARE and DRIVERS.

 

Kepler, Maxwell, and to a much less degree, Pascal (Pascal is just modded maxwell), is not built to handle Async Compute very well. Pascal improves the worst problems, but not remotely close to all of them. 

 

Whilst the 1070 is a good purchase, it has yet to show its achillies heel, as we are missing a large enough sample size of DX12 to know whether Pascal will have issues with DX12 or not.

 

As for the RX 480.

It is NOT comparable to the 1070. They are not in teh same price bracket and should as thus NEVER be compared.

 

The 1070 is a better "matchup" for a R9 Fury or R9 Fury X performance wise.