Why you should not buy AMD GPUs for DX11 games

[Monarch]

Although I can agree AMD drivers are problematic and need improvements compared to Nvidia's, you cherry picked your benchmarks and an AMD card can and will perform well in DX11 games.

 

http://www.guru3d.com/articles_pages/gta_v_pc_graphics_performance_review,5.html

 

I didn't cherry pick the benchmarks. I picked those that show how CPUs perform with both NVIDIA and AMD. It's hard to find such benhcmarks, few tech sites do tests like that. I can tell you that results they got with a 4670K + 290X are true, because I get the same performance on my PC.

Guru3d tested there only GPUs, but I don't get how they managed to get the same performance wtih 290X and 980, something's not right there. And 60+ fps for mid-range cards at 1920x1200? The heck? That's different than all other benchmarks I've seen.

 

AMD cards do not perform well in DX11 because of AMD's driver. Even Anandtech said their DX11 driver is poor (check OP).

[CoolaxGaming]

Who's comparing 780ti and 290x?

And do you know why the 970 is faster at 1080p? Because it is being utilized better, unlike 290X due to AMD's inefficient driver. At high res it's not about how much fps CPU pushes, but how fast GPUs can render pixels and since 290X isn't restricted in that department, it shows it's true potential.

 

 

 

You missed the point. 4770K + 290X produce only 54 fps average compared to 4770K + 970 which produce 72 not because the 290X is slower than the 970, but because with AMD driver 4770K bottlenecks the 290X. If 4770K + 290X produce 54, that means the 290X is capable of 54 fps for sure, right? I'm claiming it can do more, but let's say it can do 54 max.

Now let's look at FX8350 + 290X and FX8350 + 290X. With 290X it produces 35 fps average, while with 970 it produces 44. So why didn't it produce 54, when it's clear that the 290X can do 54 avg with 4770K? Because it's bottlenecked. Now, 970 is bottlenecked too, since we saw that 4770K can push it to produce 72, but clearly the 290X is bottlenecked more by the same CPU. 

 

The point is that every CPU yields better performance with NVIDIA than with AMD.

 

 

 

You don't have to control all the variables. With DX11, my fps was always around 60 average, often dropping below that, down to even 40. I switch to Mantle and viola, 60+ fps. The difference is huge. And Nvidia's DX11 comes very close to Mantle in terms of performance, I don't know how they did it but they did a good job there.

Huh funny then... At max settings with DX 11 on a 8350, I hover around 70fps bf4 MAX settings

[Monarch]

Witcher 3 recommended System Requirements:

 

Intel CPU Core i7 3770 3,4 GHz

AMD CPU AMD FX-8350 4 GHz

Nvidia GPU GeForce GTX 770

AMD GPU Radeon R9 290

RAM 8GB

OS 64-bit Windows 7 or 64-bit Windows 8 (8.1)

DirectX 11

HDD Space 40 GB

 

Anyone wondered why they made it as if 770 and 290 offer the same performance? No, it's not Nvidia or Gameworks, it's AMD's DX11, so blame AMD.

[Kinda Bottlenecked]

 

Witcher 3 recommended System Requirements:

 

Intel CPU Core i7 3770 3,4 GHz

AMD CPU AMD FX-8350 4 GHz

Nvidia GPU GeForce GTX 770

AMD GPU Radeon R9 290

RAM 8GB

OS 64-bit Windows 7 or 64-bit Windows 8 (8.1)

DirectX 11

HDD Space 40 GB

 

Anyone wondered why they made it as if 770 and 290 offer the same performance? No, it's not Nvidia or Gameworks, it's AMD's DX11, so blame AMD.

 

 

I wouldn't comment so much on a game that hasn't even come out yet if I were you... Not a strong argument point.   

[Monarch]

I wouldn't comment so much on a game that hasn't even come out yet if I were you... Not a strong argument point.   

Okay, Far Cry 4 is out.

 

FAR CRY 4 RECOMMENDED: 

OS: Windows® 7 (SP1) / Windows® 8 / Windows® 8.1 / (64-bit only) 

Processor: 2.5 GHz Intel® Core™ i5-2400S or 4.0 GHz AMD FX-8350 or better 

Memory: 8 GB RAM 

Graphics: NVIDIA GeForce GTX 680 or AMD Radeon R9 290X or better (2 GB VRAM) 

DirectX: Version 11 

Network: Broadband Internet connection 

Hard Drive: 30 GB available space 

Sound Card: DirectX-compatible (5.1 surround sound recommended) 

Additional Notes: Supported video cards at the time of release: NVIDIA GeForce GTX 460 or better, GeForce GTX 700 series; AMD Radeon HD5850 or better, Radeon R9 series. Note: Laptop versions of these cards may work but are NOT officially supported.

 

Same thing.

[Footy]

 

Most people think AMD's drivers are horrible. On this, and some other forums, it is generally thought that it is a myth. But is it? I've heard many complaints from people with AMD graphics cards, especially now with GTA V. Now, this is what I've experienced first hand, since I own a 290X.

 

BIG SNIP

 

OP, there is a lot of conflicted views in this thread however,

I just wanted to say thank you for taking the time to compile a post with so much detail it makes interesting reading.

 

Footy

[Monarch]

 

From 1:38 to the end, which is a CPU-bound situation, an i3 yields more fps with a 750ti than with a 280. It is not because the 280 is slower than the 750ti, since we can see that a 4790k is able to fully feed the 280 and produce much more fps. So this is yet another proof that AMD's DX11 driver is less efficient than Nvidia's in terms of CPU usage.

[Briggsy]

One possible explanation for AMD being less driver efficient (it's just a half-baked theory pulled from my rear-end), is that AMD GCN cores are smaller, and there are generally more of them. Nvidia Cuda cores are bigger, run at a much higher frequency, and there tends to be less of them. This means that the CPU might be doing more parallel work with AMD cards, dividing the work up into smaller chunks across a larger spread of shader cores/modules. In truth its a directx problem more than its an AMD problem, which DX12 should fix for CPU core scaling.

[Monarch]

One possible explanation for AMD being less driver efficient (it's just a half-baked theory pulled from my rear-end), is that AMD GCN cores are smaller, and there are generally more of them. Nvidia Cuda cores are bigger, run at a much higher frequency, and there tends to be less of them. This means that the CPU might be doing more parallel work with AMD cards, dividing the work up into smaller chunks across a larger spread of shader cores/modules. In truth its a directx problem more than its an AMD problem, which DX12 should fix for CPU core scaling.

 

Nvidia's driver wasn't this efficient. They optimized it in 337.50.

 

 

It was their response to Mantle. Obviously they exaggerated percentage improvements, but still the improvement is great. 

http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-33750-Driver-Analysis-Single-GPU-and-SLI-Tested/Single-GPU-Testing-and

[SleathX1]

 

From 1:38 to the end, which is a CPU-bound situation, an i3 yields more fps with a 750ti than with a 280.

Holy shit m8 you're right.

[sgloux3470]

I ran Battlefield 4 with a 270x CF rig (which is about as powerful as a single 290x) at 1080p/ultra DX11 with an average framerate of 88 and a max framerate of 120 (had it capped)  GPU usage was in the 85-95% range.  

 

https://drive.google.com/file/d/0B4HpnFBkAjlIX3ZCa2lVT2pUekU/view?usp=sharing

 

That was with an 8350 (iirc it was either stock or around 4.4GHz.)

 

So uh, no.

 

 

 

 

From 1:38 to the end, which is a CPU-bound situation, an i3 yields more fps with a 750ti than with a 280. It is not because the 280 is slower than the 750ti, since we can see that a 4790k is able to fully feed the 280 and produce much more fps. So this is yet another proof that AMD's DX11 driver is less efficient than Nvidia's in terms of CPU usage.

 

It's a CPU bottleneck, of course they're going to get similar FPS.  The video card is irrelevant.  Nvidia drivers do have slightly less CPU overhead, but it's only relevant in heavily bottlenecked scenarios like with an i3.  You'd get the same framerates with a i3 paired with a 980 and an i3 paired with a 270.  

[niofalpha]

-Snip-

It's all a conspiracy to get you to buy AMD CPU's.

[Monarch]

I just ran the API overhead test with 290X and then 970 in the same system and wow what a difference.

 

R9 290X: 674 796, GTX 970: 1 273 203. Almost 100% difference. 15.4 beta and 350.12. 

 

 

 

[AsGoodAsAHacker]

Okay, Far Cry 4 is out.

FAR CRY 4 RECOMMENDED:

OS: Windows® 7 (SP1) / Windows® 8 / Windows® 8.1 / (64-bit only)

Processor: 2.5 GHz Intel® Core™ i5-2400S or 4.0 GHz AMD FX-8350 or better

Memory: 8 GB RAM

Graphics: NVIDIA GeForce GTX 680 or AMD Radeon R9 290X or better (2 GB VRAM)

DirectX: Version 11

Network: Broadband Internet connection

Hard Drive: 30 GB available space

Sound Card: DirectX-compatible (5.1 surround sound recommended)

Additional Notes: Supported video cards at the time of release: NVIDIA GeForce GTX 460 or better, GeForce GTX 700 series; AMD Radeon HD5850 or better, Radeon R9 series. Note: Laptop versions of these cards may work but are NOT officially supported.

Same thing.

I like to think that games like that are having special relationships with card producers. No reason why nvidia performs sooooo much better when we see games like tomb raider that perform amazingly on both. I bet there are some shady deals going on occasionally with card producers and games.

[Monarch]

I like to think that games like that are having special relationships with card producers. No reason why nvidia performs sooooo much better when we see games like tomb raider that perform amazingly on both. I bet there are some shady deals going on occasionally with card producers and games.

 

No. Take a look at benchmarks in OP. BF4 is a Gaming Evolved title. And yet AMD GPUs perform worse with DX11. There's a post of mine right above yours done in 3DMark that shows that my CPU is able to make twice more draw calls with Nvidia than with AMD's driver. Tomb Raider is not a CPU-bound game and that's why this issue is not evident there, unless you're aiming for high framerates.

[xAcid9]

 

I just ran the API overhead test with 290X and then 970 in the same system and wow what a difference.

 

R9 290X: 674 796, GTX 970: 1 273 203. Almost 100% difference. 15.4 beta and 350.12. 

 

Something wrong with your PC? Both scores is pretty low.

[Monarch]

Something wrong with your PC? Both scores is pretty low.

 

Haven't seen any other scores with i5's so I can't compare with mine, but you can't expect too much from an i5. 

[WaitForMeImOnDialUp]

I like to think that games like that are having special relationships with card producers. No reason why nvidia performs sooooo much better when we see games like tomb raider that perform amazingly on both. I bet there are some shady deals going on occasionally with card producers and games.

nvidea is known to have game makers favor thier stuff over the competitor..just like with intel and cinebench.

[xAcid9]

Haven't seen any other scores with i5's so I can't compare with mine, but you can't expect too much from an i5. 

This is mine with Hyper Threading disabled@3.5Ghz

 

Your proc should be faster compare to mine since it's overclocked. 

[Kinda Bottlenecked]

I just ran the API overhead test with 290X and then 970 in the same system and wow what a difference.

 

R9 290X: 674 796, GTX 970: 1 273 203. Almost 100% difference. 15.4 beta and 350.12. 

 

The only thing that caught my eye is the 10 million draw calls from mantle. Wow.

 

Edit: Ran it on my 280x

 

Your 290x scores seem way off...

[Monarch]

This is mine with Hyper Threading disabled@3.5Ghz

 

-snip-

 

Your proc should be faster compare to mine since it's overclocked. 

 

Apparently the Xeon is faster than the i5. Even with HT disabled. If you look at Pohernori's score he got about the same score as I did with the same CPU in DX11. 3 million more with Mantle though, for some reason. I might put back the 290X and try again.

 

 

The only thing that caught my eye is the 10 million draw calls from mantle. Wow.

 

Edit: Ran it on my 280x

 

 

-snip-

 

Your 290x scores seem way off...

 

Keep in mind that we're testing API overhead for the CPU. It's not graphically intensive and is run at 720p, so GPUs will be far from fully utilized. It doesn't matter if you have 270X or 290X, the score depends on how much draw calls the CPU can make before it hits 30 fps.

 

Edit:

I ran the test a couple more times with the 970 and sometimes I got 1 mil 230k, sometimes 1 mil 300k, so there's definitely some variance, not sure why.

[Kinda Bottlenecked]

Apparently the Xeon is faster than the i5. Even with HT disabled. If you look at Pokehori's score he got about the same score as I did with the same CPU.

 

 

 

Keep in mind that this is test of API overhead on the CPU. It's not graphically intensive and is run at 720p, so GPUs will be far from fully utilized. It doesn't matter if you have 270X or 290X, the score depends on how much draw calls can the CPU make before it hits 30 fps.

 

I ran the test a couple more times with the 970 and sometimes I get 1 mil 230k, sometimes 1 mil 300k, so there's definitely some variance, not sure why.

 

That should explain the DX11 scores. But why are my mantle scores higher than yours?

[Monarch]

That should explain the DX11 scores. But why are my mantle scores higher than yours?

 

Dunno. I'll put the 290X back in and test again.

[Monarch]

That should explain the DX11 scores. But why are my mantle scores higher than yours?

 

Just ran another test with the 290X:

 

 

I don't get why your Mantle score is that high. Our DX11 scores are the same. What drivers do you have? Is your CPU OCd?

[Kinda Bottlenecked]

Just ran another test with the 290X:

 

I don't get why your Mantle score is that high. Our DX11 scores are the same. What drivers do you have? Is your CPU OCd?

 

To 4GHz. If you were running your cpu at stock, that might explain it. Also 15.4 beta driver.