Thursday, March 26th 2015

Futuremark Releases 3DMark Update with API Overhead Feature-set

Futuremark is excited to introduce our new 3DMark API Overhead feature test - the world's first independent test for measuring differences in DirectX 12, DirectX 11 and Mantle API performance. It's also the very first public application to use DirectX 12 full stop. This is cutting edge stuff! Developed with input from AMD, Intel, Microsoft, NVIDIA, and the other members of our Benchmark Development Program, the 3DMark API Overhead feature test lets you compare the performance of DirectX 12, DirectX 11, and Mantle.

The purpose of the test is to compare the relative performance of different APIs on a single system, rather than the absolute performance of different systems. The API Overhead feature test is not a general-purpose GPU benchmark, and it should not be used to compare graphics cards from different vendors. (We are working on a DirectX 12 benchmark with game-like workloads, which we expect to release soon after the public launch of Windows 10.)

DOWNLOAD: Futuremark 3DMark v1.5.884

Why is API overhead important?
There has been much talk this year of new graphics APIs that let developers code 'close to the metal.' And though it's natural to think of the GPU, the benefits of APIs with lower overhead are actually achieved by making better use of multi-core CPUs to streamline code execution and eliminate software bottlenecks, particularly for draw calls.

A draw call happens when the CPU tells the GPU to draw an object on the screen. Games typically make thousands of draw calls per frame, but each one creates performance-limiting overhead for the CPU.

As the number of draw calls rises, graphics engines become limited by API overhead. New APIs like DirectX 12 and Mantle reduce that overhead allowing more draw calls. With more draw calls, a game engine can draw more objects, textures and effects to the screen.

How does the 3DMark API Overhead feature test work?
The 3DMark API Overhead feature test measures API performance by making a steadily increasing number of draw calls. The result of the test is the maximum number of draw calls per second achieved by each API before the frame rate drops below 30 fps.

What are the system requirements?
  • The DirectX 12 test requires a PC running an up-to-date version of Windows 10 Technical Preview (build 10041 or later), 4 GB of system memory, and DirectX feature level 11_0 compatible hardware with at least at least 1 GB of graphics memory.
  • The Mantle test requires 4 GB of system memory and AMD hardware that supports the Mantle API.
  • The DirectX 11 test requires DirectX feature level 11_0 compatible hardware with at least 1 GB of graphics memory and 4 GB of system memory.
Release notes:
3DMark Windows v1.5.884
March 26, 2015

This major update adds the API Overhead feature test, the world's first independent test for comparing the performance of DirectX 12, Mantle, and DirectX 11. See how many draw calls your PC can handle with each API before the frame rate drops below 30 fps.
  • Compare DirectX 12, DirectX 11 and Mantle with the new API Overhead Feature Test, available in 3DMark Advanced Edition and 3DMark Professional Edition.
  • Added Feature Test selection screen.
  • Improved
  • Improved formatting of larger scores to make them more readable.
  • Result screen automatically shows FPS after running a single test.
Fixed
  • Fixed a bug that could cause the Sky Diver demo to hang at the cave entrance scene.
Add your own comment

46 Comments on Futuremark Releases 3DMark Update with API Overhead Feature-set

#26
Caring1
MxPhenom 216So DX12 is better then Mantle on all but 8 thread tests. Interesting.
That's subjective, it depends on what you think is better.
DX 12 has higher figures in draw call, but Mantle pulls close to the same numbers but in less than half the time.
Posted on Reply
#27
arbiter
ISI300So can I run the DX12 test on my 5870? I'm pretty sure it supports DX11 FL 11_0.
Its said that ALL DX11 cards should support the lower level part of DX12, but given age of that card don't be to surprised if that isn't the case.
Posted on Reply
#28
raptori
For DX12 users can you use it's full features ? like the use different GPUs or use cumulative memory from multi cards ? I mean does the latest version support those features ?.
Posted on Reply
#29
Caring1
Ferrum MasterWell all things considered, you know upon what you are looking at? Death of Windows 7 to 8.1 included...
Only for Gamers.
Posted on Reply
#30
MrGenius
raptoriFor DX12 users can you use it's full features ? like the use different GPUs or use cumulative memory from multi cards ? I mean does the latest version support those features ?.
I don't have multiple GPUs, so I can't verify that those features(if they exist) can be used. But the version of DX12 provided with the 10 tech preview appears to be full/complete. And to take full advantage of it only requires also installing the latest graphics drivers for 8.1, from Nvidia or AMD(or probably Windows Update too, can't confirm that though).
Posted on Reply
#31
N3M3515
Does anyone here knows how to translate draw calls/overhead to fps?
Posted on Reply
#32
Initialised
N3M3515Does anyone here knows how to translate draw calls/overhead to fps?
The test ramps up the draw calls until the system cannot hold 30fps so the draw call/second figure is the number of draw calls at the end of the test at 30fps.

It's broken down in the detailed results below:

www.3dmark.com/3dm/6380242
DX11 Multi-threaded draw calls per second 667667.2
DX11 Single-threaded draw calls per second 696097.3
DX12 draw calls per second 1.13809426E7
Mantle draw calls per second 8571791.8

DX11 Multi-threaded draw calls per frame 24576.0
DX11 Single-threaded draw calls per frame 21504.0
DX12 draw calls per frame 380928.0
Mantle draw calls per frame 294912.0

DX11 Multi-threaded final fps 27.17 fps
DX11 Single-threaded final fps 32.37 fps
DX12 final fps 29.88 fps
Mantle final fps29.07 fps

In the DX 12 run the final frame used 380928.0 draw calls and was running at 29.88 fps and made 11382128.64 draw calls to make it.

29.88 = 11382128.64/380928.0

Showing the working to make sure the units tally:
frames/second = (Draw calls/second)/(Draw calls/frame) = (1/second)/(1/frame) = frames/second
Posted on Reply
#33
R-T-B
MxPhenom 216Heres my results for just the DX11 test, on my 4770k at 4.4ghz

Considering that btarunr's system and yours have more or less equal cost segment products from AMD and NVIDIA, it's surprising NVIDIA is able to squeeze out that much more DX11 draw calls.
MxPhenom 216Not really sure either.
High level API abstraction became popular basically because software developes asked for it. It was basically developer laziness that spawned it and it has existed as long as DirectX and OpenGL have. Before these standards existed, people generally coded in DOS on the CPU and GPUs were seen as such GREAT improvements over software rendering that they could get away with high level abstraction like that.

And they could. Until AMD pointed out to gamers what was missing with Mantle. You don't know what you're missing until it's shoved in your face.

But still, DX11 API performance on AMD cards seems pisspoor. Why?
MrGeniusI don't have multiple GPUs, so I can't verify that those features(if they exist) can be used. But the version of DX12 provided with the 10 tech preview appears to be full/complete.
No. It doesn't even have FL_12 yet. Just WDDM 2.0. And no, none of the above stuff is supported yet.
Posted on Reply
#34
MxPhenom 216
ASIC Engineer
@R-T-B its because NVIDIA was able reduce overhead through drivers fairly well since their big 345.xx release or whatever.
Posted on Reply
#35
R-T-B
MxPhenom 216@R-T-B its because NVIDIA was able reduce overhead through drivers fairly well since their big 345.xx release or whatever.
I'm aware of that. I guess I'm just flabberghasted at how much they squeezed out of supposedly "flawed" DX11. AMD needs to follow suit, but given the direction of things I doubt they will.
Posted on Reply
#36
MxPhenom 216
ASIC Engineer
R-T-BI'm aware of that. I guess I'm just flabberghasted at how much they squeezed out of supposedly "flawed" DX11. AMD needs to follow suit, but given the direction of things I doubt they will.
I don't think they know how..............i kid i kid har har har
Posted on Reply
#37
R-T-B
MxPhenom 216I don't think they know how..............i kid i kid har har har
Honestly, given their condition I don't think they have the R&D bugdet, which makes your statement more true than I'd like...
Posted on Reply
#38
Ferrum Master
R-T-BJust WDDM 2.0. And no, none of the above stuff is supported yet.
I may be wrong... but IMHO drivers here is the last importance... The windows kernel and the function layers is the most important ones ensuring the thing works... as DX12 talks directly the the kernel has more possibilities to ensure the draw call specific functions execute, as it is their own creation, AMD with mantle cannot alter that, it simply must be more efficient.
Posted on Reply
#39
R-T-B
Ferrum MasterI may be wrong... but IMHO drivers here is the last importance... The windows kernel and the function layers is the most important ones ensuring the thing works... as DX12 talks directly the the kernel has more possibilities to ensure the draw call specific functions execute, as it is their own creation, AMD with mantle cannot alter that, it simply must be more efficient.
No you aren't wrong, you are very very right. While the "trim" (basically extended features) on DX12 isn't done/defined yet, the core part of it, WDDM 2.0 support in the kernel, is done and works quite well.
Posted on Reply
#40
Mussels
Freshwater Moderator
absolutely effing fascinating, THIS makes me happy - something worthwhile in benchmarks and something that could really show differences between major driver releases.
Posted on Reply
#41
Mussels
Freshwater Moderator
The results for this back up what a thread on TPU a week or two ago was speculating at


the thread was about how nvidia hardware seems to work better on weaker CPU's, but it was close - nvidia drivers are just better at multi threading in DX11
Posted on Reply
#42
R-T-B
MusselsThe results for this back up what a thread on TPU a week or two ago was speculating at


the thread was about how nvidia hardware seems to work better on weaker CPU's, but it was close - nvidia drivers are just better at multi threading in DX11
They seem massively better at DX11 draw calls in general.
Posted on Reply
#43
Prima.Vera
Are the older cards like 780/ 780Ti going to support this on Win10?
Posted on Reply
#44
R-T-B
Prima.VeraAre the older cards like 780/ 780Ti going to support this on Win10?
Pretty much anything DX11 should support the basics of DX12, and thus this test. So yeah, Fermi forward.
Posted on Reply
#45
Mussels
Freshwater Moderator
can someone do a test for me with this and an AMD card, and see what differences there are between 14.12 and newer drivers, especially the 15.3 beta (not the 15.3.1)


I've noticed a lot smoother DX11 gameplay on these drivers and i'm curious if the draw calls are why
Posted on Reply
Add your own comment
Oct 6th, 2024 04:14 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts