Quick test: Futuremark 3DMark v2.3.3663 Vulkan API Overhead Benchmarks

#5411427

Corbus

2017-03-23 15:49

Thanks for this Hilbert!

#5411430

Agonist

2017-03-23 15:51

For some reason, without touching anything, it says cancelled by user after DX11 ST runs.

#5411437

Kaarme

2017-03-23 16:02

I guess AMD forgot to send the technology consultation payment to Futuremark.

#5411449

PrMinisterGR

2017-03-23 16:30

Futuremark just released 3DMark v2.3.3663 with support for Vulkan in their API overhead test, we ran some quick tests to see what is happeing.... Quick test: Futuremark 3DMark v2.3.3663 Vulkan API Overhead Benchmarks

Thanks for the tests man. Any chance to put dots or commas to indicate thousands/millions etc? It's a bit hard to read.

#5411453

PrMinisterGR

2017-03-23 16:34

I guess AMD forgot to send the technology consultation payment to Futuremark.

No. It's more like GCN is less efficient, although it's being corrected more or less. The only "modern" AMD card on this chart is the 480 and it's faster than its equivalent (the 1060), in modern APIs. Under DX11 you can see that NVIDIA is still almost +80% more draw call efficient, but AMD was starting from something like 600k draw calls and we're currently hovering at around 1.3 million. Still a long way to go, and I'm not sure that it can be done. A friend with an untouched 6700k and 3200 RAM is telling me he's getting 2.2 million draw calls for his 280x (he had to get a CPU and he's still waiting to check Vega or drops to the 1080Ti price). EDIT: NVIDIA's Vulkan performance is impressive, their DX12 one not as much. Since both APIs are quite similar, it's probably an indication of driver states.

#5411461

Denial

2017-03-23 16:50

After a certain point I doubt the numbers even matter. Most games are operating under the draw call limits of DX11 - there is no way any are even close to the limits of 12/Vulkan or will be in the foreseeable future.

#5411473

Denial

2017-03-23 17:07

I dont trust this Futuremark 3DMark AMD is getting less performance soothing is going on. There is a history with this benchmark program that favors nvidia but they deny is but this kind of proves they were lying. No smoke without fire.

How does this prove they are lying? ... because AMD is slightly lower? That's not even to mention that both Sniper 4/GoW4 both showed similar results on Nvidia hardware to TimeSpy. No one ever writes articles about that though - guess it isn't news worthy that TimeSpy was actually accurate. Or to mention that the same test shows AMD's DX12 performance higher than Nvidia's equivalent. Why would they do that if they are so purposely anti-AMD? Should also point out that there are now two people in this thread suggesting that there is foul play against AMD - which is becoming increasingly common. It's like impossible that AMD just has a few kinks it needs to work out - must be foul play. Literally everyone in the entire world is colluding against AMD - Microsoft, 3DMark, Intel, everyone. There simply cannot be any other explanation, right?

#5411543

Anarion

2017-03-23 19:27

After a certain point I doubt the numbers even matter. Most games are operating under the draw call limits of DX11 - there is no way any are even close to the limits of 12/Vulkan or will be in the foreseeable future.

Yep. With Vulkan and DX12 draw calls shouldn't be thing that causes bottleneck.

#5411559

Denial

2017-03-23 19:58

You are talking about dx12 gameworks games so kind of shot yourself in the foot right there. Both games you speak of are dx12 not vulkan. https://blogs.nvidia.com/blog/2016/10/07/dx12-gears-of-war-4/ You not helping your case at all in fact your supporting my claim. Something is going on here because no way should vulkan be slower than dx12 ...

Just because results favor your hardware shouldn't make you instantly support such results because you are only lying to oneself then. Te hole point of these programs are to compare performance. If vulkan which is mantle re-branded is getting less performance than dx12 i say they have earned their wages this year.

Vulkan isn't Mantle rebranded. It's heavily based on Mantle, but it's not identical. You posted a comment saying that 3DMark has favored Nvidia hardware purposely. That isn't true, period. In fact the article you posted (which got removed) concludes that it doesn't favor Nvidia so I'm not even sure why you posted it. Further proof is that there are now multiple games, one Nvidia sponsored - fine.. and Sniper 4, which isn't, that both run faster on DX12 vs 11 with Async Compute enabled - which matches the TimeSpy results. And obviously there is something going on.. my issue is that you immediately make the assumption that it's foul play on 3DMark's part. It couldn't possibly be a bug in AMD's driver, right? It couldn't possibly be a bug in 3DMarks software? It couldn't possibly be because AMD's Vulkan implementation is slightly worse? It couldn't be because Nvidia's implementation is better? There are like 500,000,000 reasons for these results and you automatically assume it's foul play. If it was a one off case, I wouldn't care.. but every time AMD is slightly behind in something it's the immediate assumption by multiple users in this forum. It's getting really tiresome.

#5411563

edilsonj

2017-03-23 20:10

Is in 3D Mark's Vulkan better on GTX 1060 than RX 480? Well... can Futuremark fix Vulkan games on my GTX 1060 to make this real?

#5411607

PrMinisterGR

2017-03-23 22:25

Might i ask as to why an intel 5960 was benched against a Ryzen 1700? these are chips in completely different price brackets.

Both octocores. It was an obvious IPC comparison.

#5411639

Agent-A01

2017-03-23 23:25

Might i ask as to why an intel 5960 was benched against a Ryzen 1700? these are chips in completely different price brackets.

When have price brackets ever mattered when comparing AMD to any competitive hardware?

#5411660

mtrai

2017-03-24 00:09

These results are the amount of draw calls PER SECOND, not the maximum amount of draw calls handled. A higher number can be interpretated to an improved frametime, as less time is required for a fixed amount of draw calls to be addressed. Btw, there's something wrong with DX11 results for all AMD gpus, they should be around 1.8-2.3M for both DX11 ST and MT.

You mean this? http://i.imgur.com/x9yLEWX.jpg

#5411667

mtrai

2017-03-24 00:30

For kicks I just thought of something with how my system is set up and thought I would test it out. PC Specs: I5 6600 OC 4.9 core/4.8 Unicore 16 GB ram at 3200 ASUS Z170-A GPUs: RX 480 8 GB Red Devil and HD 7770 with just bios flashed to R7 250X Since we are looking at a lot of numbers and comparing them across manufactures and gpus etc...I thought I would run the API on just the HD 7770 GPU and see the numbers...I disabled the RX 480 to force it 3dmark to use the HD 7770. Very Interesting DX 11 results. http://www.3dmark.com/3dm/18805193 Comparison to my previous run this morning with the RX 480. Now I am very eagerly awaiting explanations on especially on the DX 11. I am not being a smart you know what. I am being serious. http://www.3dmark.com/compare/aot/199856/aot/198443 THIS IS THE HD 7770 API just a few minutes ago. http://i.imgur.com/DPLPkDV.jpg THIS IS THE SCREENSHOT for the RX 480 this morning. [img]http://i.imgur.com/x9yLEWX.jpg[/IMG]

#5411672

chumanga1

2017-03-24 00:46

Does anyone with i7 and NVIDIA card already noticed major difference with HT enabled vs disabled for DX11 multithreading in this test? This happens to me since first time 3Dmark overhead come out, take a look: Results from today with HT disabled: DirectX 11 Multi-threaded draw calls per second: 2 707 663 DirectX 11 Single-threaded draw calls per second: 2 286 665 Results with HT enabled: DirectX 11 Multi-threaded draw calls per second: 2 233 946 DirectX 11 Single-threaded draw calls per second: 2 346 820 It's always like that for me, hyperthreading decrease DX11 multi-threading draw calls result.

#5411676

mtrai

2017-03-24 01:00

I am just finding this interesting that with the Vulkan support added we are also noticing the DX11 discrepancies.

#5411708

Srsbsns

2017-03-24 03:30

Why is the Fury X coming in so much behind the RX 480 when the Fury X is a much stronger card?

#5411738

Robbo9999

2017-03-24 07:17

Well thanks 3DMark for breaking your bench and figuring out how to introduce a hardware fault... The computer has rebooted from a bugcheck. The bugcheck was: 0x000000d1 The DRIVER_IRQL_NOT_LESS_OR_EQUAL bug check has a value of 0x000000D1. This indicates that a kernel-mode driver attempted to access pageable memory at a process IRQL that was too high. Futuremark has nothing to do with fixing how an API works in games. Well, finally got it to run...who knows? DirectX 11 single-thread 2 377 030 Draw calls per second DirectX 11 multi-thread 2 454 589 Draw calls per second DirectX 12 33 556 899 Draw calls per second Vulkan 20 729 572 Draw calls per second

May as well post my results too, interesting that your DX12 results are higher than Vulkan, for me it's the other way around, yet we both have Intel/NVidia. DX11 mult:i 4,507,018 DX12: 27,231,630 Vulkan: 29,736,051 Anyone know why our results are swapped around & so different?

#5411742

tensai28

2017-03-24 07:23

Irrelevant since all of the games are dx11 and will be that way for the foreseeable future. That's why I made the switch to nvidia a few years ago.

#5411777

Undying

2017-03-24 09:40

removed