3DMark
3DMark > General Discussions > Topic Details
This topic has been locked
h0rnyfavn Jul 16, 2016 @ 10:11am
3dmark doesn't use proper Asynchronous Compute!
http://www.overclock.net/t/1605674/computerbase-de-doom-vulkan-benchmarked/220#post_25351958

All of the current games supporting Asynchronous Compute make use of parallel execution of compute and graphics tasks. 3D Mark Time Fly support concurrent. It is not the same Asynchronous Compute....

So yeah... 3D Mark does not use the same type of Asynchronous compute found in all of the recent game titles. Instead.. 3D Mark appears to be specifically tailored so as to show nVIDIA GPUs in the best light possible. It makes use of Context Switches (good because Pascal has that improved pre-emption) as well as the Dynamic Load Balancing on Maxwell through the use of concurrent rather than parallel Asynchronous compute tasks. If parallelism was used then we would see Maxwell taking a performance hit under Time Fly as admitted by nVIDIA in their GTX 1080 white paper and as we have seen from AotS.
Last edited by h0rnyfavn; Jul 17, 2016 @ 1:06pm
< >
Showing 1-15 of 513 comments
FM_Jarnis  [developer] Jul 16, 2016 @ 10:23am 
Yes it does.

http://www.futuremark.com/downloads/3DMark_Technical_Guide.pdf

It was not tailored for any specific architecture. It overlaps different rendering passes for asynchronous compute, in paraller when possible. Drivers determine how they process these - multiple paraller queues are filled by the engine.

The reason Maxwell doesn't take a hit is because NVIDIA has explictly disabled async compute in Maxwell drivers. So no matter how much we pile things to the queues, they cannot be set to run asynchronously because the driver says "no, I can't do that". Basically NV driver tells Time Spy to go "async off" for the run on that card. NV driver runs asynchronous tasks in one queue on Maxwell, similar to if they were submitted in one queue ("async off" in Time Spy). If NVIDIA enables Async Compute in the drivers on Maxwell, Time Spy will start using it. Performance gain or loss depends on the hardware & drivers.

Edit: Quoting 3DMark Technical guide

Asynchronous Compute
With DirectX 11, all rendering work is executed in one queue with the driver deciding the order of the tasks.

With DirectX 12, GPUs that support asynchronous compute can process work from multiple queues in parallel.

There are three types of queue: 3D, compute, and copy. A 3D queue executes rendering commands and can also handle other work types. A compute queue can handle compute and copy work. A copy queue only accepts copy operations. The queues all race for the same resources so the overall benefit depends on the workload.

In Time Spy, asynchronous compute is used heavily to overlap rendering passes to maximize GPU utilization. The asynchronous compute workload per frame varies between 10 - 20%.

To observe the benefit on your own hardware, you can optionally choose to disable async compute using the Custom run settings.
Last edited by FM_Jarnis; Jul 18, 2016 @ 7:36am
h0rnyfavn Jul 16, 2016 @ 10:29am 
in paraller when possible

A proper async compute is all about "parallelism" which is not possible on nvidia cars hence why all these tricks. ;)
FM_Jarnis  [developer] Jul 16, 2016 @ 10:37am 
Originally posted by h0rnyfavn:
in paraller when possible

A proper async compute is all about "parallelism" which is not possible on nvidia cars hence why all these tricks. ;)

Yes it is. But Engine cannot dictate what the hardware has available or not.

Async compute is about utilizing "idle" shader units. Slower the card, less idle ones you have. Less capable hardware may also be hard pressed to utilize all of them even if the engine asks nicely. Also there may be limitations as to what workloads in the engine *can* run in parallel. Yes, Time Spy is very graphics-heavy, since, well, its a graphics benchmark. But even there many of the rendering passes have compute tasks that can use this.

Ultimately some AMD cards gain quite a bit (ie. they have a lot of shader units idling while rendering and they are very good at using them for the available paraller loads). Some AMD cards gain less or not at all (either less capable at paralleriziing, less idle shader units or no idle shader units at all - for example a HD 7970 is hard pressed to have any to "spare")

Some NVIDIA cards cannot do this at all. The driver simply says "hold your horses, we'll do this nicely in order". Some NVIDIA cards can do some of it. They might use another way than AMD (more driver/software based), but the end result is the same - the card hardware is capable of doing more through some intelligent juggling of the work.

Last edited by FM_Jarnis; Jul 16, 2016 @ 1:20pm
db_smooth Jul 16, 2016 @ 11:17am 
You are so full of♥♥♥♥♥♥FM_Jarnis, this is being throughly researched by the best Overclockers in the world.

You are using dirty tricks to give favorable results to nvidia cards.
This is your reputation not mine.

This "TEST" is worthless
Last edited by db_smooth; Jul 16, 2016 @ 11:18am
BDK Jul 16, 2016 @ 12:25pm 
Yep, totally useless benchmark.
db_smooth Jul 16, 2016 @ 12:37pm 
My self and others are already knee deep in decompiled code, we are totally confident we will be able to prove 100% that this is a Nvidia Tailored benchmark and is using load balancing and not cocurrency.

This benchmark is completely dishonest and cannot be used to compare GPU's.
Kae Jul 16, 2016 @ 12:58pm 
Originally posted by db_smooth:
My self and others are already knee deep in decompiled code, we are totally confident we will be able to prove 100% that this is a Nvidia Tailored benchmark and is using load balancing and not cocurrency.

This benchmark is completely dishonest and cannot be used to compare GPU's.

This would be amazing if true, and would totally reshape my opinion of Futuremark. I'll be following this with interest.
BDK Jul 16, 2016 @ 1:04pm 
3DMark = Using context switching. Basically very un-optimized and old.
Games = Using parallel. Basically proper multitasking for good performance gains.

AMD can do both.
Nvidia can only do context switching. So this benchmark is completely useless.
DarkStarClassic Jul 16, 2016 @ 1:34pm 
Damn crooks and i bought this damn benchmark. This MAFIA practise bah...well no more money from me ever for this con company and ♥♥♥♥ Nvidia they are the worsed.

If this is all true can they be sued for mal practise as con company(include Nvidia also pls?)

Fix it or i will when ever i can show on forums futuremark are con artist take dirty money from Nvidiamafia company.

Sad real sad if it's true.

AMD Forever!!!
Raptor Jesus Jul 16, 2016 @ 1:40pm 
If true I demand a refund. No catering to one ♥♥♥♥ing company.

Should have stuck to pirating instead of being a good guy and paying.
Last edited by Raptor Jesus; Jul 16, 2016 @ 1:42pm
Prince Thrakkath Jul 16, 2016 @ 1:55pm 
Very disappointed in this, give proper Async support in your benchmark or give me a refund please. The results come nowhere near the reality of game benchmarks for DX12/Vulkan, these scores stink to the high heavens.
Slayix312 Jul 16, 2016 @ 1:58pm 
Doom gets around 10% performance increase with async on vs off.

IO Interactive (Hitman developers) also said they only saw a 5-10% difference with async on vs off.

source:

http://techfrag.com/2016/03/28/hitman-gains-merely-5-10-boost-amd-cards-async-compute/

Guess what performance boost Timespy gets? 9% or so (according to PCPER) for the 480 and 12% for the Fury X.

Guess what this falls in line with? Yep; Doom and Hitman performance increases.

So the performance increase Timespy gives with async compute seems to be the standard. Thus; I see no issue.
Last edited by Slayix312; Jul 16, 2016 @ 2:03pm
db_smooth Jul 16, 2016 @ 2:08pm 
Originally posted by Slayix312:
Doom gets around 10% performance increase with async on vs off.

IO Interactive (Hitman developers) also said they only saw a 5-10% difference with async on vs off.

source:

http://techfrag.com/2016/03/28/hitman-gains-merely-5-10-boost-amd-cards-async-compute/

Guess what performance boost Timespy gets? 9% or so (according to PCPER) for the 480 and 12% for the Fury X.

Guess what this falls in line with? Yep; Doom and Hitman performance increases.

So the performance increase Timespy gives with async compute seems to be the standard. Thus; I see no issue.
dont be a fanboy, the dev himself said this confirming that async is not being used on Nvidia cards.

The reason Maxwell doesn't take a hit is because NVIDIA has explictly disabled async compute in Maxwell drivers
Slayix312 Jul 16, 2016 @ 2:11pm 
Originally posted by db_smooth:
Originally posted by Slayix312:
Doom gets around 10% performance increase with async on vs off.

IO Interactive (Hitman developers) also said they only saw a 5-10% difference with async on vs off.

source:

http://techfrag.com/2016/03/28/hitman-gains-merely-5-10-boost-amd-cards-async-compute/

Guess what performance boost Timespy gets? 9% or so (according to PCPER) for the 480 and 12% for the Fury X.

Guess what this falls in line with? Yep; Doom and Hitman performance increases.

So the performance increase Timespy gives with async compute seems to be the standard. Thus; I see no issue.
dont be a fanboy, the dev himself said this confirming that async is not being used on Nvidia cards.

The reason Maxwell doesn't take a hit is because NVIDIA has explictly disabled async compute in Maxwell drivers

Don't be a fanboy about what? Wtf are you talking about? I am saying that a-sync performance gains on Timespy are directly in line with those from Doom and Hitman. BOTH get between 5-10% increase in performance with A-sync on. This is directly in line with what Timespy increases AMD card performance by. Give or take 1-2%.

No one is even talking about Nvidia cards.
FM_Jarnis  [developer] Jul 16, 2016 @ 2:21pm 
This thread should probably be locked, but then you would just keep repeating the claims in every other thread, so I'll keep it open. Only technical support is around on weekends, so anything further has to wait until at least on Monday.
< >
Showing 1-15 of 513 comments
Per page: 15 30 50

3DMark > General Discussions > Topic Details
Date Posted: Jul 16, 2016 @ 10:11am
Posts: 513