3DMark API OVERHEAD TEST

shankly1985 · 12 Jul 2015 at 15:04

Lets see how we compare now that AMD has improved there Draw Calls for DX11 over the past couple months.

DX11 Multi-threaded draw calls per second 1,045 829
DX11 Single-threaded draw calls per second 1,062 395
DX11 Multi-threaded draw calls per frame 35,328
DX11 Single-threaded draw calls per frame 35,328

i7 3770k @4.2 290 @1030/1250 Driver 15.7 - 1440p
http://www.3dmark.com/aot/42647

DX11 Multi-threaded draw calls per second 1,044 619
DX11 Single-threaded draw calls per second 1,113 866
DX11 Multi-threaded draw calls per frame 36,864
DX11 Single-threaded draw calls per frame 38,400

i7 3770k @4.2 290 @1030/1250 Driver 15.7 - 720p
http://www.3dmark.com/aot/42800

whyscotty · 12 Jul 2015 at 15:06

http://forums.overclockers.co.uk/showthread.php?t=18663220&highlight=API+OVERHEAD

shankly1985 · 12 Jul 2015 at 15:12

whyscotty said:
http://forums.overclockers.co.uk/showthread.php?t=18663220&highlight=API+OVERHEAD

Thats 720p Should have said am running 1440p

whyscotty · 12 Jul 2015 at 15:14

Not it's not, it's just the OP never updated it

shankly1985 · 12 Jul 2015 at 15:16

whyscotty said:
Not it's not, it's just the OP never updated it

My God who cares? just post some results is this going to take 10+ pages?

whyscotty · 12 Jul 2015 at 15:17

Oh dear - some ones knickers are in a twist

bru · 12 Jul 2015 at 15:20

shankly1985 said:
Thats 720p Should have said am running 1440p

humbug said:
I'm adding DX11 to this but removing all resolutions above 720P.
As we found out higher resolutions make no difference to scores.
its also far too cluttered with so many resolutions.
Its better to make room for DX11 and clean it up so that people can actually see whats going on.

i will post back when i have made the changes. thanks

shankly1985 · 12 Jul 2015 at 15:23

bru said:

Well am not sure... Because they is defo a difference changing resolution.

http://www.3dmark.com/compare/aot/42800/aot/42647

shankly1985 · 12 Jul 2015 at 15:26

So we just going to keep on crying and playing Mr Don or we going to start posting some results?

humbug · 12 Jul 2015 at 15:30

Hello

shankly1985 · 12 Jul 2015 at 15:35

humbug said:
Hello

Hi!

Humbug am seeing different results, based on different Resolutions..

humbug · 12 Jul 2015 at 15:48

shankly1985 said:
Hi!

Humbug am seeing different results, based on different Resolutions..

Your results on post #1?

It looks like you have about a 10% difference in DX11 Single threaded. And maybe Multi threaded, a bit. is that consistent?

The results i found at the time

humbug said:
Win8.1 - Mantle - FX-9590 @ 4.7Ghz - 290 @ 1180/1375.

720P: 14,405,912. http://www.3dmark.com/3dm/6411636

1080P: 14,306,375. http://www.3dmark.com/3dm/6411993

1440P: 14,159,114. http://www.3dmark.com/3dm/6411686

4K: 13,890,447. http://www.3dmark.com/3dm/6411705

5K: 13,408,873. http://www.3dmark.com/3dm/6411731

8K: 12,106,05. http://www.3dmark.com/3dm/6411752

above 4K it starts to drop but thats because it starts to become GPU bound

Its what everyone else also found, thats why i changed it, that and it was so cluttered no one could make sense of it.

What do you suggest? i mean your results look pretty consistent apart from DX11 single thread.

shankly1985 · 12 Jul 2015 at 15:54

humbug said:
Your results on post #1?

It looks like you have about a 10% difference in DX11 Single threaded. And maybe Multi threaded, a bit. is that consistent?

The results i found at the time

above 4K it starts to drop but thats because it starts to become GPU bound

Its what everyone else also found, thats why i changed it, that and it was so cluttered no one could make sense of it.

What do you suggest? i mean your results look pretty consistent apart from DX11 single thread.

If they is a difference to be had then I feel all bench results should have there own section.

humbug · 12 Jul 2015 at 15:57

shankly1985 said:
If they is a difference to be had then I feel all bench results should have there own section.

Sectioned by Driver? it looks like the Driver is better. Run that first one again at 720P

humbug · 12 Jul 2015 at 16:02

Scratch that. the draw calls per frame threw me. hang on...

shankly1985 · 12 Jul 2015 at 16:03

humbug said:
Sectioned by Driver? it looks like the Driver is better. Run that first one again at 720P

720p

DX11 Multi-threaded draw calls per second 1,076 296
DX11 Single-threaded draw calls per second 1,081 032
DX11 Multi-threaded draw calls per frame 35,328
DX11 Single-threaded draw calls per frame 38,400

ok I see what you might be getting at here.. Now my 720p test is much worst then it was before.

humbug · 12 Jul 2015 at 16:07

Your getting more DrawCalls in the Frame at higher res because the Frame is bigger, it has more information in it, when you crunch it down its still the same number of DrawCalls at any given time, thats why your DrawCalls per second are the same.

Edit: I.E the latency in calls is still the same.

shankly1985 · 12 Jul 2015 at 16:12

humbug said:
Your getting more DrawCalls in the Frame at higher res because the Frame is bigger, it has more information in it, when you crunch it down its still the same number of DrawCalls at any given time, thats why your DrawCalls per second are the same.

Edit: I.E the latency in calls is still the same.

Ok whats the most important result. Per Frame or Per Second?

humbug · 12 Jul 2015 at 16:18

shankly1985 said:
Ok whats the most important result. Per Frame or Per Second?

Per Second, as a completely inaccurate but practical example: if the API with one thread takes 20ms to call an object then it will take 100ms to call 5 objects. so if there are more objects to call it will call more objects but take longer, so at any given time the amount of objects called is the same.

Dygaza · 12 Jul 2015 at 16:23

I7-3770k @ 4.3GHz
7970 @ 1100/1500MHz
W7

15.6

DX11 MT 907 441
DX11 ST 921 565

1036:

DX11 MT 1 091 712
DX11 ST 1 184 316

1038:

DX11 MT 1 121 610
DX11 ST 1 184 317

1023.5

DX11 MT 1 137 720
DX11 ST 1 193 648

1040.0

DX11 MT 1 203 383
DX11 ST 1 252 605

1023.10

DX11 MT 1 216 165
DX11 ST 1 290 934

15.15

DX11 MT 1 185 952
DX11 ST 1 201 633

15.7

DX11 MT 1 223 682
DX11 ST 1 284 336

Shanks, resolution doesn't really matter in 3dmark api test as gpu is running on low usage.