The Shadow of Mordor Bench.

LtMatt · 26 Jul 2015 at 14:08

RavenXXX2 said:
Perfect scaling would be double your single cards fps not over 3x the fps. Dodgy.

Gregster said:
But you have gone from 39 fps to 139 fps. Sorry but that isn't perfect scaling, that is questionable scaling to me, as that is effectively running the same as 3.5 Fury Xs but only using 2 cards. Something definitely up with that.

Anyways, not my shout and Kaap can decide what to do but something is wrong there as perfect scaling would be 78 fps.

I just ran it again using the default AMD SOM profile and got 85FPS, this was with my cpu at stock and a warm run. No clue why to be honest and Kaap can test this himself using the cards to verify what I'm saying is true.

LtMatt · 26 Jul 2015 at 14:35

Technical answer from one of our Catalyst driver engineers on why AFR Friendly can sometimes be faster than the default AMD CrossFire profile.

AFR (Alternating Frame Rate) splits the workload amongst the GPUs by assigning a given frame to each GPU in the CrossFire Chain. So GPU1 renders frame 1 and GPU2 renders frame 2. In theory, you can get double the number of frames because GPU1 can jump straight to frame 3 without working on frame 2. In practice, this is reduced due to CPU overhead and other timing mechanisms which prevent you from always having something to work on.

3D Games are always working on ways to improve frame rate at the same time as providing the best possible visual experience. The easiest way is to re-use data that has already been calculated and sitting in memory from the previous frame. So Frame 1 gets rendered fully, but Frame 2 could just render the differences and copy/paste the unchanged data from frame 1. This causes a headache for MGPU configurations because the data from Frame 1 is not on the same GPU that is rendering Frame 2. If you don't do the copy (and the game has no idea that it has to do so), you end up with flickering issues because Frame 2 is lacking the information in Frame 1. Since it is the GPU driver which is running CF, it needs to detect that a resource is missing and do a copy from the memory of GPU1 to GPU2. This takes time because the path and latency across the PCI-E Bus or CF Interconnects are much greater than between a given GPU and its own onboard memory. The more resources that need to be copied, the longer it takes for Frame 2 to finish and get presented after Frame 1.

Certain CrossFire profiles will skip the transfers that are unnecessary and in that case Frame 2 finishes very quickly because it has less data to compute than Frame 1. In that case your frame rate jumps! Normal CrossFire Profiles will always have some type of extra transfer for various games, but the AFR Friendly one only has very basic settings so it will probably be faster.

andybird123 · 26 Jul 2015 at 14:40

key words in that write up matt "you can get double the number of frames"
not 3 and a half times

LtMatt · 26 Jul 2015 at 14:44

andybird123 said:
key words in that write up matt "you can get double the number of frames"
not 3 and a half times

I posted my results and was honest about the method used and i spoke with Kaap about these results prior to posting. That said, I'm happy to use the default profile only but i thought I'd post them anyway as they're interesting. With even the stock profile offering over 100% scaling though I'm not sure i can win either way, shrug.

Gregster · 26 Jul 2015 at 14:57

AMDMatt said:
I posted my results and was honest about the method used and i spoke with Kaap about these results prior to posting. That said, I'm happy to use the default profile only but i thought I'd post them anyway as they're interesting. With even the stock profile offering over 100% scaling though I'm not sure i can win either way, shrug.

All good Matt and like I said, it is Kaap's call if he allows it (and seems he will as you spoke with him). I just find it odd that 2 GPU's that you say are perfect scaling are actually scaling like 3.5 GPUs using driver tweaks.

LtMatt · 26 Jul 2015 at 15:04

Gregster said:
All good Matt and like I said, it is Kaap's call if he allows it (and seems he will as you spoke with him). I just find it odd that 2 GPU's that you say are perfect scaling are actually scaling like 3.5 GPUs using driver tweaks.

I don't know if he will, but he knew that AFR Friendly gave a big boost over the default profile. I'm happy to only use the stock profile of course so can re-run no problem, but the scaling is still over 100% with two gpu's on the stock profile so not sure what can be done. Once i ramp up the cpu and do a cold runm, that scaling is going to shoot higher as well. Are we now going to disallow that as well because it's over 100% scaling? It's a tricky one. AFR Friendly is a very common method used in multi gpu. As to why there is such a big increase though when using this particular method in this game, I've no idea.

I've been testing it on other games including Firestrike and it results in worse performance in everything i tried, aside from FarCry 4.

Gregster · 26 Jul 2015 at 15:10

AMDMatt said:
I don't know if he will, but he knew that AFR Friendly gave a big boost over the default profile. I'm happy to only use the stock profile of course so can re-run no problem, but the scaling is still over 100% with two gpu's on the stock profile so not sure what can be done. Once i ramp up the cpu and do a cold runm, that scaling is going to shoot higher as well. Are we now going to disallow that as well because it's over 100% scaling? It's a tricky one. AFR Friendly is a very common method used in multi gpu. As to why there is such a big increase though when using this particular method in this game, I've no idea.

I've been testing it on other games including Firestrike and it results in worse performance in everything i tried, aside from FarCry 4.

Not my call and it is down to Kaap if he wants to allow driver tweaks. It does leave question marks though for obvious reasons and even 2 cards scaling higher than they should isn't right. But if you have been altering driver settings, that could be the reason.

Anyways, all good

LtMatt · 26 Jul 2015 at 15:19

Gregster said:
Not my call and it is down to Kaap if he wants to allow driver tweaks. It does leave question marks though for obvious reasons and even 2 cards scaling higher than they should isn't right. But if you have been altering driver settings, that could be the reason.

Anyways, all good

Always use default settings and never cheat, not in my nature.

What can you do though if CrossFire scaling is over 100% using the stock CrossFire profile/settings?

I also wouldn't consider using a different CrossFire profile as cheating, as it's just using a different method of multi gpu, ie AFR Friendly, 1x1 Optimize, or a custom profile from AMD using one of the above plus additional optimisations.

If you were disabling Tess or altering LOD or using lower settings from within the game then that's what i would consider cheating.

Gregster · 26 Jul 2015 at 15:26

Just seems odd that no other test site has scaling over 100%, even with the default profile and not using driver tweaks.

http://www.pcworld.com/article/2947...n-fury-crossfire-review-2-fast-2-furious.html

http://www.gamersnexus.net/hwreviews/2008-amd-radeon-r9-fury-x-full-analysis-and-crossfire/Page-3

Something is deffo up for it to scale past 100% even on the default profile before you started using driver tweaks.

Anyways, it is just a bit of fun, so no worries and crack on

GoogalyMoogaly · 26 Jul 2015 at 15:26

AMDMatt said:
Always use default settings and never cheat, not in my nature.

What can you do though if CrossFire scaling is over 100% using the stock CrossFire profile/settings?

I also wouldn't consider using a different CrossFire profile as cheating, as it's just using a different method of multi gpu, ie AFR Friendly, 1x1 Optimize, or a custom profile from AMD using one of the above plus additional optimisations.

If you were disabling Tess or altering LOD or using lower settings from within the game then that's what i would consider cheating.

And yet remember the uproar when _Alatar_ got a higher score in whatever benchmark is was than people expected?
That was after he removed a higher score that he decided not to use because it was using altered drivers (or something like that). there must've been nearly a page of that thread of people saying how his score couldn't be trusted and it took days before it was.

truther · 26 Jul 2015 at 15:30

hi , I got the best 4k score single gpu and its a 980 strix beats fury amd matt and beats 980ti, how do I post screen shot , all I get is black screen from prntscrn shot of fps result

Gregster · 26 Jul 2015 at 15:32

truther said:
hi , I got the best 4k score single gpu and its a 980 strix beats fury amd matt and beats 980ti

, how do I post screen shot , all I get is black screen from prntscrn shot of fps result

Take a screenshot with MSI AB

LtMatt · 26 Jul 2015 at 15:32

Gregster said:
Just seems odd that no other test site has scaling over 100%, even with the default profile and not using driver tweaks.

http://www.pcworld.com/article/2947...n-fury-crossfire-review-2-fast-2-furious.html

http://www.gamersnexus.net/hwreviews/2008-amd-radeon-r9-fury-x-full-analysis-and-crossfire/Page-3

Something is deffo up for it to scale past 100% even on the default profile before you started using driver tweaks.

Anyways, it is just a bit of fun, so no worries and crack on

There are no driver tweaks, I'm not sure why it's scaling past 100%. I just ran it again to make sure and scored 84.5 FPS so not sure to be honest.

GoogalyMoogaly said:
And yet remember the uproar when _Alatar_ got a higher score in whatever benchmark is was than people expected?
That was after he removed a higher score that he decided not to use because it was using altered drivers (or something like that). there must've been nearly a page of that thread of people saying how his score couldn't be trusted and it took days before it was.

Yes i remember and credit to him for at least admitting he used LOD cheats. I was upfront here about the method used and people with 2x Fury X or more can test it and see for themselves.

truther · 26 Jul 2015 at 15:34

a sneak peak once I can get screen shot up

]

LtMatt · 26 Jul 2015 at 15:35

^ Nice desktop.

Wrap spoiler tags around your images like this and remove the * [spoiler*][/spoiler*]

Lokken86 · 26 Jul 2015 at 15:36

^^ Or just make it smaller. Much much smaller

.

truther · 26 Jul 2015 at 15:38

AMDMatt said:
^ Nice desktop.

Wrap spoiler tags around your images like this and remove the * [spoiler*][/spoiler*]

thanks amd matt I learnt something new and saved the poor sould with their puny resoloutions from my 48 4k 60hz 444 display

Gregster · 26 Jul 2015 at 15:40

Anyways, here is my score for the thread with no driver tweaks and everything as it should be.

GPU Titan X @ 1409/4002 - 3930K @ 4.4 - 353.38 drivers - 3840x2160

LtMatt · 26 Jul 2015 at 15:40

truther said:
thanks amd matt I learnt something new and saved the poor sould with their puny resoloutions from my 48 4k 60hz 444 display

I understand, using 4K here so you get used to using spoiler tags.

Silent_Scone · 26 Jul 2015 at 15:46

Mines almost down so will do some runs seeing as I've managed to break 3DMark FS

Roll on Windows 10.