[hardocp] R9 390x with 99.7% scaling in ROTR Crossfire

Azix

Golden Member
Apr 18, 2014
1,438
67
91
1455189919EDyKUcGV8E_10_1.gif


We are comparing the percentage increase from GTX 980 to GTX 980 SLI and then from R9 390X to R9 390X CrossFire.

Adding a second GeForce GTX 980 improves performance by 83.4%. That means we are seeing scaling efficiency of 83.4% out of GTX 980 SLI.

Adding a second Radeon R9 390X improves performance by 99.7%. That means we are seeing scaling efficiency of 99.7% out of R9 390X CrossFire.

We checked and re-tested this test three times after we saw that kind of scaling out of R9 390X CrossFire. Every run was consistent and every time this is the kind of scaling we experienced out of R9 390X CrossFire. This is by far the best and highest scaling efficiency we've ever come across for any video card in any game. This kind of scaling is what allows the R9 390X CrossFire to compete well at 4K with the competition. We aren't sure if the reason for this is due to the doubled 8GB of VRAM, or what, but it is what it is and it is incredible.

People will swear the 970 and 980 are better for Tomb Raider. Same will likely happen with the division.
 
Last edited:

Shmee

Memory & Storage, Graphics Cards Mod Elite Member
Super Moderator
Sep 13, 2008
7,381
2,414
146
Impressive scaling!
 

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136
That is some fine driver work to get such impressive scaling!
 
Feb 19, 2009
10,457
10
76
That is some fine driver work to get such impressive scaling!

Yes and no, because they fixed Fury X stutter in single card with 16.1.1 but they didn't fix it for CF so it's unplayable.

Sure, the rest of the stack is fine, but really their top cards should get priority treatment.
 
Feb 19, 2009
10,457
10
76
Fury AIR is also tested, not Fury X for CFX. Which means you have 2x $550 cards beating out the 2x $650 ones.

It's actually unplayable on Fiji CF because of the crazy stutters in 16.1.1.

If i were owners of CF Fiji GPUs, I would be extremely pissed.

AMD failed with their driver support, again. Things like this is cements the "AMD = bad drivers mantra".
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
It's actually unplayable on Fiji CF because of the crazy stutters in 16.1.1.

If i were owners of CF Fiji GPUs, I would be extremely pissed.

AMD failed with their driver support, again. Things like this is cements the "AMD = bad drivers mantra".

You are extremely pissed about Fury already and you don't even own them :D
 

Bacon1

Diamond Member
Feb 14, 2016
3,430
1,018
91
It's actually unplayable on Fiji CF because of the crazy stutters in 16.1.1.

If i were owners of CF Fiji GPUs, I would be extremely pissed.

AMD failed with their driver support, again. Things like this is cements the "AMD = bad drivers mantra".

They didn't say it was unplayable, and they never did any investigation into what was causing the stuttering. They blamed it on 4GB of ram, yet neither Single Card Fury, nore the 980 SLI had issues and they would if it was ram size. Hell single Fury had higher min FPS than 980 TI.

You could probably turn down one or two settings and have it be much better, not to mention playing @ 40fps avg doesn't sound very appealing.

HBAO+ -> On would give a good boost, and turning off motion blur, and either texture or shadow to high from VH would probably solve all issues with minimal IQ differences (I'd do that on the 980 TI SLI as well).

Also:

The first patch on February 5th was significant enough in terms of performance that we had to scrap all of our original data and start over. Three days ago we did see another patch as well that addressed, "Various performance improvements for GPU-bound situations."

They tested and saw 5-7% increase in perf @ 4k, and 1-2% in single card 1440p when testing the 980, so who knows if that fixed any issues or not with Fury, it wasn't tested.

So there is obviously a issue (driver? Game?) with Fury CFX that causes minimum spikes, maybe cpu overhead from drivers? Wish they would have tested it instead of blaming 4GB of ram like always.
 
Last edited:

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
They didn't say it was unplayable, and they never did any investigation into what was causing the stuttering. They blamed it on 4GB of ram, yet neither Single Card Fury, nore the 980 SLI had issues and they would if it was ram size. Hell single Fury had higher min FPS than 980 TI.

You could probably turn down one or two settings and have it be much better, not to mention playing @ 40fps avg doesn't sound very appealing.

HBAO+ -> On would give a good boost, and turning off motion blur, and either texture or shadow to high from VH would probably solve all issues with minimal IQ differences (I'd do that on the 980 TI SLI as well).

Also:



They tested and saw 5-7% increase in perf @ 4k, and 1-2% in single card 1440p when testing the 980, so who knows if that fixed any issues or not with Fury, it wasn't tested.

So there is obviously a issue (driver? Game?) with Fury CFX that causes minimum spikes, maybe cpu overhead from drivers? Wish they would have tested it instead of blaming 4GB of ram like always.

Blame it on 4GB of RAM because that's the perceived unfixable shortcoming of Fiji. Even though it's never been shown to be an issue. There is zero evidence that is the issue. As you said, it has higher mins than the 980 ti. If it were VRAM bottlenecked the minimums would tank.
 

redzo

Senior member
Nov 21, 2007
547
5
81
I wonder if this applies to 290 xfire as well.
So there is obviously a issue (driver? Game?) with Fury CFX that causes minimum spikes, maybe cpu overhead from drivers? Wish they would have tested it instead of blaming 4GB of ram like always.
Yes I also think that he should've just benched a 290x 4GB crossfire setup and call it a day.
 

zlatan

Senior member
Mar 15, 2011
580
291
136
The SLI bridge bandwith is too limited to get near perfect scaling. A CF XDMA solution has much higher bandwith.
 
Feb 19, 2009
10,457
10
76
The SLI bridge bandwith is too limited to get near perfect scaling. A CF XDMA solution has much higher bandwith.

Yes but it doesn't help them when their drivers are a let-down, particularly for newly released AAA titles where people who have multi-GPU need the power the most.

Makes me very hesitant to go CF again with Polaris.
 

provost

Member
Aug 7, 2013
51
1
16
Not sure what’s so great about SLI anymore, other than getting a SLI profile. I am fine waiting a couple of months to get a CF profile, if it means 80%-90% scaling, thereby effectively doubling performance, as opposed to the meagre SLI scaling gains… kind a defeats the purpose of having SLI or may be its just gimped. I beginning to question all the assumptions about by six NVidia cards, tbh
 

Ma_Deuce

Member
Jun 19, 2015
175
0
0
Yes but it doesn't help them when their drivers are a let-down, particularly for newly released AAA titles where people who have multi-GPU need the power the most.

Makes me very hesitant to go CF again with Polaris.

I was very let down by the crossfire on the 390x too. A lot of FPS left on the table there.
 

Flapdrol1337

Golden Member
May 21, 2014
1,677
93
91
If you have variable framerate and each card renders alternate frames you can only have perfect scaling if there is no frame pacing, unless the game does it's own pacing intelligently or something.

I'm always suspicious of multigpu.
 

tential

Diamond Member
May 13, 2008
7,355
642
121
Yes but it doesn't help them when their drivers are a let-down, particularly for newly released AAA titles where people who have multi-GPU need the power the most.

Makes me very hesitant to go CF again with Polaris.
I will go cf with Polaris. Fiji is a dud.
 

Bacon1

Diamond Member
Feb 14, 2016
3,430
1,018
91
The 4GB of HBM really kills it. If I were buying an AMD card today it'd be 390X, no question.

The 4GB isn't holding the card back, look at single card performance. Its just a driver bug or game bug with CFX that is causing the stutters / framedrops on Fury.

Fury X has higher min (35 vs 29 (20%)) higher max (73 vs 68 (7%)) and almost the same (48.9 vs 49.4 (1%)) FPS as the 980 TI. Somehow the 4GB of ram doesn't hold it back at all. I'd take the 20% better minimum vs 1% (.5 fps) difference in average framerates any day.

Their testing may be done well, but their observation / analysis is off a lot of the time.

Not to mention they didn't even mention that SMAA does absolutely nothing.

https://www.reddit.com/r/pcgaming/comments/434w9l/just_a_heads_up_dont_run_smaa_on_rise_of_the_tomb/
 

Azix

Golden Member
Apr 18, 2014
1,438
67
91
Maybe the memory management is not yet optimized for the game. Or there is an issue with the PCI-e interface they had. Its some issue but likely not just the VRAM capacity.
 

Headfoot

Diamond Member
Feb 28, 2008
4,444
641
126
There is additional memory overhead involved with crossfire. There were some graphs of GTAV running with decent averages and mins with a single 290 and much higher averages and very low mins on 2xCF (via 295x2). I don't know how much it is but it exists. Now, I'm not saying that is the case here. There is not evidence to say one way or the other without more specific testing. It is certainly possible if the game is right on the precipice of 4GB.

An easy test to help zero in would be 2x 4GB 390's now that they exist vs 2x 8GB 390s and see what happens.