[DX12] Fable Legends Beta Benchmarks

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

happy medium

Lifer
Jun 8, 2003
14,387
480
126
This pretty much sums it up....................from article.
http://www.pcper.com/reviews/Graphi...2-Performance-Testing-Continues/Looking-Ahead

"Clearly there is a back and forth between AMD and NVIDIA in these results - NVIDIA's GTX 980 Ti is faster than the Fury X and the GTX 950 is faster than the R7 370. But the R9 390X takes home a performance lead over the GTX 980 and the R9 380 is faster than the GTX 960. None of the margins are Earth shattering though."


One big meh, big deal.
 

TheELF

Diamond Member
Dec 22, 2012
3,974
731
126
If AT ends up using this game for testing in 2016, I hope they upgrade their CPU platform too.

Intel Core i7-4960X in 3 modes:

'Core i7' - 6 Cores, 12 Threads at 4.2 GHz
'Core i5' - 4 Cores, 4 Threads at 3.8 GHz
'Core i3' - 2 Cores, 4 Threads at 3.8 GHz
That's the problem with these tests...you need an fast I7 so it can issue all the commands to all the ACEs at the same time and you have no gamecode running(remember every VGA in ashes having halve the fps with anything lower than an i5? ) ,so it's pretty it's interesting it's nice for AMD but the conclusions will be drawn when there is an actual game and the VGAs will have to do a certain effect/work at a certain time then we will see how the cards will run on more normal CPUs and what kind of gains they will have.

Because up until now these benches are like comparing CPUs with cinebench, yes it tells you which one has more computing power but it doesn't tell you how much of this power will be usable in games.
 

Qwertilot

Golden Member
Nov 28, 2013
1,604
257
126
Especially when that review is actually doing something markedly reasonable in context :)

No reference Fury, and the 'reference' FuryX is already (very) heavily overclocked out of the box. So a sanely factory overclocked 980ti simply is the fair comparison point.

You really don't want to start randomly rewarding companies for not producing a sensibly clocked reference model.

The problem with using an overclocked card is of course that you then start wondering about what they're doing with the 960's, 980's, 3xx's etc.....
 

RampantAndroid

Diamond Member
Jun 27, 2004
6,591
3
81
I guess it's a normal practice now to test some of the best factory pre-overclocked NV cards and keep AMD cards at stock speeds. No wonder AMD didn't want to send samples to this site as they continue to show total lack of professionalism when testing products fairly. We see 980Ti beating Fury X by 17%, but the Asus Strix 980Ti OC Edition is itself 15-16% faster than a stock 980Ti. Talk about a misleading review.


Wait, what? Every review I see compares the card they're reviewing to stock competitors. It's a serious complaint I have. I don't give a flip how the Fury X compares to the stock 980TI. I care about how it compares to the OCed 980TIs that cost $20 more than stock.

It doesn't strike me as being out the ordinary. I don't like it, but it isn't something just one reviewer does.

If Fury X could overclock as well as the 980Ti, this would have been a clean sweep for AMD in Fable Legends.

So what you're saying is "they shouldn't use products readily available on the market for reviews"? It isn't the reviewer's fault that no good OCed Fury X models exist. That's on AMD, and that the Maxwell GPUs OC so well should be represented in reviews....include a factory model if you want, but the fact that you want them to omit a card that people WILL buy makes me trust your posts a bit less.
 
Last edited:

DeathReborn

Platinum Member
Oct 11, 2005
2,750
746
136
@RussianSensation - Have you seen any Nvidia drivers released for Fable Legends? Who knows maybe when there are it'll improve performance by 3-5% as you say AMD's might.

I am all for using the most up to date drivers but only those that arrive in time for the testing & with utmost fairness to both sides.
 

amenx

Diamond Member
Dec 17, 2004
3,989
2,235
136
Wait, what? Every review I see compares the card they're reviewing to stock competitors. It's a serious complaint I have. I don't give a flip how the Fury X compares to the stock 980TI. I care about how it compares to the OCed 980TIs that cost $20 more than stock.

It doesn't strike me as being out the ordinary. I don't like it, but it isn't something just one reviewer does.



So what you're saying is "they shouldn't use products readily available on the market for reviews"? It isn't the reviewer's fault that no good OCed Fury X models exist. That's on AMD, and that the Maxwell GPUs OC so well should be represented in reviews....include a factory model if you want, but the fact that you want them to omit a card that people WILL buy makes me trust your posts a bit less.
I think Nvidia should follow AMDs strategy. Jack up stock clocks of their reference cards to the hilt and call that 'stock'. An old strategy often seen with AMDs CPUs that left virtually no headroom and now finding its way in their GPUs :D.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Wait, what? Every review I see compares the card they're reviewing to stock competitors. It's a serious complaint I have. I don't give a flip how the Fury X compares to the stock 980TI. I care about how it compares to the OCed 980TIs that cost $20 more than stock.

It doesn't strike me as being out the ordinary. I don't like it, but it isn't something just one reviewer does.

Yes, but when trying to assess DX12 on its own, we need to see how a stock GTX980Ti compares to a stock Fury X and then look at their relative standing in DX11 in other games. When you introduce variables like overclocking performance, it means now we have to find DX11 reviews with Asus Strix 980Ti. We are trying to conclude on the impact of DX12 vs. DX11, not trying to argue if 980TI is faster than the Fury X overall. We are not trying to prove whether or not Fury X is better than 980Ti as that's not the point (we know 980Ti is better because it has 20-25% OCing headroom and 50% more VRAM).

Look at what I am saying:

1080P Asus Strix 980Ti is 27% faster than the Fury X in DX11 games

perfrel_1920.gif


But under a DX12 game, that advantage drops to 17%.

fable-fps.gif


Ashes of Singularity shows the same scenario where GCN benefits much more from DX12 than Maxwell does.

Furthermore, Fury X beating 390X by only 3 fps at TR could suggest some driver optimization issue as well here.

So what you're saying is "they shouldn't use products readily available on the market for reviews"? It isn't the reviewer's fault that no good OCed Fury X models exist. .

That's not what I am saying at all. The test is used to compare the impact of DX12 games on existing architectures/products. Therefore, we need a stock 980Ti data point as a reference to make this point clearer. There is nothing wrong with considering/including an after market 980Ti cards in the real world and no one says it shouldn't be done, but the point is to see the impact of DX12, we need stock basis (it's OK to include an after-market 980Ti that's 15-16% faster but then it should be labelled on the graph as an after-market card).

Just imagine the alternative scenario if we compared a stock GTX660Ti to an 1200mhz HD7950 in Fable Legends and concluded that 7950 is faster by 30-40%. That's telling us more about overclocking capabilities of the latter, rather than DX11 vs. DX12.

Again, we already know that an after-market 980Ti beats the Fury X but a DX12 benchmark goes far beyond that. In other words, imagine if we included a 1.5Ghz 980Ti in the benchmarks and focused solely on 980Ti trashing the Fury X. Would you then conclude that Maxwell's DX12 performance is as good as AMD's? Not necessarily because now you are comparing overclocking performance above 980Ti's stock, not isolating DX12 variable. You need to look at other cards too to see whether or not 980Ti's crazy overclocking is skewing the data. In this case, it clearly is.

This is exactly why using a factory pre-overclocked 980Ti that shows Maxwell in the best light possible is misleading for the overall conclusion because we clearly see all other NV cards like 960/970/980 are doing much worse against R9 380/285/390/390X.

Do you see my point now?

I mean sure it's real world that a $650 980Ti with a 15-16% factory pre-overclock is beating a Fury X by 17%, but looking at that only overshadows how an after-market $230 R9 290 is 70%+ faster than a $200 960, how $280 390 is on the heels of a $450 980, and how an R9 290X is beating a 970. 980Ti's excellent performance is a testament to NV's awesome GM200 design here but the rest of NV's line-up is not fairing good at all given the price/performance points of their $160-500 cards.
 
Last edited:

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
You do realize the graph you posted is for the CPU scaling.

Yes, for two reasons.

You alluded to DX12 vindicating AMD, specifically that CPU is now unimportant. The graph shows that to be false.

It also shows that unlike what you claim, Nvidia is still faster than AMD.

You definitely didn't answer my question, how does AMD's scaling look?
 

Techhog

Platinum Member
Sep 11, 2013
2,834
2
26
No reference Fury, and the 'reference' FuryX is already (very) heavily overclocked out of the box. So a sanely factory overclocked 980ti simply is the fair comparison point.

... *sigh* You guys will say anything to support your favorite company. THis is just a ridiculous arguement.
 
Feb 19, 2009
10,457
10
76
That's a complier issue. The D3D bytecode not designed for GCN, and the FXC optimization make more hurt than good for AMD. So the Catalyst must deoptimize the compiled bytecode to compile a good binary for GCN.
On Xbox One the HLSL complier can compile a much faster code. The same HLSL code for PC is 30 percent slower avg. And the reason behind this is the D3D bytecode. SPIR-V will be an ultimate weapon for AMD because it will be a modern IR.

Microsoft will probably replace the D3D bytecode in the future, because most companies will build a GCN-like architecture, and the change is really needed in the complier space for this.

So you are saying the bench is using async compute, and its unoptimized for GCN. But what about Maxwell? Did NV finally added async compute support in their drivers? Last time we heard any on this, apparently the feature wasn't ready and not enabled yet.

Edit: Got some more info, Joel Hruska at Extremetech said that Lionhead Studios did inform him that the benchmark has a low amount of async compute usage, within each frame is 5% worth of compute that could be performed asynchronously. Given that Zlatan says the AC is 30% slower than normal on the PC build vs Xbone (compiler issue needs driver optimization), that 5% is therefore about 3% performance gain that GCN can flex over anything that doesn't support AC. Which is so small, it's pointless in the overall performance.
 
Last edited:

kawi6rr

Senior member
Oct 17, 2013
567
156
116
I think Nvidia should follow AMDs strategy. Jack up stock clocks of their reference cards to the hilt and call that 'stock'. An old strategy often seen with AMDs CPUs that left virtually no headroom and now finding its way in their GPUs :D.

And for us people who don't overclock anymore or at all it's a win win!
 

Makaveli

Diamond Member
Feb 8, 2002
4,733
1,072
136
1080P
280X is 19% faster than GTX770

Thanks for posting the link I didn't see that review yet.

So it looks like the 7970Ghz/280x is doing 50fps in this benchmark.

Give or take 5fps here and there with final code and overclocking I will take 55fps good enough for me at 1200p.
 

dogen1

Senior member
Oct 14, 2014
739
40
91
I don't see how a driver update will do much for supposedly thin driver.

You miss zlatan's post?

Given that Zlatan says the AC is 30% slower than normal on the PC build vs Xbone (compiler issue needs driver optimization), that 5% is therefore about 3% performance gain that GCN can flex over anything that doesn't support AC. Which is so small, it's pointless in the overall performance.

He said that the HLSL compiler produces 30% slower code, not AC is 30% slower.
 
Last edited:
Feb 19, 2009
10,457
10
76
You miss zlatan's post?

He said that the compiler produces 30% slower HLSL code, not AC is 30% slower.

In the context of AC as was the discussion and his prior post brought up.

Either way, in the best case scenario, AC for Fable will lift performance by 5% for GPUs that can execute & process compute in parallel vs GPUs that do it in serial.

Compared to how UE4 games normally run on AMD vs NV, the results we're seeing are nothing short of outstanding for DX12 & AMD. Imagine the likes of Project Cars in DX11 and suddenly we see a massive lift in AMD performance in DX12. Because the UE4 gap is about that bad for AMD normally.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Thanks for posting the link I didn't see that review yet.

So it looks like the 7970Ghz/280x is doing 50fps in this benchmark.

That's probably a conservative figure. 1150mhz 7970 is a 24% OC over the 925mhz version. 44.68 x 1.15X scaling = 51 fps.

Compared to how UE4 games normally run on AMD vs NV, the results we're seeing are nothing short of outstanding for DX12 & AMD. Imagine the likes of Project Cars in DX11 and suddenly we see a massive lift in AMD performance in DX12. Because the UE4 gap is about that bad for AMD normally.


It's also interesting how some people are so keen to focus on 980Ti's performance numbers and ignoring the elephants in the room.

1) UE4 DX11 performance up to now showed massive advantage for Maxwell. Fable Legends UE4 DX12 benchmarks do not show any inherent advantage for Maxwell over GCN. What happens in a brand agnostic game engine like Frostbite, CryEngine under DX12? For months we heard how DX12 was an AMD marketing gimmick.

2) 925mhz HD7970 is smashing a GTX960 by 32% on AT.

What's 32%? Take that and increase GPU clocks yet another 13.5%, and you end up with a $190 R9 280X that's going to be 40-45% faster than a GTX960.

Not a big deal for a budget gamer, right? :hmm:

HardOCP, TR and tons of people online kept recommending GTX960 2GB cards and giving them Gold Awards, fully ignoring the raw GPU horsepower deficits, fully ignoring 2GB VRAM limits, and now DX12 completely hammers the 960 in the first two DX12 benchmarks. When are all these 'professional' sites and forum members who kept recommending GTX960 2-4GB over R9 280/280X/R9 290 going to admit they were wrong all along?

All I see is marketing hype of perf/watt and doing everything possible to not recommend the superior products. Gold Award for a GTX960 2GB. Are you kidding me? These 'professionals' should be ashamed of themselves for not seeing what the rest of the PC gaming community was predicting. They have lost our trust but being so short-sighted and not seeing all the key trends/forum posts around them for months (heck 960 came out in January 2015 but is still recommended even now!)

They should admit publicly when they are wrong and pay more attention to opinions of gamers worldwide who call them out on their horrid recommendations. I feel sorry for anyone who bought into the biased reviewer's hype and bought GTX950/960 close to launch with the intention of keeping them for 2-3 years.

Even if NV improves performance in future drivers, are they going to add 30-40% more performance to the GTX960? Doubtful.
 
Last edited:

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
So you are saying the bench is using async compute, and its unoptimized for GCN. But what about Maxwell? Did NV finally added async compute support in their drivers? Last time we heard any on this, apparently the feature wasn't ready and not enabled yet.

That's not what he specifically addressed in his post ...

He's just saying the reason why the GI kernel runs slower is because the HLSL compiler does not generate the ideal intermediate code sequence on GCN and there maybe some truth to that ...

Back when vec4 was still a thing, the HLSL compiler was built around that programming model in mind but now almost every GPU micro-architecture that you see nowadays are scalar based architectures however the optimizations that the compiler does to vectorize the code just hurts these architectures when it's emitting useless instructions and wasting registers ...
 

Azix

Golden Member
Apr 18, 2014
1,438
67
91
All I want to know is if this game makes proper use of async or not. If not, they failed me. All that talk about it seems to have just been for the xbox version.
 

antihelten

Golden Member
Feb 2, 2012
1,764
274
126
Not showing the same as AOTS. How...unexpected.

Actually, as far as I can tell this show almost exactly the same as AOTS, large relative gains for AMDs midrange GPUs (i.e. Tahiti/Hawaii), but comparatively poor gains for AMDs high end GPUs (Fiji).

But let's start with the elephant in the room:

1. "The Test
The software provided to us is a prerelease version of Fable Legends, with early drivers, so ultimately the performance at this point is most likely not representative of the game at launch and should improve before release. What we will see here is more of a broad picture painting how different GPUs will scale when DX12 features are thrown into the mix. In fact, AMD sent us a note that there is a new driver available specifically for this benchmark which should improve the scores on the Fury X, although it arrived too late for this pre-release look at Fable Legends"


So to start off that means the AT results online aren't even using the latest driver from AMD that has optimizations for this title.

This only becomes even more ironic, when a few paraghraphs above this Anandtech writes this:

"We tackled two synthetic tests earlier this year, Star Swarm and 3DMark, but due to timing and other industry events, we are waiting for a better time to test the Ashes of the Singularity benchmark as the game nears completion"

So apparently they have no problems waiting weeks (if not months) for updated drivers/benchmarks for AOTS, but when it comes to Fable they can't even wait a few days to redo the tests with AMDs updated drivers.

Fury X scaling in comparison to Hawaii is pretty poor ...

They'll really need to reconsider on lowering the occupancy required to get high utilization on their next micro-architecture ...

Why are the top tier cards (Fury X and 980Ti) so close to 390X and 980 in performance?

We can also see from AT's review and TR's review that Fury/Fury X cards needs a newer driver to optimize their performance of there is an architectural bottleneck in this title.

Barely any performance difference between the 390X/Nano/Fury/Fury X at 4K is also very odd. This game should be retested at a later date with updated AMD drivers to see if this mysterious bottleneck disappears.

This lack of scaling from Fiji relative to Hawaii was also present in AOTS, so it seems to be a global problem with the GPU and/or drivers:

http://www.pcgameshardware.de/Ashes...ials/Benchmark-DirectX-12-DirectX-11-1167997/
http://www.computerbase.de/2015-08/...of-the-singularity-unterschiede-amd-nvidia/2/
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
This lack of scaling from Fiji relative to Hawaii was also present in AOTS, so it seems to be a global problem with the GPU and/or drivers:

http://www.pcgameshardware.de/Ashes...ials/Benchmark-DirectX-12-DirectX-11-1167997/
http://www.computerbase.de/2015-08/...of-the-singularity-unterschiede-amd-nvidia/2/

That's a good point. It's possible that Fiji has some underlying bottlenecks in the architecture.

So in essence, 980Ti continues to be the best flagship card but in the $550 and below market segments, AMD's GCN now has a complete sweep in Ashes and Fable Legends, the only 2 DX12 benchmarks we have.
 

Rvenger

Elite Member <br> Super Moderator <br> Video Cards
Apr 6, 2004
6,283
5
81
All I want to know is if this game makes proper use of async or not. If not, they failed me. All that talk about it seems to have just been for the xbox version.

Zlatan said it does here.

Both camps do well in this game but it seems that the 390x is really showing its power. 2013 Full Hawaii chip is the elephant in the room.
 
Feb 19, 2009
10,457
10
76
That's a good point. It's possible that Fiji has some underlying bottlenecks in the architecture.

Fiji is front-end bottleneck (refer to original Fury X review at Computerbase.de where they talked to AMD directly about design challenges & tradeoffs, AMD is fully aware of the potential bottleneck), same setup for many more shaders. AMD do need to optimize it better to extract peak performance.
 

guskline

Diamond Member
Apr 17, 2006
5,338
476
126
Glad I kept the 2 R9 290s in CF for my 4790k rig.
 
Last edited:

Azix

Golden Member
Apr 18, 2014
1,438
67
91
Zlatan said it does here.

yeah I noticed silverforce said the extremetech author mentioned it was around 5% so went to check the comments. This is likely lower than Ashes of the singularity. Not what I expected at all.

If nvidia manages to hinder AMD GPU owners in this I am going to kill a cat. A very cute one.
 

Rvenger

Elite Member <br> Super Moderator <br> Video Cards
Apr 6, 2004
6,283
5
81
It appears for DX12 the driver overhead appears to be higher in DX12. Seems that the tables turned so we may see Nvidia eeking out AMD at the higher resolutions now, while AMD winning at the lower resolutions. Its mostly about the architecture at this point.


Fiji at this point is ROP limited which proves it because the memory bandwidth and shaders are not helping. Hawaii is the winner in performance gains.
 
Last edited: