[DX12] Fable Legends Beta Benchmarks

[Mondozei]

I was surprised to see AMD doing better at 720p initially. For anyone who hasn't been living under a rock, NV has had an advantage in lower resolutions for a long while now:

Notice how the AMD cards just do increasingly better and better as the resolution goes up. So I guess order is restored after seeing the benches from Extreme Tech.

That being said, two points well repeated. The 7970 is absolutely the GTX 8800 of this decade unless we get a similar card next year. I sold my 290 and if I hadn't been planning to upgrade next year I would have regretted it. It cost me 350 dollars in early 2014 and I looking back that was probably one of my best GPU buying decisions ever.

Also, WTF is going on with the Fury? This is something that needs to be investigated further.

 

[Glo.]

By the looks of things the difference in DX12 between Fury X and R9 390X is too small for the increased amount of shaders and much higher bandwidth in Fiji chip.


And its constant in all of the DX12 benchmarks: Ashes of Singularity and Fable Legends.

Im wondering if it has something to do with Fiji being GCN 1.2 architecture...

 

[Techhog]

By the looks of things the difference in DX12 between Fury X and R9 390X is too small for the increased amount of shaders and much higher bandwidth in Fiji chip.


And its constant in all of the DX12 benchmarks: Ashes of Singularity and Fable Legends.

Im wondering if it has something to do with Fiji being GCN 1.2 architecture...

Well, we have to look at the 285/380 in that case too.

 

[Azix]

By the looks of things the difference in DX12 between Fury X and R9 390X is too small for the increased amount of shaders and much higher bandwidth in Fiji chip.


And its constant in all of the DX12 benchmarks: Ashes of Singularity and Fable Legends.

Im wondering if it has something to do with Fiji being GCN 1.2 architecture...

I'm thinking Fury's bottleneck will only be minimized with heavy compute use. They're close to what they can put through the graphics queue and see improvements on the current 28nm and gcn.

Fiji could see the most gains with asyn compute.

 

[dogen1]

I'm thinking Fury's bottleneck will only be minimized with heavy compute use. They're close to what they can put through the graphics queue and see improvements on the current 28nm and gcn.

Fiji could see the most gains with asyn compute.

Fiji should only be ROP bottlenecked with pixel shader heavy engines, and even then you wouldn't expect most pixel shaders to be very ROP bound. Games are commonly requiring DX11 or 12 now, so they should act like it and move as much as possible to compute shaders.

 

[SlowSpyder]

Looks like GCN 1.0 does ok in this. I'd like a new card, but this one keeps on trucking at 19x12 for me.

 

[Makaveli]

Looks like GCN 1.0 does ok in this. I'd like a new card, but this one keeps on trucking at 19x12 for me.

Agreed.

 

[railven]

The async compute & DMA is ready and enabled in the AMD/NV drivers - if the hardware support it.

I have tested the benchmark on a Fury X:
1080p async on: 105,8 fps
1080p async off: 87,8 fps

I don't have Maxwell so I can't provide data on it.

So was this just a drive by post, or are we gonna get some verification?

I wonder if NV has negative scaling like it did for AOTS. Be interesting to see if this post is factual.

 

[antihelten]

So was this just a drive by post, or are we gonna get some verification?

I wonder if NV has negative scaling like it did for AOTS. Be interesting to see if this post is factual.

Zlatan may have a build of the game that's still under NDA (he obviously has a different build than the one available to tech sites), so he might not be able to provide any verification. Just guessing here.

 

[lilltesaito]

Zlatan may have a build of the game that's still under NDA (he obviously has a different build than the one available to tech sites), so he might not be able to provide any verification. Just guessing here.

I think he does give a lot of information that is borderline of NDA breaking.

Also starting think he might just leave this forum, he seems to get focused on when he post stuff.

 

[Silverforce11]

Well it's interesting that Lionhead briefed tech journalists that the bench has ~5% of compute that can function in async mode (thus, accelerated). So its a surprise to see a big difference for GCN with it on and off.

Look at it more closely:

According to Lionhead, only the Compute Shader Simulation & Culling can be run in AC mode.

On Fury X that operation takes ~0.47ms to complete. The sum of all the other task is: ~12.2ms, so the AC compute usage is close enough to the 5% that Lionhead mentioned.

So whatever build Zlatan has (he's a PS4 dev), it uses a lot more than 5% async compute.

Given the discrepancy between AMD and NV in UE4 games in DX11 mode (where GCN is anywhere from 25-50% slower than the comparative NV GPU), AMD needs every bit of % enhancement from Async Compute they can get to catch up.

 

[3DVagabond]

By the looks of things the difference in DX12 between Fury X and R9 390X is too small for the increased amount of shaders and much higher bandwidth in Fiji chip.


And its constant in all of the DX12 benchmarks: Ashes of Singularity and Fable Legends.

Im wondering if it has something to do with Fiji being GCN 1.2 architecture...

It might be ROP limited? Fiji and Hawaii have the same # of ROP's. ACE's are the same as well, but I don't think they are being fully saturated to cause the limitation.

 

[tential]

Looks like GCN 1.0 does ok in this. I'd like a new card, but this one keeps on trucking at 19x12 for me.

1440p VSR was a no go for me.

 

[poofyhairguy]

Damn, should have kept my 7970.

 

[Silverforce11]

It might be ROP limited? Fiji and Hawaii have the same # of ROP's. ACE's are the same as well, but I don't think they are being fully saturated to cause the limitation.

Normally ROP limited implies a dropping off of performance at higher resolution or AA, as seen with neutered variants with less ROPs. Fiji show opposite behavior, it scales better with higher resolution.

I think its CP or command processor limited. At lower resolution where the frame times are higher or rather, each shader is occupied less frequently due to completing the task faster, the CP is unable to keep the 4K shaders fed quick enough. We see at 4K where the it slows down, it scales better.

Remember their 1 CP is almost unchanged from GCN 1 to 1.2 while shader count keep going up.

This was addressed by AMD themselves when asked by Computerbase.de, they know their design is a compromise for the front-end and at lower resolution they have poor efficiency. Fiji screams stop-gap to me, an unbalanced design that was made to test the waters for HBM integration.

In theory, the 1 CP + 8 ACE design means that DX12 games with a lot of async compute will allow Fiji to flex its 4K shaders via the ACEs and scale much better compared to Hawaii.

 

[dogen1]

Normally ROP limited implies a dropping off of performance at higher resolution or AA, as seen with neutered variants with less ROPs. Fiji show opposite behavior, it scales better with higher resolution.

Higher resolutions scale ALU and TMU workload just as much as ROPs actually. It's probably bandwidth that helps AMD cards at higher resolutions.

 

[railven]

Zlatan may have a build of the game that's still under NDA (he obviously has a different build than the one available to tech sites), so he might not be able to provide any verification. Just guessing here.

Issue is without verification we're back to "290X with Mantle is gonna beat SLI GTX 780."

Yes, the claims are not equal, but when something is stated (and then regurgitated as fact) we're back at the AOTS issue of "Maxwell doesn't have Asynch Compute."

I think he does give a lot of information that is borderline of NDA breaking.

Also starting think he might just leave this forum, he seems to get focused on when he post stuff.

What kind of NDA does a PS4 dev have on a MSFT 1st party title?

 

[lilltesaito]

Issue is without verification we're back to "290X with Mantle is gonna beat SLI GTX 780."

Yes, the claims are not equal, but when something is stated (and then regurgitated as fact) we're back at the AOTS issue of "Maxwell doesn't have Asynch Compute."



What kind of NDA does a PS4 dev have on a MSFT 1st party title?

Maybe Beta testing it? I was speaking more so all his post not just this one.

 

[96Firebird]

Yes, the claims are not equal, but when something is stated (and then regurgitated as fact) we're back at the AOTS issue of "Maxwell doesn't have Asynch Compute."

Bolded part is the real problem... If people see something posted that they hope is true, they run with it as if it is true. Call me cynical, but I'd rather see evidence than follow with blind loyalty.

 

[railven]

Maybe Beta testing it? I was speaking more so all his post not just this one.

I wouldn't say he's any exception. I think I've seen perhaps 4 posts "Link?" "Source?" just today browsing CPU/GPU sub-forums.

Someone (anyone) makes a claim, someone else is gonna ask you to back it up. More so if someone claims/reveals they are in the "industry."

My constant "290X with Mantle will be faster than SLI GTX 780" was a claim made be someone who was saying they had close ties to AMD.

And don't forget the now infamous JFAMD (poor chap, he's just the always go to example).

This goes for both sides. (Don't forget the poor user that tried to introduce discussion on the memory issue for GTX 970 only to be met by a pitchfork & torch mob at the door).

Bolded part is the real problem... If people see something posted that they hope is true, they run with it as if it is true. Call me cynical, but I'd rather see evidence than follow with blind loyalty.

I can understand the feeling. Guess I'm just tired of it. More so because I buy into the hype and just seem to always end up with egg on my face. Haha.

 

[N7SpectreElite]

I'm new to the Forums at Anandtech but I do have Dual Titan X @ 1400Mhz OC'ed if anyone wants me to test something for them. I just wish we could get our hands on the benchmark to test for ourselves.

 

[guskline]

When, if ever, is the benchmark going to be available for us to test?

 

[tential]

Railven zlatan has some access to people in the industry judging from him getting to see nano before release as well.

But that Said, wait for more results/info

 

[Zodiark1593]

This has been a know Windows "issue". By default Windows sets Balanced power management instead of High Performance to prioritize laptops. This setting should never be used for desktops, nevermind for high-end GPU/CPU benchmarking.

Too bad many non-professional PC gamers probably have Windows using the Balanced setting.

For years I tested NV and AMD GPUs and Intel's high CPU over clocks and the conclusion remains the same from Windows 7->8->10: Balanced setting does not allow for maximum CPU/GPU boost and % utilization over the same loads as a High Performance setting does. Even running distributed computing on my i7 laptops wipes out a 10-15% chunk of my CPU performance.

I will test this further, though my gpu monitoring software shows the boost clock of my EVGA SSC GTX 960 consistently pegged out at ~1440 MHz (which is above what is advertised, and without the second six-pin to boot).

 

[SlowSpyder]

1440p VSR was a no go for me.

I see 1440P in my near-ish future.