[ComputerBase.de] Forza 7 Benchmark: Vega has more gasoline in the blood than Pascal

[amenx]

The interesting result is at 1080p, there is still a lingering CPU bottleneck for NV GPUs there.

https://www.computerbase.de/2017-09/forza-7-benchmark/2/#diagramm-forza-7-1920-1080

I guess the myth that developers are fully responsible for performance on these low level API aren't true. Drivers still matters a great deal!

Still, nearly a 20% gain with the latest driver. And still puts the 1060 slightly above furyx. Yep, drivers do matter.

 

[guskline]

Just bought Forza 7. It took forever to download from the Microsoft store. A bit of stutter on the start of some races when in 4k Ultra but it settles down. I'm presently running it on my 5960x/GTX1080TI rig with an Acer ET430k monitor and a Thrustmater T150 racing wheel and pedals. I thought Dirt 4 was a hoot but Forza is really amazing.

In fairness to Vega64, I'm glad it's punching it out with both the GTX1080 and even the GTX1080TI.
It makes Nvidia step up to the plate and further optimize (which it did).

From the latest graphs, it appears to now be in a slug contest where the GTYX1080TI comes out ahead at 4k but not by a lot.

I think what really hurts is the price of Vega 64 vs the GTX1080. However, it appears the limited quantity and seemingly endless mining demand have helped AMD.

 

[StrangerGuy]

I guess the myth that developers are fully responsible for performance on these low level API aren't true. Drivers still matters a great deal!

What, NV/AMD knows their hardware better than game devs? Shocking!

 

[SirDinadan]

What, NV/AMD knows their hardware better than game devs? Shocking!

It was a pretty popular concept / belief - IIRC, I read it on dev whitepapers / slides as well - that driver optimizations for DX12 games from AMD / NV can't bring much performance. DX12 is closer to the metal, low-level, etc. This magic NV driver debunks this idea, granted, "ForzaTech" is a fully-fledged DX12 engine.

 

[Muhammed]

The interesting result is at 1080p, there is still a lingering CPU bottleneck for NV GPUs there.

It was rather clear anomaly. I guess NVIDIA may be able to squeeze more out (judging by 1080p results).

Exactly, I am expecting more performance increase is coming either from NVIDIA again or from the developer.

From the latest graphs, it appears to now be in a slug contest where the GTYX1080TI comes out ahead at 4k but not by a lot.

The 1080Ti is significantly ahead with the new driver, it's just not retested in the graphs, only the 1080 and 1060. And the 1080 is beating Vega 64 at 4K too.

 

[Trovaricon]

It was a pretty popular concept / belief - IIRC, I read it on dev whitepapers / slides as well - that driver optimizations for DX12 games from AMD / NV can't bring much performance. DX12 is closer to the metal, low-level, etc. This magic NV driver debunks this idea, granted, "ForzaTech" is a fully-fledged DX12 engine.

This is the result when technical topic gets handled / is promoted for technical-buzzword-aware audience (I am on the edge if it is worse if it's intentional or simply caused by not-so-much technical staff - marketing).

If you are in control of any link in the chain of execution closer to the system / hardware than the application which calls any API (D3D12 / Vulkan included), then you might completely change WHO and HOW executes called functions. You might as well execute rendering on the CPU - with a few lines of code which will ignore what the application requested.

 

[SirDinadan]

GTX 1060 gains more than 20% with NV 387.92 vs NV 385.69.

Spoiler: TC @ Brands Hatch - 1440p ultra, MSAA Off

 

[Phynaz]

It was a pretty popular concept / belief - IIRC, I read it on dev whitepapers / slides as well - that driver optimizations for DX12 games from AMD / NV can't bring much performance. DX12 is closer to the metal, low-level, etc. This magic NV driver debunks this idea, granted, "ForzaTech" is a fully-fledged DX12 engine.

Nvidia could very well be intercepting dumb stuff done by the developer and fixing it. This happens a lot with shader replacement for example. It's also why driver packages have grown so large.

 

[Despoiler]

I just think it's funny that Nvidia released a driver that apparently did nothing. Then released the actual driver later. Marketing Ready driver to be sure.

 

[4K_shmoorK]

I just think it's funny that Nvidia released a driver that apparently did nothing. Then released the actual driver later. Marketing Ready driver to be sure.

And you know this how? What purpose does posting this serve other than trying to get it one more stir of the pot before the thread dies?

 

[Despoiler]

And you know this how? What purpose does posting this serve other than trying to get it one more stir of the pot before the thread dies?

They released two successive drivers. The first was clearly a marketing check box if the second boosted perf 15-25% a week later. They could have waited a week and then released the driver that was actually "game ready."

Stir of the pot? Calling out Nvidia's deceptive marketing practices seems fair game. It's not like AMD's marketing gets a pass purely to not stir the pot.

 

[Phynaz]

They released two successive drivers. The first was clearly a marketing check box if the second boosted perf 15-25% a week later. They could have waited a week and then released the driver that was actually "game ready."

Stir of the pot? Calling out Nvidia's deceptive marketing practices seems fair game. It's not like AMD's marketing gets a pass purely to not stir the pot.

Nvidia releasing a new driver is "deceptive marketing"?

You should really take some time to consider how deep AMD has burrowed into your psyche.

 

[Muhammed]

They released two successive drivers. The first was clearly a marketing check box if the second boosted perf 15-25% a week later. They could have waited a week and then released the driver that was actually "game ready."

LOL, apply that to your AMD fine wine mentality and you are set with an AMD deceptive marketing practices for life!

Not a while ago you even denied the possibility of them releasing a driver to fix this game, hanging on desperately to the idea that they already released a game ready driver, despite several others stating it will happen, but now instead of you sucking up your mistakes, you call them on it when they actually release a driver that fixes the issue?!

 

[CPUGuy]

The profile for Forza 7 was released a while back. Not until recently was that profile "corrected"
http://us.download.nvidia.com/Windows/385.69/385.69-win10-win8-win7-desktop-release-notes.pdf

 

[Despoiler]

Nvidia releasing a new driver is "deceptive marketing"?

You should really take some time to consider how deep AMD has burrowed into your psyche.

Reading comprehension. The new driver wouldn't be deceptive marketing. Tell me, what was the point of the first when the second clearly was the one that should have been released?

LOL, apply that to your AMD fine wine mentality and you are set with an AMD deceptive marketing practices for life!

Not a while ago you even denied the possibility of them releasing a driver to fix this game, hanging on desperately to the idea that they already released a game ready driver, despite several others stating it will happen, but now instead of you sucking up your mistakes, you call them on it when they actually release a driver that fixes the issue?!

Finewine has nothing to do with game specific driver up dates. It definitely doesn't apply to drivers updates released a week from each other. Finewine is an over time observation. Not only that I never said Nvidia wouldn't release another driver. I said it "doesn't seem likely". That leaves a window for things to change. Reading comprehension? It was actually giving Nvidia the benefit of the doubt that they would only release a proper driver. Obviously it's not the case. The first one was just for marketing purposes.

 

[Muhammed]

The first one was just for marketing purposes.

The first one didn't release for just Forza 7, it was for 9 other games.

Finewine has nothing to do with game specific driver up dates.

It has according to your flawed logic, AMD should just delay releasing drivers that gradually improve gaming performance until a final driver is ready that comes with all the improvements in one go!

I said it "doesn't seem likely". That leaves a window for things to change. Reading comprehension? It was actually giving Nvidia the benefit of the doubt that they would only release a proper driver.

Yeah yeah I am sure you did:

Nvidia released their Forza 7 drivers a week ago. What did they release if it's not hitting the fast path in their drivers?

So wait for unannounced future drivers after the "game ready" drivers they already released? This is in a game they have a marketing deal for, but their performance is well behind AMD @ certain resolutions? Wouldn't they have wanted to get it right going out the door to reviewers? I think you just started the Nvidia version of the "wait for vega/wait for navi" meme? Nvidia:Wait for drivers.

 

[pj-]

Reading comprehension. The new driver wouldn't be deceptive marketing. Tell me, what was the point of the first when the second clearly was the one that should have been released?

Use your imagination. Maybe they screwed up, maybe they found the optimization after the last driver was released. "Game ready" isn't a big enough deal for 99% of people that nvidia would intentionally release a driver that does nothing only to release the real one a week later. To anyone who actually cares, doing it this way makes them look worse than if they had just waited a week.

 

[Hail The Brain Slug]

Use your imagination. Maybe they screwed up, maybe they found the optimization after the last driver was released. "Game ready" isn't a big enough deal for 99% of people that nvidia would intentionally release a driver that does nothing only to release the real one a week later. To anyone who actually cares, doing it this way makes them look worse than if they had just waited a week.

IIRC, Nvidia game-ready drivers have been released days after game launch in some cases. I'm betting on Nvidia not knowing about the optimization until after the first one came out.

 

[PhoBoChai]

It was a pretty popular concept / belief - IIRC, I read it on dev whitepapers / slides as well - that driver optimizations for DX12 games from AMD / NV can't bring much performance. DX12 is closer to the metal, low-level, etc. This magic NV driver debunks this idea, granted, "ForzaTech" is a fully-fledged DX12 engine.

Correct. This was the standard myth not long ago.

But we all know that it isn't true. DX12 & Vulkan still has to interface with the drivers to function, it's not actual code to the metal as marketing would have tried to make people believe.

This game is pretty interesting to analyze on a technical level, seems like NV suffer CPU overhead at lower resolution while RX Vega drop off suggest it's bandwidth constrained (less bandwidth than Fury X for 50% more Tflops ins very unbalanced given no major architecture improvement to bw efficiency).

 

[Ancalagon44]

Correct. This was the standard myth not long ago.

But we all know that it isn't true. DX12 & Vulkan still has to interface with the drivers to function, it's not actual code to the metal as marketing would have tried to make people believe.

This game is pretty interesting to analyze on a technical level, seems like NV suffer CPU overhead at lower resolution while RX Vega drop off suggest it's bandwidth constrained (less bandwidth than Fury X for 50% more Tflops ins very unbalanced given no major architecture improvement to bw efficiency).

I think AMD mentioned that Polaris had some color compression magic, but yeah, Vega is still likely memory bandwidth constrained. This is likely why running Vega 56 and Vega 64 at the same clocks (core and memory) yields nearly identical performance.

 

[Tup3x]

I just think it's funny that Nvidia released a driver that apparently did nothing. Then released the actual driver later. Marketing Ready driver to be sure.

Game ready driver basically means that the game will work. It doesn't mean that they are done optimising it.

 

[nathanddrews]

So if I had to summarize this thread to a not-so-tech-savvy friend, it would go something like this:

"At launch, Vega showed significant performance over Pascal, but after a few updates, they are very close in performance. It's a brand new game and will probably keep getting refined updates over the next few weeks, so I wouldn't make a big deal about it."

Like @Tup3x said, game ready drivers are simply drivers that have been validated to work with a game. Stability prioritized over performance.

 

[SirDinadan]

"At launch, Vega showed significant performance over Pascal, but after a few updates, they are very close in performance. It's a brand new game and will probably keep getting refined updates over the next few weeks, so I wouldn't make a big deal about it."

Release date performance is still a huge deal. Established tech-sites and the so called TechTubers review games as soon as they are out so they don't miss out on the early hype - huge chunk of the overall traffic for a review / video analysis of a game is generated around this time. Youtube prioritizes upload time and trending in searching results, if you miss the release-window, you are screwed. You never gonna hit many views with a thorough review after some days after launch date. Furthermore, game performance data bases are usually not updated with new driver releases.

As I said, search engines favor upload date and trending, so anyone (Average Joe) looking for a performance review of a game will stumble upon many reviews that were created after launch.

 

[Carfax83]

You were saying?

What I said hasn't really changed. The GTX 1080 is still CPU bottlenecked with the 387.92 drivers, because even though it outperforms Vega 64 at 4K with the new drivers, it's still significantly slower at 1080p.

 

[Hail The Brain Slug]

What I said hasn't really changed. The GTX 1080 is still CPU bottlenecked with the 387.92 drivers, because even though it outperforms Vega 64 at 4K with the new drivers, it's still significantly slower at 1080p.

While I agree with your overall sentiment that there's still something funny going on because the pascal cards are scaling much differently than the rest, I think you're wrong about the 1080p results.

1.1% slower is not "significantly slower".

Edit: Just herpin the derp, looking at wrong results. Nothing to see here.

I'm interested to see if Nvidia continues to improve the results with further updates.