One possible reason why AMD has the overhead advantage in this game compared to Nvidia is that the game makes extensive use of the D3D12 resource binding architecture but by design it's not completely hardware agnostic ...
Even if Forza 7 can't use the fully bindless feature in DX12, making use of partially bindless with structured buffers is enough to cause Nvidia hardware some slowdowns in these specific cases ...
One of the things that makes me question the heavily utilized CPU core theory is the 7700K doesn't do much better than the R7 1800X and in fact trails it in frame times slightly. Normally the 7700K has quite a bit better per core performance over Ryzen in games.
This is definitely one of the weirder gaming results, both GPU and CPU.
Forza 7, like Forza Horizon 3 uses resource binding tier 2. But if FH3 had no performance issues after the parallel rendering patch on NVidia hardware, and in fact runs faster than on comparable AMD hardware, then we can't really blame the resource binding architecture for the performance issues.
You are comparing a 3.5GB card to an 8GB card, on a beta code.A better comparison point for Forza 7 would Forza 6: Apex and it's AMD hardware that had the advantage to comparable Nvidia hardware in Forza 6 ...
You are comparing a 3.5GB card to an 8GB card, on a beta code.
This is the full game:
*snip*
CPU limitation, at 4K it's quite different:Not a whole lot different than I imagined, AMD hardware is fairly competitive in Forza 6 ... (poor scaling from GP104 to GP102 which is also observed in Forza 6 just like Forza 7)
CPU limitation, at 4K it's quite different:
*snip*
4K does need AA unless you're on a tiny 23" monitor. I game on a 40inch at 4K and you can definitely see the crawlies/jaggies without at least 4x MSAA ( I use 8X). I've got old eyes8xMSAA? Why? 2x at MOST. 4k doesn't even need AA. I hate tests that use it it isn't accurate for me.
8xMSAA? Why? 2x at MOST. 4k doesn't even need AA.
Yet they are ahead of NVIDIA as well, the 390 is far ahead of the 970, the RX 580 is far ahead of the 1060, and Vega 64 is far ahead of the 1080. This isn't about Vega alone.The previous architectures are not built for these scenarios, even if the mentioned limitations are more common in today's engines.
Yup, the game doesn't even use D3D12, it's basically D3D11 wrapped in 12.I think this is not really a "D3D12 loves Vega" thing.
Forza 7 was developed by Turn 10 Studios and is based on the in-house ForzaTech engine. The game uses only the DirectX 12 API, but only the feature level 11_0 and, like all other DirectX 12 games so far, no new hardware features of the API. Whether Async Compute is used, is unknown, but appears in view of the test results as probable.
I've heard some developers support this notion, but they also say that a properly implemented multithreaded renderer would have a minimal increase on input latency. Just look at Doom for instance. Doom's Vulkan renderer is amazing, and has extremely low input latency. It's actually one of the best games to play on a high refresh rate monitor because the game is so responsive.
And then there's FH3's post parallel rendering patch, which all would agree performed much better than the original model which main loaded one or two cores.
Yet, doom actually doesn't have nearly as good input latency as other games like cod, as tested by digital foundry. Idk what forza would measure at, but these guys live and breathe 60 fps. I'm pretty sure they know what they're doing. Plus, just because one core is at or near 100% load doesn't mean it's actually being utilized fully (check core load while using a frame limiter like RTSS. it's just a wait loop), nor does it automatically mean it's acting as a bottleneck for the game. If that one core is really just running a input loop it's not gonna bottleneck anything else. The entire point is to guarantee that input is being captured as often as possible without any interference.
Forza Horizon 3 has different bottlenecks compared to Forza 7 and Turn 10 studios wasn't the main developer behind Forza Horizon 3, that was Playground Games ...
A better comparison point for Forza 7 would Forza 6: Apex and it's AMD hardware that had the advantage to comparable Nvidia hardware in Forza 6 ...
Overhead didn't matter much with Forza Horizon 3 since it's built with a lower performance profile in mind (30FPS on consoles) but it starts to matter much more in trying to reach lower frametime targets such as Forza 6/7 ...
D3D12 resource binding is not entirely agnostic by design since some vendors have more input on different parts of the gfx API. Fully bindless exists in D3D12 not only because of AMD but because Intel had new hardware (Gen 9) on the way to expose the feature so you don't really see fully bindless exposed in Metal 2. D3D12 multi-engine is where AMD and Nvidia (you'd be surprised they would collaborate on this but they share common functionality so it's good to expose separate copy queues for async copy on both sides) had more input than Intel (no DMA engine or compute engines) ... (it's normal for gfx APIs to have vendor bias, in fact it's almost impossible to be perfectly agnostic)
Yet, doom actually doesn't have nearly as good input latency as other games like cod, as tested by digital foundry. Idk what forza would measure at, but these guys live and breathe 60 fps. I'm pretty sure they know what they're doing. Plus, just because one core is at or near 100% load doesn't mean it's actually being utilized fully (check core load while using a frame limiter like RTSS. it's just a wait loop), nor does it automatically mean it's acting as a bottleneck for the game. If that one core is really just running a input loop it's not gonna bottleneck anything else. The entire point is to guarantee that input is being captured as often as possible without any interference.
That's actually a fairly common side effect of multi-threading in games ...
What developers will often do to 'use' more cores is that they'll buffer up multiple frames so that game logic and rendering can be done in parallel across multiple frames along with the synchronization thus spiking usage across more CPU cores ...
Instead of the input lag costs simply being game logic + rendering it becomes game logic/rendering (1st frame) + game logic/rendering (2nd frame) which effectively doubles the input latency ...
At first multi-threaded rendering sounds like a great idea since higher CPU utilization = win but you're arguably getting a less responsive experience in the end ...
I'd liken the comparison between framerate and frametime. Higher framerate =/= smoother experience but lower frametimes does translate to smoother experience ...
It's not just about the numbers, one has to ask themselves if they are also getting a better experience in the end ...
That would mean that the game would only check for input after every second frame,what's the reasoning for that if the input thread can,and probably is, a separate thread?Instead of the input lag costs simply being game logic + rendering it becomes game logic/rendering (1st frame) + game logic/rendering (2nd frame) which effectively doubles the input latency ...
Actually a whole lot different, as NVIDIA cards don't suffer as badly @1080p, they stay ahead of AMD GPUs.Not a whole lot different than I imagined, AMD hardware is fairly competitive in Forza 6 ... (poor scaling from GP104 to GP102 which is also observed in Forza 6 just like Forza 7)
The second chart you posted shows Fiji having the smoothest experience out of them all since it has the best frametimes ...
Nvidia using sofware scheduler in dx12 same as in dx11.So they runs dx12 with sofware way with drivers.This causing more cpu overhead under dx12 compare to AMD but it is better under dx11 than AMD.So you think Phynaz's comments (and others) are about damage control and that's it? Sorry but that's ridiculous. This is a legitimate complaint. It's obvious that something is amiss with these scores, as the GTX 1080 Ti is only 7% faster than the GTX 1080 at 1080p, which means that it's being bottlenecked somewhere.
It could be the driver, or it could be the game's CPU thread model, we don't know. But what we do know is that it's not normal. And since this FH3 had some serious CPU performance issues when it first released that was eventually resolved, my bet is that the game's CPU usage model is at fault. NVidia's driver uses more CPU than AMD's, which probably explains why NVidia is more affected by this than AMD.
Nvidia using sofware scheduler in dx12 same as in dx11.So they runs dx12 with sofware way with drivers.This causing more cpu overhead under dx12 compare to AMD but it is better under dx11 than AMD.
AMD using hardware schedulers in dx12.SO they dont have any cpu overhead under dx12.That why vega is faster in lower resolutions.
1080TI is only 7% faster than GTX1080 in 1920x1080 and 2560x1440 and only 20% faster in 4k.That indicate CPU bottleneck even in 4k(1080TI still pushing 90+fps in 4k)
no it is 30-35% faster.Without any bottlenecks the 1080ti is only a little over 20% faster than the 1080. It's like 11tf to 9tf